N

ình
n hB
Mạ
Hype Cycle for Artificial Intelligence, 2024

ng
à
Ho
17 June 2024 - ID G00809438 - 126 min read

_
MB
By Analyst(s): Afraz Jaffri, Haritha Khandabattu

NH
Initiatives:Artificial Intelligence; Analytics and Artificial Intelligence; Evolve Technology and

E
UY
Process Capabilities to Support D&A; Generative AI Resource Center

NG
Investment in AI has reached a new high with a focus on

4
generative AI, which, in most cases, has yet to deliver its

02
2/2
anticipated business value. This research helps AI leaders identify

4/1
other techniques worthy of investment by profiling a wide range of

10
2:2
AI innovations, many of which are advancing fast.

_1
ên
uy
Analysis
Ng
nh
What You Need to Know
Bì
nh

Generative AI (GenAI) has passed the Peak of Inflated Expectations, although hype about
Mạ

it continues. In 2024, more value will derive from projects based on other AI techniques,
ng

either stand-alone or in combination with GenAI, that have standardized processes to aid
oà

implementation. To deliver maximum benefit, AI leaders should base future system

_H

architectures on composite AI techniques by combining approaches from innovations at

MB

all stages of the Hype Cycle.

NH
UYE

As the volume and scale of AI projects have increased, second-order effects have come
NG

into play. Increasing attention is therefore being paid to governance, risk, ownership,
safety and mitigation of technical debt. These factors are being addressed at national,
24
/20

enterprise, team and individual practitioner levels, but, even with regulations reaching
/12

advanced stages, maturity is far from being achieved.

04
:21

The Hype Cycle

12
n_

The two biggest movers on this year’s Hype Cycle, AI engineering and knowledge graphs,
ê

highlight the need for means of handling AI models at scale in a robust manner. AI
uy
Ng

engineering is a fundamental requirement for delivery, at scale, of enterprise AI solutions

ình

that demand new team topologies. Knowledge graphs provide dependable logic and
hB

explainable reasoning, in contrast to the fallible, yet powerful, predictive capabilities of the
ạn

deep-learning techniques used by GenAI.

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 1 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Innovations at the Innovation Trigger include composite AI, AI-ready data, causal AI,

ng
decision intelligence, AI simulation and multiagent systems. These reflect the growing

à
Ho
need to advance process and decision automation beyond single-model outputs into

_
orchestrated multiturn composite services.

MB
NH
At the Peak of Inflated Expectations, responsible AI, AI TRiSM, prompt engineering and

E
UY
sovereign AI point to increasing concerns about the governance and safety aspects of the

NG
rapidly expanding use of AI by enterprises and individuals.

4
02
Soon to leave the peak or already in the Trough of Disillusionment are synthetic data,

2/2
ModelOps, edge AI, neuromorphic computing and smart robots. These innovations still

4/1
have momentum, but levels of implementation vary, and they are frequently used

10
incorrectly or subject to inflated expectations of business value. Neuromorphic computing

2:2
and smart robots have advanced significantly in the past year, indicating the potential for

_1
ên
rapid progression through the rest of the Hype Cycle.

uy
Ng
Cloud AI services have regressed on the Hype Cycle since last year, due to the number of
nh
GenAI-based cloud AI services that have come to market. Vendors and end users of these
Bì
nh

services have experienced problems with service capacity, reliability, model update
Mạ

frequency and cost fluctuation, which may, however, be considered growing pains.
ng
oà

On the Slope of Enlightenment are AI technologies that have many years of innovation
_H
MB

behind them and are getting nearer to mainstream adoption. Usage of autonomous
NH

vehicles has increased in some locations, despite severe skepticism in certain quarters,
YE

the imposition of restrictions and the withdrawal of some operating licenses. Intelligent
U

applications, now powered by GenAI, have entered the workforce, but more time is needed
NG

to objectively quantify their impact on productivity.

24
/20

New entries on this year’s Hype Cycle include quantum AI, embodied AI and sovereign AI,
/12

as companies and governments are starting to respond to the potential, and dangers, of
04

an AI-dominated future.
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 2 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Figure 1: Hype Cycle for Artificial Intelligence, 2024

ng
à
Ho
_
MB
NH
E
UY
NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng

The Priority Matrix

oà
_H

Compared with many other Hype Cycles, this one is unusual in having so many
MB

innovations of transformational or high benefit, none of moderate benefit, and only one of
NH

low benefit.
UYE
NG

Gartner expects that, within two years, composite AI will be the standard methodology for
developing AI systems, and to be widely adopted. Another transformational innovation,
24

computer vision, is already the subject of mass consumer adoption through smart
/20

devices.
/12
04
:21

Innovations two to five years away from mainstream adoption that merit particular
12

attention include decision intelligence, embodied AI, foundation models, GenAI, intelligent
n_

applications and responsible AI. Early adoption of these will lead to significant
ê
uy

competitive advantage and ease the problems associated with using AI models within
Ng

business processes.
ình
hB

Among the innovations five to 10 years away from mainstream adoption, neuromorphic
ạn
gM

computing could open doors to novel AI architectures. An influx of new ideas and
entrepreneurial ventures will be essential for further development of this technology.
àn
Ho
B_

Gartner, Inc. | G00809438 Page 3 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
AI leaders should balance strategic exploration of potentially transformative or highly

ng
beneficial innovations with investigation of innovations that do not require extensive

à
Ho
proficiency in engineering or data science, and that have been commoditized both as

_
stand-alone applications and as components of packaged business solutions.

MB
NH
Table 1: Priority Matrix for Artificial Intelligence, 2024

E
UY
(Enlarged table in Appendix)

NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG

Off the Hype Cycle

24

The following innovations have been dropped from this year’s Hype Cycle:
/20
/12

■ Operational AI systems: Subsumed by AI engineering.

04
:21

■ Data labeling and annotation: Dropped because it is more relevant to the Hype Cycle
12

for Data Science and Machine Learning, 2024.

ên_
uy

■ AI maker and teaching kits: Dropped due to a lack of hype.

Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 4 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
On the Rise

ng
Autonomic Systems

à
Ho
Analysis By: Erick Brethenoux, Nick Jones

_
MB
NH
Benefit Rating: Transformational

E
UY
NG
Market Penetration: Less than 1% of target audience

4
Maturity: Emerging

02
2/2
4/1
Definition:

10
Autonomic systems are self-managing physical or software systems, performing domain-

2:2
bounded tasks, that exhibit three fundamental characteristics: autonomy (execute their

_1
ên
own decisions and tasks autonomously without external assistance); learning (modify

uy
their behavior and internal operations based on experience, changing conditions or goals);
Ng
and agency (have a sense of their own internal state and purpose that guides how and
nh
what they learn and enables them to act independently).
Bì
nh
Mạ

Why This Is Important

ng

Autonomic systems are emerging as an important trend as they enable levels of business
oà

adaptability, flexibility and agility that can’t be achieved with traditional AI techniques
_H
MB

alone. Their flexibility is valuable in situations where the operating environment is

NH

unpredictable and real-time monitoring and control aren’t practical. Their learning ability is
YE

valuable in situations where a task can be learned even though there is no well-
U

understood algorithm (composite AI) to implement it.

NG

Business Impact
24
/20

Autonomic systems excel where:

/12
04

■ Conventional automation applying composite AI techniques is inadequate, or using

:21

fixed training data is impractical or not agile.

12
n_

It is impractical to provide real-time human guidance, or training conditions can’t be

ê

■
uy

anticipated.
Ng
ình

■ We cannot program the exact learning algorithm, but the task is continuously
hB

learnable.
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 5 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Continuously or rapidly changing tasks or environments make frequent retraining

ng
and testing of machine learning systems too slow or costly.

à
Ho
_
MB
Drivers

NH
Autonomic systems are the culmination of a three-part trend:

E
UY
NG
■ Automated systems are a very mature concept. They perform well-defined tasks and
have fixed deterministic behavior (such as an assembly robot welding cars). The

4
increasing number of use cases around automation using AI techniques is a strong

02
2/2
base for autonomous systems.

4/1
Autonomous systems go beyond simple automation to add independent behavior.

10
■

2:2
They may exhibit some degree of adaptive behavior, but are predominantly under

_1
algorithmic control (such as self-driving cars or a Boston Dynamics’ Spot robot that

ên
has its overall route and goals set by a remote human operator but has substantial

uy
local autonomy — that is, for a very specific task). Adaptive AI capabilities are a
Ng
necessary foundation for autonomic systems and should accelerate the adoption of
nh
Bì

autonomic systems.
nh
Mạ

■ Autonomic systems exhibit adaptive behavior through learning and self-modifying

ng

algorithms. For example, Ericsson has demonstrated the use of reinforcement

oà

learning and digital twins to create an autonomic system that dynamically optimizes
_H

5G network performance while creating optimization rules. This trend is showing the
MB

feasibility of such systems. Early learning about carefully bounded autonomic

NH
YE

systems will build trust in their capabilities to operate independently.

U
NG

Other drivers include:

24
/20

■ Autonomic behavior is a spectrum. For example, chatbots learn from internet

/12

discussions; streaming services learn which content you like; and delivery robots
04

share information about paths and obstructions to optimize fleet routes. The
:21

advantages of systems that can learn and adapt their behavior will be compelling.
12
n_

Agent-based systems are seeing an adoption renaissance fueled by the increasing

ê

■
uy

complexity of existing applications and the advent of large action models.

Ng
ình

■ Substantial academic research is underway on autonomics, which will result in more

hB

widespread use.
ạn
gM
àn

Obstacles
Ho
B_

Gartner, Inc. | G00809438 Page 6 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Nondeterminism: Systems that continuously learn and adapt their behavior aren’t

ng
predictable. This will pose challenges (such as legal) for employees and customers

à
Ho
who may not understand how and why a system performed as it did.

_
MB
■ Immaturity: Skills in the area will be lacking until autonomics becomes more

NH
mainstream. New types of professional services may be required (like autonomous

E
UY
business skills).

NG
■ Social concerns: Misbehavior, nondeterminism or lack of understanding could
generate public resistance when systems interact with people.

4
02
2/2
■ Digital ethics and safety: Autonomic systems will require architectures and

4/1
guardrails to prevent them from learning undesirable, dangerous, unethical or even

10
illegal behavior when no human is validating the system.

2:2
_1
■ Legal liability: It may be difficult for the supplier of an autonomic system to take

ên
uy
total responsibility for its behavior because that will depend on the goals it has set,

Ng
its operating conditions and what it learned. nh
Bì
nh

User Recommendations
Mạ

■ Start by building experience with autonomous systems first to understand the

ng

constraints and requirements (legal, technical and cultural) that the organization is
oà
_H

subjected to. Pilot autonomic technologies in cases where early adoption will deliver
MB

agility and performance benefits in software or physical systems.

NH
YE

■ Manage risk in autonomic system deployments by analyzing the business, legal and
U

ethical consequences of deploying autonomic systems — which are partially

NG

nondeterministic. Do so by creating a multidisciplinary task force.

24

■ Optimize the benefits of autonomic technologies by piloting them in situations such

/20

as complex and rapidly changing environments where early adoption will deliver
/12

agility and performance benefits in either software or physical systems.

04
:21
12

Sample Vendors
ên_

Aspire; IBM; Latent AI; Playtika Holding; Vanti.

uy
Ng
ình

Gartner Recommended Reading

hB

Top Strategic Technology Trends for 2022: Autonomic Systems

ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 7 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Quantum AI

ng
Analysis By: Chirag Dekate, Soyeb Barot

à
Ho
_
MB
Benefit Rating: Low

NH
E
Market Penetration: Less than 1% of target audience

UY
NG
Maturity: Embryonic

4
02
Definition:

2/2
4/1
Quantum artificial intelligence is an embryonic field of research emerging at the

10
intersection of quantum technologies and AI. Quantum AI aims to exploit unique

2:2
properties of quantum mechanics to develop new and more powerful AI algorithms that

_1
deliver better than classical performance, potentially resulting in new types of AI

ên
algorithms designed to run on quantum systems.

uy
Ng
nh
Why This Is Important
Bì

Quantum AI is an area of active research. Once commercialized, quantum AI could

nh
Mạ

potentially help in:

ng
oà

■ Enabling organizations to use quantum systems to address advanced AI analytics

_H

faster while using a fraction of the resources used in conventional AI

MB

supercomputing resources.
NH
YE

■ Developing new AI algorithms that exploit quantum mechanics to deliver capabilities

U
NG

beyond ones that can be executed on classical systems.

Unlocking disruptive applications that include drug discovery, energy industry and
24

■
/20

logistics.
/12
04

Business Impact
:21
12

While the business impact of the embryonic quantum AI field today is low, when validated
n_

techniques mature, quantum AI will enable competitive advantage across industries; for
ê
uy

instance:
Ng
ình

■ Life sciences: Transform drug discovery by shortening timelines, lowering costs and
hB

improving outcomes.
ạn
gM

■ Finance: Optimize portfolios, minimize risk and improve fraud detection systems.
àn
Ho
B_

Gartner, Inc. | G00809438 Page 8 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Material science: Discover new materials that revolutionize energy transportation,

ng
manufacturing and create new revenue streams.

à
Ho
_
MB
Drivers

NH
■ Hype around quantum technologies is driving more businesses and researchers to

E
UY
explore the intersection of quantum and AI.

NG
■ The accelerated pace of innovation in quantum systems (including larger volume of
higher quality qubits, and greater stability and reliability of quantum systems) is

4
02
driving greater interest in applicability in areas, including quantum AI.

2/2
4/1
■ Access to quantum computing as a service is lowering the barrier to entry,

10
encouraging greater collaboration among researchers and enabling exploration of

2:2
new algorithms and techniques.

_1
ên
Governments and enterprises globally are increasing funding for quantum (and

uy
■

Ng
quantum AI) research, resulting in accelerated innovation.nh
Bì

■ The halo effect of increased hype around generative AI is driving new focus on
nh

alternative research techniques, including quantum AI, that could potentially deliver
Mạ

new disruptive results.

ng
oà

■ Universities and training programs are developing programs and curricula to develop
_H

a quantum-ready workforce.
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 9 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Hardware limitations: Current quantum systems, while getting stabler, are still error-

à
Ho
prone and inherently noisy, limiting their utility and impact on practical quantum AI.

_
MB
■ Algorithm limitations: While several quantum AI algorithms have been proposed,

NH
very few have been vetted and proven, and they are nowhere close to being

E
UY
enterprise-ready.

NG
■ Cost: Despite their limited utility and widespread accessibility, rapidly evolving noisy

4
intermediate-scale quantum (NISQ) systems are relatively expensive, which could

02
2/2
inhibit research and development efforts needed to devise quantum AI algorithms.

4/1
Scalability of systems: Scaling quantum systems to the level necessary for

10
■

2:2
enterprise-ready quantum AI continues to be a major technical hurdle.

_1
ên
■ Compute paradigms: Integrating traditional data and analytics pipelines with

uy
quantum is inherently challenging because quantum systems operate on a
Ng
fundamentally different paradigm both from a data representation perspective and
nh
from a compute (non-von Neumann model) perspective.
Bì
nh
Mạ

User Recommendations
ng
oà

■ Prioritize investments in AI and GenAI over any quantum AI investments. Quantum AI

_H

is too nascent to warrant focused investments and unlikely to yield material gains in
MB

the next two to three years.

NH
YE

■ Partner with local universities by sponsoring academic research as a means of

U

derisking your quantum AI investments and create a university-to-industry talent

NG

pipeline.
24

Create a quantum AI opportunity radar that enables you to track progress of

/20

■
underlying technologies and quantum AI algorithms, enabling you to maximize value
/12
04

creation as the embryonic field of quantum technologies evolves.

:21
12

Sample Vendors
ên_

Amazon; Google; IBM; IonQ; Multiverse Computing; Pasqal; SandboxAQ; Zapata AI

uy
Ng
ình

Gartner Recommended Reading

hB

Infographic: How Use Cases Are Developed and Executed on a Quantum Computer
ạn
gM

Predicts 2024: Emerging Defense Technology and New Domains

àn
Ho
B_

Gartner, Inc. | G00809438 Page 10 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Innovation Insight for Quantum Computing for the Automotive Industry

ng
à
Ho
Cool Vendors in Quantum Computing

_
MB
First-Principles AI

NH
E
Analysis By: Erick Brethenoux, Svetlana Sicular

UY
NG
Benefit Rating: Transformational

4
02
Market Penetration: 5% to 20% of target audience

2/2
4/1
Maturity: Emerging

10
2:2
_1
Definition:

ên
First-principles AI (FPAI; aka physics-informed AI) incorporates physical and analog

uy
Ng
principles, governing laws and domain knowledge into AI models. In contrast, purely
nh
“digital” AI models do not necessarily obey the fundamental governing laws of physical
Bì

systems and first principles — nor generalize well to scenarios on which they have not
nh

been trained. FPAI extends AI engineering to complex systems engineering and model-
Mạ

based systems (like agent-based systems).

ng
oà
_H

Why This Is Important

MB

As AI expands in engineering and scientific use cases, it needs a stronger ability to model
NH

problems and better represent their context. Digital-only AI solutions cannot generalize
YE

well enough beyond training, limiting their adaptability. FPAI instills a more reliable
U
NG

representation of the context and the physical reality, yielding more adaptive systems.
This leads to reduced training time, improved data efficiency, better generalization and
24

greater physical consistency.

/20
/12

Business Impact
04
:21

Physically consistent and scientifically sound AI models can significantly improve

12

applicability, especially in engineering use cases. FPAI helps train models with fewer data
n_

points and accelerates the training process, helping models converge faster to optimal
ê
uy

solutions. It improves the generalizability of models to make reliable predictions for

Ng

unseen scenarios, including applicability to nonstationary systems, and enhances

ình

transparency and interpretability boosting trustworthiness.

hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 11 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ FPAI approaches instill a more flexible representation of the context and conditions

à
Ho
in which systems operate, allowing developers to build more adaptive systems.

_
MB
Traditional business modeling approaches have been brittle. This is because the

NH
digital building blocks making up solutions cannot generalize well enough beyond

E
their initial training data, therefore limiting the adaptability of those solutions.

UY
NG
■ FPAI approaches provide additional physical knowledge representations, such as
partial differential equations to guide or bound AI models. Traditional AI techniques,

4
02
particularly in the machine learning family, have been confronted with severe

2/2
limitations — especially for causality and dependency analysis, admissible values,

4/1
context flexibility and memory retention mechanisms. Asset-centric industries have

10
already started leveraging FPAI in physical prototyping, predictive maintenance or

2:2
composite materials analysis, in conjunction with augmented reality

_1
ên
implementations.

uy
Ng
■ Complex systems like climate models, large-scale digital twins and complex health
nh
science problems are particularly challenging to model. Composite AI approaches
Bì

provide more concrete answers and manageable solutions to these problems, but
nh

their engineering remains a significant challenge. FPAI provides more immediate

Mạ

answers to these problems.

ng
oà
_H

■ First principles knowledge simplify and enrich AI approaches by defining problem

MB

and solution boundaries, reducing the scope of traditionally brute force approach
NH

employed by ML; for example, known trajectories of physical objects simplify AI-
YE

enabled sky monitoring. First-principles-based semantics reveal deepfakes.

U
NG

■ The need for more robust and adaptable business simulation systems will also
promote the adoption of FPAI approaches. With a better range of context
24
/20

modelization and more accurate knowledge representation techniques, simulations

/12

will be more reliable and account for a wider range of possible scenarios — all better
04

anchored in reality.
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 12 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ The development of systematic tests and standardized evaluation for these models

à
Ho
across benchmark datasets and problems could slow down the adoption of FPAI

_
MB
capabilities.

NH
■ Computationally, the scaling of the training, testing and deployment of complex FPAI

E
UY
models on large datasets in an efficient manner will be an issue.

NG
■ Resource-wise, collaboration across many diverse communities (physicists,

4
mathematicians, computer scientists, statisticians, AI experts and domain scientists)

02
2/2
will be a challenge.

4/1
Brute force approaches are prevalent in AI, and are easy to implement for data

10
■

2:2
scientists, while first principles require additional fundamental knowledge of a

_1
subject, calling for a multidisciplinary team.

ên
uy
Ng
User Recommendations nh
■ Set realistic development objectives by identifying errors that cannot be reduced and
Bì

discrepancies that cannot be addressed, including data quality.

nh
Mạ

■ Encourage reproducible and verifiable models starting with small-scoped problems;

ng
oà

complex systems (in the scientific sense of the term) are generally good candidates
_H

for this approach.

MB
NH

■ Enforce standards for testing accuracy and physical consistency for physics and
YE

first-principles-based models of the relevant domain, while characterizing sources of

U

uncertainty.
NG

■ Promote model-consistent training for FPAI models and train models with data
24

characteristics representative of the application, such as noise, sparsity and

/20

incompleteness.
/12
04

■ Quantify generalizability in terms of how performance degrades with degree of

:21

extrapolation to unseen initial and boundary conditions and scenarios.

12
ên_

■ Ensure relevant roles and education in a multidisciplinary AI team (with domain

uy
Ng

expertise), so that the team can develop effective and verifiable solutions.
ình
hB

Sample Vendors
ạn

Abzu; IntelliSense.io; MathWorks; NNAISENSE; NVIDIA; VERSES

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 13 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Innovation Insight: AI Simulation

à
Ho
_
MB
Innovation Insight for Composite AI

NH
E
Go Beyond Machine Learning and Leverage Other AI Approaches

UY
NG
Innovation Insight: Causal AI

4
02
Predicts 2023: Simulation Combined With Advanced AI Techniques Will Drive Future AI

2/2
Investments

4/1
10
Embodied AI

2:2
_1
Analysis By: Pieter den Hamer

ên
uy
Ng
Benefit Rating: Transformational nh
Bì

Market Penetration: Less than 1% of target audience

nh
Mạ

Maturity: Embryonic
ng
oà
_H

Definition:
MB

Embodied AI is based on the view that intelligence and embodiment in a certain context
NH

are inextricably linked — one shapes the other. It is an approach where a physical or virtual
YE

AI agent’s models are trained and co-engineered with its user interface, sensors,
U
NG

appearance, actuators or other capabilities required to interact with a specific, real or

simulated environment. This enables robust, reliable and adaptive execution of intelligent
24

tasks.
/20
/12

Why This Is Important

04
:21

Embodied AI aims to create AI agents that can act autonomously or to augment humans
12

in practical, dynamic contexts — much more so than current AI, including abstract large
n_

language models with limited reliability and effectiveness in decision-making and action-
ê
uy

taking. This is achieved through active perception and adaptive behavior, orchestrated by
Ng

an AI agent’s intelligence that is in symbiosis with the capabilities and constraints of the
ình

AI agent’s host or body in a certain environment.

hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 14 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
Embodied AI will further value creation with AI across various use cases. Particularly

à
Ho
where there is a need for more practical know-how, a better representation of the physical,

_
MB
social or other characteristics of its environment, and a greater resilience to deal with

NH
unexpected or disruptive events. Example use cases include virtual assistants, avatars,

E
gaming characters, autonomous vehicles and smart robots. This will pave the way toward

UY
more effective and trusted AI and more game-changing use of AI to enable new products,

NG
services and business models.

4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 15 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ The advent of generative AI (GenAI) has catalyzed AI adoption in general. Yet it has

à
Ho
also highlighted the limitations of current AI, particularly with respect to reliability

_
MB
and the challenges with contextualization and grounding of AI in reality.

NH
■ Embodied AI benefits from advances in compute power and GenAI to support

E
UY
realistic simulations with reinforcement learning for adaptive behavior training. This

NG
also supports approaches to co-evolving baseline versions of both embodiment and
intelligence of AI agents, before further improving and deploying them in the real

4
02
world.

2/2
4/1
■ Embodied AI is enabled by emerging approaches such as physics-informed or first

10
principles AI (representing among others, the laws of physics or engineering

2:2
heuristics), composite AI (for example, using neuro-symbolic AI for spatiotemporal

_1
reasoning) and causal AI (representing cause-and-effect relations).

ên
uy
The interest for embodied AI is further fueled by innovations in areas such as
Ng
■
virtual/augmented/mixed reality, gaming, smart robotics, autonomous systems,
nh
Bì

natural language generation and emotion AI. All of which are related to the improved
nh

design of AI agents, both physical and virtual. Physical agents also benefit from
Mạ

advances in sensor technology, robotics engineering, and, for example, new

ng

materials for more natural mechanics and haptic interfaces.

oà
_H

Advances in embodied AI are underpinned by evolving scientific insights about

MB

■
intelligence, which is no longer seen as a centralized brain-only concept. Cognitive
NH
YE

traits like perception, emotion, reasoning and behavior are actually distributed and
U

co-evolved in multiple parts of the body. This also aligns well with distributed AI
NG

system architectures, including multiagent systems.

24

■ Embodied AI is seen as a critical step toward possible future artificial general

/20

intelligence as it is inseparable from its operational entity that interacts with its
/12

environment. This means it is not abstracted from but grounded in reality by design,
04

holding the promise of providing intrinsic meaning or semantics to its knowledge

:21
12

representations and ‘native’ common sense.

ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 16 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ The world is a very complex, unpredictable and even chaotic place. Which is why the

à
Ho
development of realistic simulations, effective robotics and — for example — truly

_
MB
autonomous cars has proven to be elusive.

NH
■ Real-world interaction requires real-time, highly responsive AI, even with limited

E
UY
energy and compute resources (for example, on mobile or edge devices). However,

NG
more lightweight and energy-efficient AI are not easily achievable.

4
■ Embodied AI holds the promise of even more powerful and autonomous AI.

02
2/2
Unfortunately, this may not only facilitate benevolent but also malevolent use.

4/1
Effective regulation and risk management for responsible AI are, however, not a

10
given.

2:2
_1
■ AI embodiments can be unnecessarily anthropomorphic in their design (a body with

ên
two legs and two arms), bringing in additional complexity and challenges.

uy
■ Ng
Embodied AI requires multidisciplinary collaboration between experts in areas as
nh
diverse as machine learning, graphical design, mechanical engineering and still
Bì
nh

others, depending on the use case and type of AI agent.

Mạ
ng

User Recommendations
oà
_H

■ Identify use cases that may benefit from applying embodied AI, both in more virtual
MB

domains, such as online customer interaction or knowledge worker augmentation,

NH

and in more physical domains, such as manufacturing or logistics.

UYE

Explore the value that embodied AI can add by reducing the limitations of current AI
NG

■
in terms of better interpretation of, for example, physical constraints in a warehouse
or cultural norms in client interaction. This may result in increased safety or
24
/20

decreased bias in the use of AI, respectively.

/12
04

■ Extend the mindset of how AI agents should be developed or trained. Move from a
:21

modeling-only approach toward one that considers how intelligence can be a

12

synergy between AI models and the design of the agent’s embodiment. This could,
n_

for example, relate to the facial expression of virtual agents, or the coordination of
ê
uy

movement in physical agents.

Ng
ình
hB

Sample Vendors
ạn

Amazon; Figure; Google; Hanson Robotics; Intel; Intrinsic; NNAISENSE; Qualcomm; Tesla
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 17 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Innovation Insight: AI Agents

à
Ho
_
MB
Innovation Insight: AI Simulation

NH
E
Hype Cycle for Mobile Robots and Drones, 2023

UY
NG
Building a Digital Future: The Metaverse

4
02
Emerging Technologies: Introducing the Artificial Intelligence Roadmap for Virtual

2/2
Assistants

4/1
10
Multiagent Systems

2:2
_1
Analysis By: Leinar Ramos, Pieter den Hamer, Anthony Mullen

ên
uy
Ng
Benefit Rating: High nh
Bì

Market Penetration: Less than 1% of target audience

nh
Mạ

Maturity: Embryonic
ng
oà
_H

Definition:
MB

A multiagent system (MAS) is a type of AI system composed of multiple, independent (but

NH

interactive) agents, each capable of perceiving their environment and taking actions.
YE

Agents can be AI models, software programs, robots and other computational entities.
U
NG

Multiple agents can work toward a common goal that goes beyond the ability of
individual agents, with increased adaptability and robustness.
24
/20

Why This Is Important

/12
04

Current AI is focused on the creation of individual agents built for specific use cases,
:21

limiting the potential business value of AI to simpler problems that can be solved by
12

single monolithic models. The combined application of multiple autonomous agents can
n_

tackle complex tasks that individual agents cannot, while creating more adaptable,
ê
uy

scalable and robust solutions. It is also able to succeed in environments where

Ng

decentralized decision making is required.

ình
hB

Business Impact
ạn
gM

Multiagent systems can be used in:

àn
Ho
B_

Gartner, Inc. | G00809438 Page 18 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Generative AI: Orchestrating AI agents for complex tasks

ng
à
■ Robotics: Swarms of robots and drones for warehouse optimization, search and

Ho
rescue, environment monitoring, and other use cases

_
MB
NH
■ Energy and utilities: Smart grid optimization and load balancing

E
UY
■ Supply chain: Optimizing scheduling, planning, routing and supply chain

NG
optimization

4
Telecom: Network optimization and fault detection

02
■

2/2
Healthcare: Using agents to model actors (individuals, households, professionals)

4/1
■

10
2:2
Drivers

_1
ên
■ Generative AI agents: Large language models (LLMs) are increasingly augmented

uy
with additional capabilities, such as an internal memory and plug-ins to external
Ng
applications, to implement AI agents. An emerging design pattern is to assemble and
nh
combine these LLM-based AI agents into more powerful systems, which is
Bì
nh

increasing the feasibility of and interest in multiagent systems.

Mạ

Increased decision-making complexity: AI is increasingly used in real-world

ng

■
oà

engineering problems containing complex systems, where large networks of

_H

interacting parts exhibit emergent behavior that cannot be easily predicted. The
MB

decentralized nature of multiagent systems makes them more resilient and

NH

adaptable to complex decision making.

UYE

Simulation and multiagent reinforcement learning: Advances in the realism and

NG

■
performance of simulation engines, as well as the use of new multiagent
24

reinforcement learning techniques, allow for the training of multiagent AI systems in

/20

simulation environments, which can then be deployed in the real world.

/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 19 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Training complexity: Multiagent systems are typically harder to train and build than

à
Ho
individual AI agents. These systems can exhibit emergent behavior that is hard to

_
MB
predict in advance, which increases the need for robust training and testing.

NH
■ Monitoring and governing multiple agents: Coordination and collaboration between

E
UY
agents is challenging. Careful monitoring, governance and a common grounding are

NG
required to ensure that the combined multiagent system behavior achieves its
intended goals.

4
02
2/2
■ Limited adoption and readiness: Despite its benefits, the application of multiagent

4/1
systems to real-world problems is not yet widespread, which creates a lack of

10
enterprise awareness and readiness to implement.

2:2
_1
■ Specialized skills required: Building and deploying multiagent systems requires

ên
specialized skills beyond traditional AI skills, particularly the use of reinforcement

uy
learning and simulation.
Ng
nh
Fragmented vendor landscape: A fragmented vendor landscape inhibits customer
Bì

■
nh

adoption and engagement.

Mạ
ng

User Recommendations
oà
_H

■ Use multiagent systems for complex problems that require decentralized decision
MB

making and cannot be solved by single AI agents. This includes problems with
NH

changing environments where agents need to adapt and problems where a diverse
YE

set of agents with different expertise can be combined to accomplish a goal.

U
NG

■ Shift to a multiagent approach gradually since this is an emerging area of research

and the risks and benefits are not yet fully understood.
24
/20

Establish clear guardrails when implementing multiagent systems, including legal

/12

■
04

and ethical guidelines around autonomy, liability, robust security measures and data
:21

privacy protocols.
12
n_

■ Invest in the use of simulation technologies for AI training, as simulation is the

ê
uy

primary environment to build and test multiagent systems.

Ng
ình

■ Educate your AI teams on multiagent systems, how they differ from single-agent AI
hB

design, and some of the available techniques to train and build these systems.
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 20 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sample Vendors

ng
Alphabet; Ansys; Cosmo Tech; FLAME GPU; MathWorks; Microsoft; OpenAI; The AnyLogic

à
Ho
Company

_
MB
NH
Gartner Recommended Reading

E
Innovation Insight: AI Agents

UY
NG
Innovation Insight: AI Simulation

4
02
AI Design Patterns for Large Language Models

2/2
4/1
10
AI Simulation

2:2
Analysis By: Leinar Ramos, Anthony Mullen, Pieter den Hamer, Jim Hare

_1
ên
uy
Benefit Rating: High

Ng
nh
Market Penetration: 1% to 5% of target audience
Bì
nh
Mạ

Maturity: Emerging
ng
oà

Definition:
_H

AI simulation is the combined application of AI and simulation technologies to jointly

MB

develop AI agents and the simulated environments in which they can be trained, tested
NH
YE

and sometimes deployed. It includes both the use of AI to make simulations more efficient
U

and useful, and the use of a wide range of simulation models to develop more versatile
NG

and adaptive AI systems.

24

Why This Is Important

/20
/12

Increased complexity in decision making is driving demand for both AI and simulation.
04

However, current AI faces challenges, as it is brittle to change and usually requires a lot of
:21

data. Conversely, realistic simulations can be expensive and difficult to build and run. To
12

resolve these challenges, a growing approach is to combine AI and simulation: Simulation

ên_

is used to make AI more robust and compensate for a lack of training data, and AI is used
uy
Ng

to make simulations more efficient and realistic.

ình
hB

Business Impact
ạn

AI simulation can bring:

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 21 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Increased value by broadening AI use to cases where data is scarce, using

ng
simulation to generate synthetic data (for example, synthetic data for generative AI

à
Ho
[GenAI])

_
MB
■ Greater efficiency by leveraging AI to decrease the time and cost to create and use

NH
complex and realistic simulations

E
UY
Greater robustness by using simulation to generate diverse scenarios, increasing AI

NG
■
performance in uncertain environments

4
02
■ Decreased technical debt by reusing simulation environments to train future AI

2/2
models

4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 22 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Limited availability of AI training data is increasing the need for synthetic data

à
Ho
techniques, such as simulation. Simulation techniques, like physics-based 3D

_
MB
simulation, are uniquely positioned to generate diverse AI training datasets. This is

NH
increasingly important for GenAI as training data becomes more scarce.

E
UY
■ Advances in capabilities are making simulation increasingly useful for AI.

NG
Simulation capabilities have been rapidly improving, driven both by increased
computing performance and more-efficient techniques.

4
02
2/2
■ The growing complexity of decision making is increasing interest in AI simulation.

4/1
Simulation is able to generate diverse “corner case” scenarios that do not appear

10
frequently in real-world data, but that are still crucial to train and test AI so it can

2:2
perform well in uncertain environments.

_1
ên
■ Increased technical debt in AI is driving the need for the reusable environments

uy
that simulation provides. Organizations will increasingly deploy hundreds of AI
Ng
models, which requires a shift in focus toward building persistent, reusable
nh
Bì

environments where many AI models can be trained, customized and validated.

nh

Simulation environments are ideal since they are reusable, scalable and enable the
Mạ

training of many AI models at once.

ng
oà

■ The growing sophistication of simulation drives the use of AI, making it more
_H

efficient. Modern simulations are resource intensive. This is driving the use of AI to
MB

accelerate simulation, typically by employing AI models that can replace parts of the
NH
YE

simulation without running resource-intensive, step-by-step numerical computations.

U
NG

■ Research in learned simulations (known as “world models”) is driving interest in AI

simulation: Research is increasing on training world models that can learn to predict
24

how the environment will evolve, based on its current state and agents’ actions.
/20

These learned simulations could make AI simulation more feasible by not having to
/12

directly specify simulation parameters.

04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 23 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Gap between simulation and reality: Simulations can only emulate — not fully

à
Ho
replicate — real-world systems. This gap will reduce as simulation capabilities

_
MB
improve, but it will remain a key factor. Given this gap, AI models trained in

NH
simulation might not have the same performance once they are deployed;

E
differences in the simulation training dataset and real-world data can impact

UY
models’ accuracy.

NG
■ Complexity of AI simulation pipelines: The combination of AI and simulation

4
02
techniques can result in more-complex pipelines that are harder to test, validate,

2/2
maintain and troubleshoot.

4/1
10
■ Limited readiness to adopt AI simulation: A lack of awareness among AI

2:2
practitioners about leveraging simulation capabilities can prevent organizations

_1
from implementing an AI simulation approach.

ên
uy
Fragmented vendor market: The AI and simulation markets are fragmented, with
Ng
■
few vendors offering combined AI simulation solutions, potentially slowing down the
nh
Bì

deployment of this capability.

nh
Mạ

User Recommendations
ng
oà

■ Complement AI with simulation to optimize business decision making or to

_H

overcome a lack of real-world data by offering a simulated environment for synthetic

MB

data generation or reinforcement learning.

NH
YE

■ Complement simulation with AI by applying deep learning to accelerate simulation,

U
NG

and generative AI to augment simulation.

Create synergies between AI and simulation teams, projects and solutions to enable
24

■
/20

a new generation of more-adaptive solutions for ever-more-complex use cases.

/12

Incrementally build a common foundation of more-generalized and complementary

04

models that are reused across different use cases, business circumstances and
:21

ecosystems.
12
n_

■ Prepare for the combined use of AI, simulation and other relevant techniques —such
ê
uy

as graphs, natural language processing or geospatial analytics — by prioritizing

Ng

vendors that offer platforms that integrate different AI techniques (composite AI), as
ình

well as simulation.
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 24 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sample Vendors

ng
Altair; Ansys; The AnyLogic Company; Cosmo Tech; Epic Games; MathWorks; Microsoft;

à
Ho
NVIDIA; Rockwell Automation; Unity

_
MB
NH
Gartner Recommended Reading

E
Innovation Insight: AI Simulation

UY
NG
Predicts 2023: Simulation Combined With Advanced AI Techniques Will Drive Future AI

4
Investments

02
2/2
4/1
Causal AI

10
Analysis By: Pieter den Hamer, Ben Yan, Leinar Ramos

2:2
_1
Benefit Rating: High

ên
uy
Ng
Market Penetration: 1% to 5% of target audience nh
Bì

Maturity: Emerging
nh
Mạ

Definition:
ng
oà

Causal AI identifies and utilizes cause-and-effect relationships to go beyond correlation-

_H

based predictive models and toward AI systems that can prescribe actions more
MB

effectively and act more autonomously. It includes different techniques, such as causal
NH
YE

graphs and simulation, that help uncover causal relationships to improve decision
U

making.
NG

Why This Is Important

24
/20

AI’s ultimate value comes from making better decisions and taking effective actions.
/12

However, the current correlation-based approach has its limitations. It may be fine for
04

prediction, assuming that past and future do not deviate too much, but predicting an
:21

outcome is not the same as understanding what causes it and how to improve it. Causal
12

AI is crucial when we need to be more robust in forecasting and more prescriptive to

ên_

determine the best actions to influence specific outcomes.

uy
Ng
ình

Business Impact
hB

Causal AI leads to:

ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 25 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Greater decision augmentation and autonomy in AI systems by estimating

ng
intervention effects.

à
Ho
■ Greater efficiency by adding domain knowledge to bootstrap AI models with smaller

_
MB
datasets.

NH
E
■ Better explainability by capturing easy-to-interpret cause-and-effect relationships.

UY
NG
■ More robustness and adaptabilty by leveraging causal relationships that remain
valid in changing environments.

4
02
2/2
■ The ability to extract causal knowledge with less costly and time-consuming

4/1
experiments.

10
2:2
■ Reduced bias in AI systems by making causal links more explicit.

_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 26 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Analytics demand is shifting from predictive to more prescriptive capabilities.

à
Ho
Making accurate predictions will remain key, but a causal understanding of how to

_
MB
affect predicted outcomes is increasingly important.

NH
■ AI systems increasingly need to act autonomously, particularly for time-sensitive

E
UY
and complex use cases where human intervention is not feasible. This will only be

NG
possible by AI understanding what impact actions will have and how to make
effective interventions.

4
02
2/2
■ Limited data availability for certain use cases require more data-efficient

4/1
techniques like causal AI. Causal AI leverages human domain knowledge of cause-

10
and-effect relationships to bootstrap AI models in small-data situations.

2:2
_1
■ Growing complexity and dynamics of business requires more robust AI techniques.

ên
The volatility of the last few years has exposed the brittleness of correlation-based

uy
AI models across industries.Causal structure changes much more slowly than
Ng
statistical correlations, making causal AI more robust and adaptable in fast-
nh
Bì

changing environments.
nh
Mạ

■ The need for greater AI trust and explainability is driving interest in models that are
ng

more intuitive to humans. Causal AI techniques, such as causal graphs, make it

oà

possible to be explicit about causes and explain models in terms that humans
_H

understand.
MB
NH

■ Generative AI (GenAI) can accelerate causal AI implementation. GenAI is emerging

YE

as an aid to explore documents and other data sources for existing causal
U
NG

knowledge. This can then be used to generate candidate causal graphs, which, while
still requiring human validation or completion, may reduce time-consuming manual
24

work.
/20
/12

■ The next step in AI requires causal AI. Current deep learning models and, in
04

particular, GenAI, have limitations in terms of their reliability and ability to reason. A
:21

composite AI approach that complements GenAI with causal AI — in particular,

12

causal knowledge graphs — offers a promising avenue to bring AI to a higher level.

ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 27 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Causality is not trivial. Not every phenomenon is easy to model in terms of its

à
Ho
causes and effects. Causality might be unknown, regardless of AI use.

_
MB
■ The quality of a causal AI model depends on its causal assumptions and on the

NH
data used to build it. This data is susceptible to bias and imbalance, and may be

E
UY
incomplete in terms of representing all causal factors, known or unknown.

NG
■ Causal AI requires technical and domain expertise to properly estimate causal

4
effects. Building causal AI models is often more difficult than building correlation-

02
2/2
based predictive models, requiring active collaboration between domain experts and

4/1
AI experts.

10
2:2
■ AI experts might be unaware of causality methods. If AI experts are overly reliant on

_1
data-driven models like machine learning (ML), organizations could get pushback

ên
when looking to implement causal AI.

uy
■ Ng
The vendor landscape is nascent, and enterprise adoption is currently low. This
nh
represents a challenge when organizations run initial causal AI pilots and identify
Bì
nh

specific use cases where causal AI is most relevant.

Mạ
ng

User Recommendations
oà
_H

■ Acknowledge the limitations of correlation-based AI and ML approaches which

MB

focus on leveraging correlations and mostly ignore causality. These limitations also
NH

apply to most GenAI techniques, including foundation models and large language
YE

models.
U
NG

■ Use causal AI when you require more augmentation and automation in decision
intelligence, that is, when AI is needed not only to generate predictions but also to
24
/20

understand how to affect the predicted outcomes. Examples include customer

/12

retention programs, marketing campaign allocation and financial portfolio

04

optimization, as well as smart robotics and autonomous systems.

:21
12

■ Select different causal AI techniques depending on the complexity of the specific use
n_

case. These include causal rules, causal graphs and Bayesian networks, simulation,
ê
uy

and ML for causal learning.

Ng
ình

■ Educate your data science teams on causal AI. Explain the difference between
hB

causal and correlation-based AI, and cover the range of techniques available to
ạn

incorporate causality.
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 28 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sample Vendors

ng
Actable AI; causaLens; Causality Link; Geminos Software; IBM; Microsoft; Parabole.ai;

à
Ho
Scalnyx; Vizuro; Xplain Data

_
MB
NH
Gartner Recommended Reading

E
Innovation Insight: Causal AI

UY
NG
Innovation Insight for Composite AI

4
02
Innovation Insight for Decision Intelligence Platforms

2/2
4/1
10
Case Study: Causal AI to Maximize the Efficiency of Business Investments (HDFC Bank)

2:2
_1
AI-Ready Data

ên
uy
Analysis By: Roxane Edjlali, Svetlana Sicular, Mark Beyer

Ng
nh
Benefit Rating: High
Bì
nh
Mạ

Market Penetration: 1% to 5% of target audience

ng
oà

Maturity: Embryonic
_H
MB

Definition:
NH
YE

The ability to prove the fitness of data for the specific AI use case determines if data is AI-
U

ready data. Proof of readiness comes from the ability to continuously meet AI
NG

requirements by assessing its alignment to the use case, enabling data qualification and
ensuring data governance. As a result, AI-ready data can only be determined contextually
24
/20

to the use case and AI technique used, which forces new approaches to data
/12

management.
04
:21

Why This Is Important

12

With the rise of pretrained off-the-shelf models and hype from generative AI, organizations
ên_

and their data management leaders are at the forefront of creating data strategies for AI
uy
Ng

to ensure that their data is ready to serve AI and underpin data-driven applications. Chief
ình

data and analytics officers and data management leaders need to be able to quickly
hB

respond to AI-ready data demands. It all starts by delivering AI-ready data to support AI
ạn

use cases.
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 29 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
The ability to deliver AI-ready data to support enterprises’ AI strategies will be critical to

à
Ho
delivering on the business value of AI. As a result, all organizations that invest in AI at

_
MB
scale will need to evolve their data management practices and capabilities not only to

NH
preserve the evergreen classical ideas of data management but also to extend them to AI.

E
It will be critical to provision AI-ready data iteratively to cater to existing and upcoming

UY
demands of the business, ensure trust, preserve intellectual property (IP) and reduce bias

NG
and hallucinations.

4
02
Drivers

2/2
4/1
■ Models, especially for generative AI, increasingly come from the vendors rather than

10
being delivered in-house. Data is becoming the main means for enterprises to get

2:2
value from these pretrained models.

_1
ên
■ Most commonly delivered AI solutions depend on data availability, quality and

uy
understanding, not just AI model building. Many enterprises attempt to tackle AI
Ng
without considering AI-specific data management issues. The importance of data
nh
Bì

management in AI is often underestimated, so data management solutions must

nh

now be adjusted for AI needs.

Mạ
ng

■ Classical data management is ripe for disruption to support AI efforts. Rapid

oà

progress of AI poses new challenges in organizing and managing the data for AI. We
_H

expect a cycle of augmented data management techniques that are better suited for
MB

meeting the data requirements of AI. Data ecosystems on the foundation of data
NH

fabric indicate the beginning of this new cycle.

UYE
NG

■ Data management capabilities, practices and tools greatly benefit AI development

and deployment. The AI community invents new data-centric approaches, such as
24

federated machine learning (ML) and retrieval-augmented generation (RAG), which

/20

benefit from data management innovations like data fabric and lakehouse. For
/12

example, implementing a knowledge graph as part of the data fabric allows for a
04

large-language-model-led query to benefit from the context provided by the graph,

:21

which will increase the accuracy of the code generated.

12
ên_

■ New data management solutions mitigate AI-amplified bias originating in data

uy
Ng

interpretation, labeling and human actions recorded in the data. Bias mitigation and
ình

hallucinations are an acute, AI-specific problem that forces data management to

hB

determine how to structure, analyze and prepare data.

ạn
gM

■ Generative AI is removing the distinction between structured and unstructured data,

thereby requiring data management to adapt to these new uses.
àn
Ho
B_

Gartner, Inc. | G00809438 Page 30 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ AI is disconnected from data management. The AI community remains mostly

à
Ho
unaware of data management capabilities, practices and tools that can greatly

_
MB
benefit AI development and deployment, which can lead to challenges when scaling

NH
prototypes in production. Traditional data management also ignores the AI-specific

E
considerations, such as data bias, labeling and drift; this is changing, but slowly.

UY
NG
■ Even though the data side of AI is essential, it is underestimated. It includes tasks
such as preparing datasets and developing a clear understanding of why the data

4
02
was collected a certain way, what the data means and what biases exist in the data.

2/2
4/1
■ Responsible AI requires new governance approaches of both the data and AI model.

10
These are AI-specific data practices that many enterprises want to solve through

2:2
tooling rather than governance.

_1
ên
■ Data management activities don’t end once the model has been developed.

uy
Deployment considerations and ongoing drift monitoring require dedicated data
management activities and practices. Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 31 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Formalize AI-ready data as part of your data management strategy. Implement

à
Ho
active metadata management, data observability and data fabric as foundational

_
MB
components of this strategy. Combine foundational and new capabilities to meet AI

NH
needs. Establish roles and responsibilities to manage data in support of AI.

E
UY
■ Approach AI model development in a data-centric way due to the dependency of AI

NG
models on representative data. Diversify data, models and people to ensure AI value
and ethics.

4
02
2/2
■ Utilize data management expertise, AI engineering, DataOps and MLOps approaches

4/1
to support the AI life cycle. Include data management requirements when deploying

10
models. Develop data monitoring and data governance metrics to ensure that your

2:2
AI models produce the correct output continuously.

_1
ên
■ Enforce policies on data fitness for AI. Define and measure minimum data standards

uy
for AI readiness of data early on for each use case and continuously prove data
Ng
fitness when taking AI to scale. These include checking lineage, quality and
nh
Bì

governance assessment, versioning and automated testing.

nh
Mạ

■ Investigate those data management tools that are rich in augmented data
ng

management capabilities and can integrate well with the AI tools that have created
oà

disruptive data-centric AI capabilities like federated ML and RAG.

_H
MB
NH

Sample Vendors
YE

Databricks; Explorium; Google; illumex; Landing AI; LatticeFlow; Microsoft; MOSTLY AI;
U
NG

Protopia AI; YData

24

Gartner Recommended Reading

/20

Quick Answer: What Makes Data AI-Ready?

/12
04
:21

Innovation Insight: How Generative AI Is Transforming Data Management Solutions

12
n_

Quick Answer: Options for Using Your Data With Generative AI Models
ê
uy
Ng

Successful Generative AI Projects Require Better Metadata Management

ình
hB

When Not to Use Generative AI

ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 32 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Decision Intelligence

ng
Analysis By: Erick Brethenoux, David Pidsley, Pieter den Hamer

à
Ho
_
MB
Benefit Rating: Transformational

NH
E
Market Penetration: 5% to 20% of target audience

UY
NG
Maturity: Emerging

4
02
Definition:

2/2
4/1
Decision intelligence (DI) is a practical discipline that advances decision making by

10
explicitly understanding and engineering how decisions are made, and how outcomes are

2:2
evaluated, managed and improved via feedback.

_1
ên
Why This Is Important

uy
Ng
The current hype around automated decision making and augmented intelligence, fueled
nh
by AI techniques in decision making (including generative AI [GenAI]), is pushing DI toward
Bì

the Peak of Inflated Expectations. Recent crises have revealed the brittleness of business
nh
Mạ

processes. Reengineering those processes to be resilient, adaptable and flexible will

ng

require the discipline of methods and techniques. The fast-emerging decision intelligence
oà

platforms (DIPs) market is starting to provide resilient solutions for decision makers.
_H
MB

Business Impact
NH
YE

■ DI provides better, more timely and optimized decision making by making decisions
U

explicit and transparent. It reduces the unpredictability of decision outcomes by

NG

capturing the business context.

24

■ DI reduces technical debt and increases visibility. It improves the impact of

/20

business processes by materially enhancing the consistency of decision models

/12

based on the power of their relevance and the quality of their transparency, making
04

decisions transparent and auditable.

:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 33 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ A dynamic and complex business environment with an increasingly unpredictable

à
Ho
and uncertain pace of business: Two forces are creating a new market around DIPs.

_
MB
First is the combination of AI techniques such as rules, knowledge graphs and

NH
machine learning (ML). Second is the confluence of technology clusters around

E
composite AI, smart business processes, real-time event processing, insight engines,

UY
decision management and advanced personalization platforms.

NG
■ The need to curtail unstructured, ad hoc decisions that are siloed and disjointed:

4
02
Often uncoordinated, such decisions promote local optimizations at the expense of

2/2
global efficiency.

4/1
10
■ Expanding collaboration between humans and machines: This collaboration,

2:2
supplemented by a lack of trust in technologies, is increasingly replacing tasks and

_1
promoting uneasiness from a human perspective. DI practices promote

ên
transparency, interpretability, fairness, reliability and accountability of decision

uy
Ng
models, critical for the adoption of business-differentiating techniques.
nh
Bì

■ Tighter regulations making risk management more prevalent: From privacy and
nh

ethical guidelines to new laws and government mandates, organizations are facing
Mạ

difficulty in understanding the risk impacts of their decisions. DI promotes explicit

ng

decision models, reducing the risk.

oà
_H

Uncertainty regarding decision consistency across the organization: Lack of

MB

■
explicit representation of decisions prevents proper harmonization of collective
NH
YE

decision outcomes; DI remedies this issue.

U
NG

■ Emergence of software tools in the form of DIPs: DIPs will enable organizations to
practically implement DI projects and strategies.
24
/20

■ GenAI and its synergy with existing DI techniques and practices: The advent of
/12

GenAI offers more efficient and richer context to decision making. It accelerates the
04

research and adoption of composite AI models, which are the foundation of DIPs.
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 34 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Lack of proper coordination between business units: The inability to impartially

à
Ho
reconsider critical decision flows within and across departments diminishes the

_
MB
effectiveness of early DI efforts.

NH
■ Fragmentation: Decision-making silos have created data, competencies and

E
UY
technology clusters that are difficult to reconcile and that could slow down the

NG
implementation of decision models.

4
■ Subpar operational structure: An inadequate organizational structure around

02
2/2
advanced techniques, such as the lack of an AI center of excellence, could impair DI

4/1
progress.

10
2:2
■ Lack of modeling in a wider context: In organizations that have focused almost

_1
exclusively on technical skills, the other critical parts of human decision making —

ên
psychological, social, economic and organizational factors — have gone

uy
unaddressed.
Ng
nh
Lack of AI literacy: Many organizations still suffer from a lack of understanding of
Bì

■
nh

AI techniques. This AI illiteracy could slow down the development of DI projects.

Mạ
ng

User Recommendations
oà
_H

■ Promote the resilience and sustainability of cross-organizational decisions by

MB

building models using principles to enhance traceability, replicability, pertinence and

NH

trustworthiness.
UYE

Improve the predictability and alignment of decision agents by simulating their

NG

■
collective behavior while also estimating their global contribution versus local
optimization.
24
/20

Develop staff expertise in traditional and emerging decision augmentation and

/12

■
04

decision automation techniques, including predictive and prescriptive (optimization

:21

and business rules) analytics. Upskill business analysts, and develop new roles such
12

as decision engineer and decision steward.

ên_
uy

■ Tailor the choice of decision-making technique to the particular requirements of

Ng

each decision situation by collaborating with subject matter experts, AI experts and
ình

business process analysts.

hB
ạn

■ Accelerate the development of DI projects by encouraging experimentation with

gM

GenAI and expediting the deployment of composite AI solutions.

àn
Ho
B_

Gartner, Inc. | G00809438 Page 35 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Innovation Insight for Decision Intelligence Platforms

à
Ho
_
MB
Predicts 2024: How Artificial Intelligence Will Impact Analytics Users

NH
E
Reengineer Your Decision-Making Processes for More Relevant, Transparent and Resilient

UY
Outcomes

NG
Emerging Tech: Venture Capital Growth Insights for Decision Intelligence Platforms

4
02
2/2
Video: How Decision Intelligence Improves Business Outcomes

4/1
10
Neuro-Symbolic AI

2:2
_1
Analysis By: Erick Brethenoux, Afraz Jaffri

ên
uy
Ng
Benefit Rating: High nh
Bì

Market Penetration: Less than 1% of target audience

nh
Mạ

Maturity: Emerging
ng
oà
_H

Definition:
MB

Neuro-symbolic AI is a form of composite AI that combines machine learning (ML)

NH

methods and symbolic systems (for example, knowledge graphs) to create more robust
YE

and trustworthy AI models. This fusion enables the combination of probabilistic models
U
NG

with explicitly defined rules and knowledge to give AI systems the ability to better
represent, reason and generalize concepts. This approach provides a reasoning
24

infrastructure for solving a wider range of business problems more effectively.

/20
/12

Why This Is Important

04
:21

Neuro-symbolic AI is important because it addresses limitations in current AI systems,

12

such as incorrect outputs, lack of generalization to a variety of tasks and an inability to

n_

explain the steps that led to an output. This leads to more powerful, versatile and
ê
uy

interpretable AI solutions and allows AI systems to tackle more complex tasks with
Ng

humanlike reasoning.
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 36 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
Neuro-symbolic AI will have an impact on the efficiency, adaptability and reliability of AI

à
Ho
systems used across business processes. The integration of logic and multiple reasoning

_
MB
mechanisms brings down the need for ever larger AI models and their supporting

NH
infrastructure. These systems will rely less on the processing of huge amounts of data,

E
making AI agile and resilient. Neuro-symbolic approaches can augment and automate

UY
decision making with less risk of unintended consequences.

NG
Drivers

4
02
Limitations of AI models based exclusively on ML techniques that focus on

2/2
■

4/1
correlation over understanding and reasoning. The newest generation of large

10
language models is well-known for its tendency to give factually incorrect answers

2:2
or produce unexpected results. Neuro-symbolic AI addresses these limitations.

_1
ên
■ The need for explanation and interpretability of AI outputs that are especially

uy
important in the regulated industry use cases and in systems that use private data.
Ng
nh
■ The need to prioritize understanding the meanings behind words, not just their
Bì

arrangement (semantics over syntax) in systems that deal with real-world entities to
nh
Mạ

ground meaning to words and terms in specific domains.

ng
oà

■ The set of tools available to combine different types of AI models is increasing and
_H

becoming easier to use for developers, data scientists and end users. The dominant
MB

approach is to chain together results from different models (composite AI) rather
NH

than using single models.

UYE

The integration of multiple reasoning mechanisms necessary to provide agile AI

NG

■
systems eventually leads to adaptive AI systems, notably through blackboardlike
mechanisms.
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 37 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Most neuro-symbolic AI methods and techniques are being developed in academia

à
Ho
or industry research labs. Despite the increased availability of tools, there are still

_
MB
limited implementations in business or enterprise settings.

NH
■ There are no agreed-upon techniques for implementing neuro-symbolic AI, and

E
UY
disagreements continue between researchers and practitioners on the effectiveness

NG
of combining approaches, despite the emergence of real-world use cases.

4
■ The commercial and investment trajectories for AI startups allocate almost all

02
2/2
capital to deep learning approaches, leaving only those willing to bet on the future to

4/1
invest in neuro-symbolic AI development.

10
2:2
■ Currently, popular media and academic conferences do not give as much exposure

_1
to the neuro-symbolic AI movement as compared to other approaches.

ên
uy
Ng
User Recommendations nh
■ Adopt composite AI approaches when building AI systems by utilizing a range of
Bì

techniques that increase the robustness and reliability of AI models. Neuro-symbolic

nh
Mạ

AI approaches will fit into a composite AI architecture.

ng
oà

■ Dedicate time to learning and applying neuro-symbolic AI approaches by identifying

_H

use cases that can benefit from these approaches.

MB
NH

■ Invest in data architecture that can leverage the building blocks for neuro-symbolic
YE

AI techniques, such as knowledge graphs and agent-based techniques.

U
NG

■ Consider neuro-symbolic AI architectures when the limitations of generative AI

models prevent their implementation in the organization.
24
/20
/12

Sample Vendors
04

Elemental Cognition; Franz; Google DeepMind; IBM; Microsoft; RelationalAI;

:21

Wolfram|Alpha
12
ên_

Gartner Recommended Reading

uy
Ng

Innovation Insight: AI Simulation

ình
hB

Go Beyond Machine Learning and Leverage Other AI Approaches

ạn
gM

When Not to Use Generative AI

àn
Ho
B_

Gartner, Inc. | G00809438 Page 38 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Predicts 2023: Simulation Combined With Advanced AI Techniques Will Drive Future AI

ng
Investments

à
Ho
_
Composite AI

MB
NH
Analysis By: Erick Brethenoux, Pieter den Hamer

E
UY
Benefit Rating: Transformational

NG
Market Penetration: 20% to 50% of target audience

4
02
2/2
Maturity: Early mainstream

4/1
10
2:2
Definition:

_1
Composite AI refers to the combined application (or fusion) of different AI techniques to

ên
improve the efficiency of learning to broaden the level of knowledge representations. It

uy
Ng
broadens AI abstraction mechanisms and, ultimately, provides a platform to solve a wider
nh
range of business problems effectively.
Bì
nh

Why This Is Important

Mạ
ng

Composite AI recognizes that no single AI technique is a panacea. It aims to combine

oà

“connectionist” AI approaches, like machine learning (ML) and deep learning, with
_H

“symbolic” and other AI approaches, like rule-based reasoning, graph analysis or

MB

optimization techniques. The goal is to enable AI solutions that require less data and
NH

energy to learn, embodying more abstraction mechanisms. Composite AI is at the center

YE

of the generative AI (GenAI) and decision intelligence (DI) markets.

U
NG

Business Impact
24
/20

Composite AI brings the power of AI to a broader group of organizations that do not have
/12

access to large amounts of historical or labeled data but possess significant human
04

expertise. It helps to expand the scope and quality of AI applications (that is, more types
:21

of reasoning challenges). Other benefits include better interpretability and embedded

12

resilience and the support of augmented intelligence. The new wave of GenAI
n_

implementations heavily relies on composite AI.

ê
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 39 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ The growing reliance on AI for decision making is driving organizations toward

à
Ho
composite AI. The most appropriate actions can be further determined by combining

_
MB
rule-based and optimization models — a combination often referred to as

NH
prescriptive analytics.

E
UY
■ Small datasets, or the limited availability of data, have pushed organizations to

NG
combine multiple AI techniques. Enterprises have started to complement scarce raw
historical data with additional AI techniques, such as knowledge graphs and

4
02
generative adversarial networks (GANs), to generate synthetic data.

2/2
4/1
■ Combining AI techniques is much more effective than relying only on heuristics or a

10
fully data-driven approach. A heuristic or rule-based approach can be combined with

2:2
a deep learning model (for example, predictive maintenance). Rules coming from

_1
human experts, or the application of physical/engineering model analysis, may

ên
specify that certain sensor readings indicate inefficient asset operations. This can be

uy
Ng
used as a feature to train a neural network to assess and predict the asset’s health,
nh
also integrating causal AI capabilities.
Bì
nh

■ The proliferation of computer vision and natural language processing (NLP)

Mạ

solutions is used for identifying or categorizing people or objects in an image. This

ng

output can be used to enrich or generate a graph, representing the image entities and
oà
_H

their relationships.
MB

Agent-based modeling is the next wave of composite AI. A composite AI solution is

NH

■
YE

composed of multiple agents, each representing an actor in the ecosystem.

U

Combining these agents into a “swarm” enables the creation of common situation
NG

awareness, more global planning optimization, responsive scheduling and process

resilience.
24
/20

■ The advent of GenAI is accelerating the research and adoption of composite AI

/12

models through artifacts, process and collaboration generations, which are the
04

foundation of DI platforms.
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 40 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Lack of awareness and skills in leveraging multiple AI methods: This could prevent

à
Ho
organizations from considering the techniques particularly suited to solving specific

_
MB
problem types.

NH
■ Deploying ModelOps: The ModelOps domain (that is, the operationalization of

E
UY
multiple AI models, such as optimization models, rule models and graph models)

NG
remains an art much more than a science. A robust ModelOps approach is required
to efficiently govern composite AI environments and harmonize it with other

4
02
disciplines, such as DevOps and DataOps.

2/2
4/1
■ Trust and risk barriers: The AI engineering discipline is starting to take shape, but

10
only mature organizations apply its benefits in operationalizing AI techniques.

2:2
Security, ethical model behaviors, observability, model autonomy and change

_1
management practices must be addressed across the combined AI techniques.

ên
uy
User Recommendations Ng
nh
Bì

■ Identify projects in which a fully data-driven, ML-only approach is inefficient or ill-

nh

fitted. For example, in cases when enough data is not available or when the pattern
Mạ

cannot be represented through current ML models.

ng
oà

■ Capture domain knowledge and human expertise to provide context for data-driven
_H

insights by applying decision management with business rules and knowledge

MB

graphs, in conjunction with ML and/or causal models.

NH
YE

■ Combine the power of ML, image recognition or NLP with graph analytics to add
U
NG

higher-level, symbolic and relational intelligence.

Extend the skills of ML experts, or recruit/upskill additional AI experts, to cover graph

24

■
/20

analytics, optimization or other techniques for composite AI. For rules and heuristics,
/12

consider knowledge engineering skills, as well as emerging skills such as prompt

04

engineering.
:21
12

■ Accelerate the development of DI projects by encouraging experimentation with

n_

GenAI, which will in turn accelerate the deployment of composite AI solutions.

ê
uy
Ng
ình

Sample Vendors
hB

ACTICO; Aera Technology; FICO; Frontline Systems; IBM; Indico Data; Peak; SAS
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 41 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Go Beyond Machine Learning and Leverage Other AI Approaches

à
Ho
_
MB
When Not to Use Generative AI

NH
E
Top Strategic Technology Trends for 2022: AI Engineering

UY
NG
How to Choose Your Best-Fit Decision Management Suite Vendor

4
02
Artificial General Intelligence

2/2
4/1
Analysis By: Pieter den Hamer

10
2:2
Benefit Rating: Transformational

_1
ên
Market Penetration: Less than 1% of target audience

uy
Ng
nh
Maturity: Embryonic
Bì
nh

Definition:
Mạ
ng

Artificial general intelligence (AGI), also known as strong AI, is the (currently hypothetical)
oà

intelligence of a machine that can accomplish any intellectual task that a human can
_H

perform. AGI is a trait attributed to future autonomous AI systems that can achieve goals
MB

in a wide range of real or virtual environments at least as effectively as humans can.

NH
YE

Why This Is Important

U
NG

As AI becomes more sophisticated and powerful, with recent great advances in generative
AI (GenAI) in particular, a growing number of people see AGI as no longer purely
24
/20

hypothetical. Improving the understanding of at least the concept of AGI is critical for
/12

steering and regulating AI’s further evolution. It is also important to manage realistic
04

expectations and to avoid prematurely anthropomorphizing AI. However, if AGI becomes

:21

real, its impact on the economy, (geo)politics, culture and society cannot be
12

underestimated.
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 42 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
In the short term, organizations must know that the hype about AGI exists today among

à
Ho
many stakeholders, stoking fears and unrealistic expectations about current AI’s true

_
MB
capabilities. This AGI anticipation is already accelerating the emergence of more AI

NH
regulations and affects people’s trust and willingness to apply AI today. In the long term,

E
AI continues to grow in power and, with or without AGI, will increasingly impact

UY
organizations, including the advent of machine customers and autonomous business.

NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 43 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Recent great advances in applications of GenAI and the use of foundation models

à
Ho
and large language or multimodal models drive considerable hype about AGI. These

_
MB
advances have been enabled largely by the massive scaling of deep learning, as well

NH
as by the availability of huge amounts of data and compute power. To further evolve

E
AI toward AGI, however, current AI will need to be complemented by other (partially

UY
new) approaches, such as knowledge graphs, multiagent systems, simulations,

NG
evolutionary algorithms, causal AI, composite AI and likely other innovations yet
unknown.

4
02
2/2
■ Vendors such as Google, IBM and OpenAI are openly discussing and actively

4/1
researching the field of AGI, creating the impression that AGI lies within reach.

10
However, their definitions of AGI vary greatly and are often open to multiple

2:2
interpretations.

_1
ên
Humans’ innate desire to set lofty goals is also a major driver for AGI. At one point in

uy
■

Ng
history, humans wanted to fly by mimicking bird flight. Today, airplane travel is a
nh
reality. The inquisitiveness of the human mind, taking inspiration from nature and
Bì

from itself, is not going to fizzle out.

nh
Mạ

■ People’s tendency to anthropomorphize nonliving entities also applies to AI-powered

ng

machines. This has been fueled by the humanlike responses of ChatGPT and similar
oà
_H

AI, as well as AI being able to pass several higher-level education exams.

MB

Complex AI systems display behavior that has not been explicitly programmed.
NH

■
YE

Among other reasons, this results from the dynamic interactions between many
U

system components. Consequently, AI is increasingly attributed with humanlike

NG

characteristics, such as understanding. Although many philosophers,

neuropsychologists and other scientists consider this attribution as going too far or
24
/20

being highly uncertain, it has created a sense that AGI is within reach or at least is
/12

getting closer. In turn, this has triggered massive media attention, several calls for
04

regulation to manage the risks of AGI and a great appetite to invest in AI for
:21

economic, societal and geopolitical reasons.

12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 44 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Unreliability, lack of transparency and limited reasoning capabilities in current AI are

à
Ho
not easy to overcome with the intrinsically probabilistic approach of deep learning.

_
MB
More data or more compute power for ever-bigger models are unlikely to resolve

NH
these issues, let alone to achieve AGI.

E
UY
■ The meanings of “intelligence” and related terminology like “understanding” have

NG
little scientific consensus, including the definition and interpretation of AGI. Scientific
understanding about human intelligence is still challenged by the enormous

4
02
complexity of the human brain and mind. Any claims about AGI — in whatever form

2/2
it may emerge — are hard to validate when we don’t even understand human

4/1
intelligence. However, even when AGI will be properly understood and defined, further

10
technological innovations will likely be needed to implement AGI. Therefore, AGI as

2:2
defined here is unlikely to emerge in the near future.

_1
ên
If AGI materializes, it is likely to lead to the emergence of autonomous actors that, in

uy
■

Ng
time, will be attributed with full self-learning, agency, identity and perhaps even
nh
morality. This will open up a bevy of legal rights of AI and trigger profound ethical
Bì

and even religious discussions. AGI also brings the risk of negative impacts on
nh

humans, from job losses to a new, AI-triggered arms race and more. This may lead to
Mạ

serious backlash, and regulations to ban or control AGI are likely to emerge in the
ng
oà

near future.
_H
MB

User Recommendations
NH
YE

■ Engage with stakeholders to address their concerns and create or maintain realistic
U

expectations. Today, people may be either overly concerned about future AI replacing
NG

humanity or overly excited about current AI’s capabilities and impact on business.
Both cases hamper a realistic and effective approach to using AI today.
24
/20

Stay apprised of scientific and innovative breakthroughs that may indicate the
/12

■
04

possible emergence of AGI. Meanwhile, keep applying current AI to learn, reap its
:21

benefits and develop practices for its responsible use.

12
n_

■ Although AGI is not a reality now, current AI already poses significant risks regarding
ê
uy

bias, reliability and other areas. Prepare for emerging AI regulations and promote
Ng

internal AI governance to manage current and emerging future risks of AI.

ình
hB

Sample Vendors
ạn
gM

Aigo; Google; IBM; Microsoft; OpenAI

àn
Ho
B_

Gartner, Inc. | G00809438 Page 45 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
The Future of AI: Reshaping Society

à
Ho
_
MB
Innovation Insight for Generative AI

NH
E
Innovation Insight: AI Simulation

UY
NG
Applying AI — Key Trends and Futures

4
02
Innovation Insight for Artificial Intelligence Foundation Models

2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 46 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
At the Peak

ng
Sovereign AI

à
Ho
Analysis By: Lydia Clougherty Jones, Clementine Valayer

_
MB
NH
Benefit Rating: High

E
UY
NG
Market Penetration: 20% to 50% of target audience

4
Maturity: Adolescent

02
2/2
4/1
Definition:

10
Sovereign AI is the effort by nation-states to self-operationalize their development and use

2:2
of AI with less dependence on the commercial market. It embodies political and cultural

_1
ên
differences to advance sovereign objectives, including when developing AI strategy for

uy
value alongside sovereign-appropriate harm reduction. Given the wide variances across
Ng
sovereign AI innovation to loss ratios, sovereign AI impacts international relationships,
nh
global trade and economic markets in unexpected ways.
Bì
nh
Mạ

Why This Is Important

ng

Sovereign AI reflects the rapid acceleration of nation-states’ adoption of AI techniques for

oà

their own use to improve alignment of nearly all of their internal government functions
_H
MB

and activities with operational goals. While it could enhance an individual state’s military
NH

defense, AI use by other nations could undermine those national security efforts.
YE

Sovereign AI aims to maximize AI value while decreasing AI risk, including for those
U

sovereign states who collaborate to achieve common goals such as decreasing the
NG

impact of AI-generated “deepfakes” in political environments.

24
/20

Business Impact
/12

Sovereign AI impacts nearly all aspects of government and the enterprises with whom it
04

interacts. It improves the effectiveness of operations by automating tasks, such as within

:21

government contact centers. Because sovereign AI modernizes the business of

12
n_

government, it can improve employee experience and accelerate citizen engagement.

ê

When sovereign states control their own AI systems, they reduce their dependence on
uy
Ng

other sovereign states and on the private tech market.

ình
hB

Drivers
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 47 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ An increasing number of countries are actively planning and building their own AI

ng
infrastructure and capabilities to gear up their competitiveness and safeguard their

à
Ho
future; they are developing what they consider their sustainable “sovereign AI.”

_
MB
■ Known and unknown risks of harms to citizens and society from irresponsible uses

NH
of AI drive sovereign states to want more control over the development of AI

E
UY
systems, and more so over generative AI (GenAI) government use cases.

NG
■ Increasing need for a sovereign entity to self-regulate including how its data is used
to train large language models (LLMs). For example, nation-states are increasingly

4
02
using AI tools to make important government decisions, but we hear from clients

2/2
that these decisions are often outsourced to private companies without public input

4/1
10
or oversight. This lack of transparency and accountability drives sovereign states to

2:2
develop the AI tools themselves to address concerns about potential biases and

_1
conflicts of interest in these critical decision-making processes.

ên
uy
■ Desire to decrease dependencies on other nations and the tech market, including
Ng
when underrepresentation of cultural and linguistic inputs cannot be achieved.
nh
Bì

■ Sovereign AI plays an important role in digital sovereignty as it focuses on the

nh
Mạ

sovereign control of AI data and systems, including control over computing capacity,
ng

data storage, access to human resources and proprietary knowledge for AI

oà

application development. Digital sovereignty can also significantly impact sovereign

_H

AI development, with the availability of locally stored data, for example, to train AI
MB

models.
NH
YE

■ Sovereign AI is different from sovereign data strategies as the former’s core focus is
U
NG

being the developer and user of AI technologies, not the regulator of it. Sovereign
data strategies reflect state efforts to regulate data and AI use by and about its
24

citizens, private industry and its economy.

/20
/12

■ Combating threats to political stability from the proliferation of deepfakes.

04
:21

■ Upskilling government workers today for a more AI-ready government workforce

12

tomorrow.
ên_
uy

■ Meeting the increasing demand to advance local and national defense strategies.
Ng

Progressing and maintaining leadership in the emerging technologies space.

ình

■
hB
ạn

Obstacles
gM

■ Preparing an AI-ready IT infrastructure.

àn
Ho
B_

Gartner, Inc. | G00809438 Page 48 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Creating an AI-skilled government workforce.

ng
à
■ Modernizing the government culture to embrace advanced analytics and

Ho
automation.

_
MB
NH
■ Overcoming the pressures on an already-taxed IT infrastructure and fragmented

E
business networks to develop and implement its own AI systems.

UY
NG
■ Lack of the right data for training LLMs, resulting in AI output with varying utility.

4
Lack of technically skilled humans to loop into the AI development and use life cycle,

02
■

2/2
resulting in an increase of unintended negative outcomes from AI use.

4/1
10
■ Diversity of needs and support across political and cultural variations impede

2:2
nation-states from accelerating AI adoption across its government functions.

_1
ên
■ Differences in political and cultural values will create inconsistent AI-value versus AI-

uy
harms analysis, leading to unpredictable impacts on international trade and global
markets. Ng
nh
Bì

■ Development of AI by nation-states across the world will lead to a fragmentation and

nh
Mạ

possibly contradiction in the requirements for AI solutions, many of which cannot be

ng

met by either the public or private sector.

oà
_H
MB

User Recommendations
NH

Sovereign states seeking a self-governed and controlled approach to the development of

YE

AI systems aligned with their strategic objectives should:

U
NG

■ Start small and prioritize the AI uses aligned with maximum advancement of your
24

stakeholder and government business goals.

/20
/12

■ Build an AI strategic roadmap that progresses from internal use cases to citizen-
04

facing ones.
:21
12

■ Ensure that the AI strategy identifies key value opportunities and risks.
ên_

Monitor and learn from sovereign AI already underway, including from New Zealand,
uy

■
Ng

the European Commission, India the United States and the United Kingdom.
ình

Collaborate with (friendly) sovereign states to accelerate the learning curve, sharing
hB

■
failure analysis and positive narratives of unexpected success.
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 49 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Use Gartner’s The Pillars of a Successful Artificial Intelligence Strategy to guide the

ng
nation toward self-governance of AI development, creating tangible value and

à
Ho
achieving competitive national leader status.

_
MB
NH
Gartner Recommended Reading

E
UY
Top Trends in AI Public Policy and Regulations for 2024

NG
The Future of AI: Reshaping Society

4
02
2/2
Quick Answer: Why Is Empathy Critical for Postdigital Government?

4/1
10
Government Insight: U.S. Federal AI Executive Order Opportunities and Risks

2:2
_1
The Impact of the ‘U.S. Executive Order on AI’

ên
uy
Ng
AI TRiSM nh
Analysis By: Avivah Litan, Bart Willemsen, Jeremy D'Hoinne
Bì
nh
Mạ

Benefit Rating: High

ng
oà

Market Penetration: 5% to 20% of target audience

_H
MB

Maturity: Early mainstream

NH
UYE

Definition:
NG

AI trust, risk and security management (AI TRiSM) ensures AI governance, trustworthiness,
fairness, reliability, robustness, efficacy and data protection. AI TRiSM includes solutions
24
/20

and techniques for model and application transparency, content anomaly detection, AI
/12

data protection, model and application monitoring and operations, adversarial attack
04

resistance, and AI application security.

:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 50 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Why This Is Important

ng
AI models and applications should be subject to protection mechanisms during

à
Ho
development and at runtime. Doing so ensures sustained value generation and acceptable

_
MB
use based on predetermined intentions. Accordingly, AI TRiSM is a framework that

NH
comprises a set of risk, privacy and security controls and trust enablers that helps

E
enterprises govern and manage AI models and applications’ life cycles — and accomplish

UY
business goals. The benefit is improved outcomes and performance and enhanced

NG
compliance with regulations such as the EU AI Act.

4
02
Business Impact

2/2
4/1
Organizations that do not consistently manage AI risks are exponentially inclined to

10
experience adverse outcomes such as project failures, AI misperformance and

2:2
compromised data confidentiality. Inaccurate, unethical or unintended AI outcomes,

_1
process errors, uncontrolled biases, and interference from benign or malicious actors can

ên
result in security failures, financial and reputational loss, or liability and social harm. AI

uy
Ng
misperformance can also lead organizations to make suboptimal or incorrect business
nh
decisions.
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 51 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ OpenAI’s ChatGPT democratized third-party generative AI applications and

à
Ho
transformed how enterprises compete and do work. Accordingly, the risks associated

_
MB
with hosted, cloud-based generative AI applications are significant and rapidly

NH
evolving.

E
UY
■ Democratized, third-party AI applications often pose considerable data

NG
confidentiality risks. This is partly because large, sensitive datasets used to train AI
models often come from various sources, including data shared by users of these

4
02
applications.

2/2
4/1
■ Confidential data access must be carefully controlled to avoid adverse regulatory,

10
commercial and reputational consequences.

2:2
_1
■ AI risk and security management imposes new operational requirements that are not

ên
fully understood and cannot be addressed by existing systems. New vendors are

uy
filling this gap.
Ng
nh
AI models and applications must be constantly monitored to ensure that
Bì

■
nh

implementations are compliant, fair and ethical. Risk management tools can identify
Mạ

and adjust bias controls where needed in both (training) data and algorithmic
ng

functions.
oà
_H

■ AI outputs that are unchecked can steer organizations into faulty decision making or
MB

harmful acts because of inaccurate, illegal or fictional information driving business

NH

decisions.
UYE

AI model and application explainability and expected behavior must be constantly

NG

■
tested through observation and testing of model and application outputs. Doing so
24

ensures original explanations, interpretations and expectations of AI models and

/20

applications remain active during model and application operations. If they don’t,
/12

corrective actions must be taken.

04
:21

■ Detecting and stopping adversarial attacks on AI requires new methods that most
12

enterprise security systems do not offer.

ên_
uy

■ Regulations for AI risk management — such as the EU AI Act and other regulatory
Ng

frameworks in North America, China and India — are driving businesses to institute
ình

measures for managing AI model and application risk. Such regulations define the
hB

new compliance requirements organizations will have to meet on top of existing

ạn

ones, like those pertaining to privacy protection.

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 52 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ AI TRiSM is often an afterthought. Organizations generally don’t consider it until

à
Ho
models or applications are in production.

_
MB
■ Enterprises interfacing with hosted, large language models (LLMs) are missing

NH
native capabilities to automatically filter inputs and outputs — for example,

E
UY
confidential data policy violations or inaccurate information used for decision

NG
making. Also, enterprises must rely on vendor licensing agreements to ensure their
confidential data remains private in the host environment.

4
02
2/2
■ Once models and applications are in production, AI TRiSM becomes more

4/1
challenging to retrofit to the AI workflow, thus creating inefficiencies and opening the

10
process to potential risks.

2:2
_1
■ Off-the-shelf software that embeds AI is often closed and does not support APIs to

ên
third-party products that can enforce enterprise policies.

uy
■ Ng
Most AI threats are not fully understood and not effectively addressed.
nh
Bì

■ AI TRiSM requires a cross-functional team, including legal, compliance, security, IT

nh

and data analytics staff, to establish common goals and use common frameworks,
Mạ

which is difficult to achieve.

ng
oà
_H

■ Although challenging, the integration of life cycle controls can be done with AI
MB

TRiSM.
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 53 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Set up an organizational unit to manage AI TRiSM. Include members with a vested

à
Ho
interest in AI projects.

_
MB
■ Define acceptable use policies at a level granular enough to enforce.

NH
E
UY
■ Implement data classification and permissioning systems to ensure enterprise

NG
policies can be enforced.

■ Establish a system to record and approve all AI-based applications and gain periodic

4
02
user attestation that they are used according to preset intentions.

2/2
4/1
■ Use appropriate AI TRiSM toolsets to manage AI model, application, and agent trust

10
risk and security.

2:2
_1
■ Require vendors with AI components to provide verifiable attestations of expected AI

ên
behavior.

uy
Ng
Implement AI data protection solutions and use different methods for different use
nh
■
Bì

cases and components.

nh
Mạ

■ Establish data protection and privacy assurances in license agreements with

ng

vendors hosting LLMs.

oà
_H

■ Constantly validate and test the security, safety and risk posture of all AI used in
MB

your organization, no matter the footprint.

NH
UYE

Sample Vendors
NG

Aporia; Bosch Global Software Technologies (AIShield); Harmonic; Lasso Security;

ModelOp; Prompt Security; Protopia AI; TrojAI
24
/20
/12

Gartner Recommended Reading

04

Top Strategic Technology Trends for 2024: AI Trust, Risk and Security Management
:21
12
n_

Innovation Guide for Generative AI in Trust, Risk and Security Management

ê
uy
Ng

Market Guide for AI Trust, Risk and Security Management

ình
hB

Tool: Generative AI Policy Kit

ạn
gM

Quick Answer: The EU AI Act and Its Anticipated Impact

àn
Ho
B_

Gartner, Inc. | G00809438 Page 54 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Prompt Engineering

ng
Analysis By: Frances Karamouzis, Jim Hare, Afraz Jaffri

à
Ho
_
MB
Benefit Rating: High

NH
E
Market Penetration: 5% to 20% of target audience

UY
NG
Maturity: Emerging

4
02
Definition:

2/2
4/1
Prompt engineering is the discipline of providing inputs, in the form of text or images, to

10
generative AI (GenAI) models to specify and confine the set of responses the model can

2:2
produce. The inputs prompt a set that produces a desired outcome without updating the

_1
actual weights of the model (as done with fine-tuning). Prompt engineering is also referred

ên
to as “in-context learning,” where examples are provided to further guide the model.

uy
Ng
nh
Why This Is Important
Bì

Prompt engineering is the linchpin to business alignment for desired outcomes. It is

nh
Mạ

important because large language models (LLMs) and GenAI models in general are
ng

extremely sensitive to nuances and small variations in input. A slight tweak can change
oà

an incorrect answer to one that is usable as an output. Each model has its own sensitivity
_H

level, and the discipline of prompt engineering is to uncover the sensitivity through
MB

iterative testing and evaluation.

NH
YE

Business Impact
U
NG

Prompt engineering has the following business impacts:

24
/20

■ Performance: It helps improve model performance and reduce hallucinations.

/12

■ Business alignment: It allows subject data scientists, subject matter experts and
04

software engineers to steer foundation models, which are general-purpose in nature,

:21
12

to align to the business, domain and industry.

ên_

Time to market, quality, efficiency and effectiveness: There are a number of

uy

■
Ng

architecture options as well as execution options that AI leaders must balance. There
ình

is also a myriad of prompt optimization tools that will diminish (or at the very least
hB

shift) the need for manual engineering.

ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 55 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Balance and efficiency: The fundamental driver for prompt engineering is it allows

à
Ho
organizations to strike a balance between consuming an “as is” offering versus

_
MB
pursuing a more expensive and time-consuming approach of fine-tuning. GenAI

NH
models, and in particular LLMs, are pretrained, so the data that enterprises want to

E
use with these models cannot be added to the training set. Instead, prompts can be

UY
used to feed content to the model with an instruction to carry out a function.

NG
■ Process or task-specific customizations or new use cases: The insertion of context

4
02
and patterns that a model uses to influence the output generated allows for

2/2
customizations for a particular enterprise or domain, or regulatory items. Prompts

4/1
are created to help improve the quality for different use cases — such as domain-

10
specific question answering, summarization, categorization, and so on — with or

2:2
without the need for fine-tuning a model, which can be expensive or impractical. This

_1
ên
would also apply to creating and designing new use cases that utilize the model’s

uy
capability for image and text generation.
Ng
nh
■ Validation and verification: It is important to test, understand and document the
Bì

limits and weaknesses of the models to ensure a reduced risk of hallucination and
nh

unwanted outputs.
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 56 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Prompt engineering is a new discipline: The craft of designing and optimizing user

à
Ho
requests to an LLM or LLM-based chatbot to get the most effective result is still

_
MB
emerging. Engineers are finding that desired outputs using GenAI can be challenging

NH
to create, debug, validate and repeat. Communities worldwide are developing new

E
prompt engineering methods and techniques to help achieve these desirable

UY
outcomes.

NG
■ Approaches, techniques and scalability: A unified approach to performing prompt

4
02
engineering does not exist. Complex scenarios need to be broken down into smaller

2/2
elements. It is challenging to debug complex prompts. Understanding how specific

4/1
prompt elements influence the logic of the LLM is vital. Scalable and maintainable

10
methods of prompt engineering are still a work in progress for most organizations.

2:2
_1
■ Role alignment: Data scientists are critical to understanding the capabilities and

ên
limits of models, and to determining whether to pursue a purely prompt-based or

uy
Ng
fine-tuning-based approach (or combination of approaches) for customization. The
nh
ultimate goal is to use machine learning itself to generate the best prompts and
Bì

achieve automated prompt optimization. This is in contrast to an end user of an

nh

LLM who concentrates on prompt design to manually alter prompts to give better
Mạ

responses.
ng
oà
_H

User Recommendations
MB
NH

■ Build awareness and understanding of prompt engineering to quickly start the

YE

journey of shape-shifting the appropriate prompt engineering discipline and teams.

U
NG

■ Build critical skills among different team members that will synergistically contribute
critical elements. For example, there are important roles for data scientists, business
24

users, domain experts, software engineers and citizen developers.

/20
/12

■ Educate the team with the myriad of options of prompt optimization tools that will
04

diminish (or at the very least shift) the need for manual engineering.
:21
12

■ Communicate and cascade the message that prompt engineering is not foolproof.
ên_

Enterprise teams apply rigor and diligence to permeate and work to ensure
uy
Ng

successful solutions.
ình
hB

Sample Vendors
ạn

FlowGPT; Google; HoneyHive; Magniv; Microsoft; PromptBase; Salesforce

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 57 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
How to Engineer Effective Prompts for Large Language Models

à
Ho
_
MB
Prompt Engineering With Enterprise Information for LLMs and GenAI

NH
E
Quick Answer: How Will Prompt Engineering Impact the Work of Data Scientists?

UY
NG
Generative AI Changes Software Engineering Leaders’ Responsibilities

4
02
Responsible AI

2/2
4/1
Analysis By: Svetlana Sicular, Philip Walsh

10
2:2
Benefit Rating: Transformational

_1
ên
Market Penetration: 5% to 20% of target audience

uy
Ng
nh
Maturity: Adolescent
Bì
nh

Definition:
Mạ
ng

Responsible artificial intelligence (RAI) is an umbrella term for aspects of making

oà

appropriate business and ethical choices when adopting AI. These include business and
_H

societal value, risk, trust, transparency, fairness, bias mitigation, explainability,

MB

sustainability, accountability, safety, privacy, and regulatory compliance. RAI

NH

encompasses organizational responsibilities and practices that ensure positive,

YE

accountable and ethical AI development and operation.

U
NG

Why This Is Important

24
/20

Early exploitation of generative AI resulted in the re-emergence of RAI as a key AI topic. As

/12

AI amplifies at a huge scale, with both good and bad outcomes, RAI enables the right
04

outcomes by ensuring business value while mitigating risks. RAI can employ a set of tools
:21

and approaches, including industry-specific methods, adopted by vendors and enterprises.

12

More jurisdictions introduce new regulations that drive and challenge organizations to
n_

adopt RAI practices.

ê
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 58 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
RAI assumes accountability for AI development and use at the individual, organizational

à
Ho
and societal levels. If AI governance is practiced by designated groups, RAI extends its

_
MB
reach to all stakeholders involved in the AI process. RAI helps achieve fairness, even

NH
though biases are often baked into the data; gain trust, although transparency and

E
explainability methods are evolving; and ensure regulatory compliance, despite the AI’s

UY
probabilistic nature.

NG
Drivers

4
02
RAI helps AI participants develop, implement, utilize and address the various drivers they

2/2
4/1
face. With further AI adoption, the RAI drivers are becoming more important and are better

10
understood by vendors, buyers, society and legislators:

2:2
_1
■ The adoption of GenAI raises new concerns, such as hallucinations, leaked sensitive

ên
data, copyright issues and reputational risks that bring new actors in RAI (for

uy
Ng
example, in security, legal and procurement). nh
Bì

■ Leading vendors are offering indemnification of their GenAI offerings, making

nh

customers more confident as part of their RAI approaches: although a good step,
Mạ

these are still incomplete.

ng
oà

■ The organizational driver of RAI assumes the need to strike a balance between the
_H

business value and associated risks within regulatory, business and ethical
MB

boundaries. This includes considerations such as reskilling employees to adapt to AI

NH

technologies and safeguarding intellectual property.

UYE
NG

■ The societal driver includes resolving AI safety for societal well-being versus limiting
human freedoms. Existing and pending legal guidelines and regulations, such as the
24

EU’s Artificial Intelligence Act, make RAI a necessity.

/20
/12

■ The customer/citizen driver is based on fairness and ethics and requires reconciling
04

privacy with convenience. Customers/citizens may be willing to share their data in

:21

exchange for certain benefits.

12
n_

■ AI affects all ways of life and touches all societal strata; hence, the RAI challenges
ê
uy

are multifaceted and cannot be easily generalized. New problems will continue to
Ng

arise with rapidly evolving technologies and their uses.

ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 59 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Poorly defined accountability for RAI makes it look good on paper but renders it

à
Ho
ineffective in reality.

_
MB
■ Organizations lack awareness of AI’s unintended consequences. Many turn to RAI

NH
only after they experience AI’s negative effects, whereas prevention is simpler.

E
UY
NG
■ Most AI regulations are still in draft. AI products’ adoption of regulations for privacy
and intellectual property makes it challenging for organizations to ensure

4
compliance and avoid all possible liability risks.

02
2/2
■ Rapidly evolving AI technologies, including tools for explainability, bias detection,

4/1
privacy protection and some regulatory compliance, lull organizations into a false

10
2:2
sense of responsibility, while mere technology is not enough. A disciplined AI ethics

_1
and governance approach is necessary, in addition to technology.

ên
uy
■ Measuring success is difficult. Creating RAI principles and operationalizing them
Ng
without regularly measuring the progress makes it hard to sustain RAI practices.
nh
Bì
nh

User Recommendations
Mạ

■ Publicize consistent approaches across all RAI focus areas. The most typical areas
ng
oà

of RAI in the enterprise are fairness, bias mitigation, ethics, risk management,
_H

security, privacy, reliability, sustainability and regulatory compliance.

MB
NH

■ Designate a champion for each use case who will be accountable for the responsible
YE

development and use of AI.

U
NG

■ Define the AI life cycle framework. Address RAI in all phases of this cycle. Address
hard trade-off questions.
24
/20

■ Provide RAI training to personnel. Include AI literacy and critical thinking as part of
/12

the training.
04
:21

■ Operationalize RAI principles. Ensure diversity of participants and enable them to

12

easily voice AI concerns.

ên_
uy

■ Participate in industry or societal AI groups. Learn best practices and contribute your
Ng

own because everybody will benefit from this exchange. Ensure that policies account
ình

for the needs of any internal or external stakeholders.

hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 60 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sample Vendors

ng
Adobe; Arthur; Fiddler AI; Google; H2O.ai; IBM; Microsoft; Responsible AI Institute; SolasAI;

à
Ho
TruEra

_
MB
NH
Gartner Recommended Reading

E
Expert Insight Video: What Is Responsible AI and Why You Should Care About It?

UY
NG
Top Trends in AI Public Policy and Regulations for 2024

4
02
Software Engineering Leaders Must Help Drive Responsible AI

2/2
4/1
10
Best Practices for the Responsible Use of Natural Language Technologies

2:2
_1
How to Ensure Your Vendors Are Accountable for Governance of Responsible AI

ên
uy
Ng
AI Engineering nh
Analysis By: Soyeb Barot, Anthony Mullen, Leinar Ramos, Joe Antelmi
Bì
nh
Mạ

Benefit Rating: High

ng
oà

Market Penetration: 1% to 5% of target audience

_H
MB

Maturity: Adolescent
NH
YE

Definition:
U
NG

AI engineering is the foundation for enterprise delivery of AI and generative AI (GenAI)

solutions at scale. The discipline unifies DataOps, MLOps and DevOps pipelines to create
24
/20

coherent enterprise development, delivery (hybrid, multicloud, edge), and operational

/12

(streaming, batch) AI-based systems.

04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 61 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Why This Is Important

ng
The demand for AI solutions has dramatically increased, driven by the unrelenting hype

à
Ho
surrounding GenAI. Few organizations have built the data, analytics and software

_
MB
foundations needed to move individual pilot projects to production at scale, much less

NH
operate portfolios of AI solutions at scale. There are significant engineering, process and

E
culture challenges to address. To meet the demands for scaling AI solutions, enterprises

UY
must establish consistent AI pipelines supporting the development, deployment, reuse,

NG
governance and maintenance of AI models (statistical, machine learning, generative, deep
learning, graph, linguistic and rule-based).

4
02
2/2
Business Impact

4/1
10
AI engineering enables organizations to establish and grow high-value portfolios of AI

2:2
solutions consistently and securely. Most AI developments are currently limited by

_1
operational and cultural bottlenecks. With AI engineering approaches — DataOps,

ên
ModelOps and DevOps — it is possible to deploy models into production in a structured,

uy
Ng
repeatable factory-model framework. nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 62 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ DataOps, ModelOps and DevOps provide best practices for moving artifacts through

à
Ho
the AI development life cycle. Standardization across data and model pipelines

_
MB
accelerates the delivery of AI solutions.

NH
■ The elimination of traditional siloed approaches to data management and AI

E
UY
engineering doubles the data engineering effort and reduces impedance mismatches

NG
across data ingestion, processing, model engineering and deployment, which
inevitably drift once the AI models are in production.

4
02
2/2
■ AI engineering enables discoverable, composable and reusable AI artifacts (data

4/1
catalogs, knowledge graphs, code repositories, reference architectures, feature

10
stores, model stores and others) across the enterprise context. These are essential

2:2
for scaling AI enterprisewide.

_1
ên
■ AI engineering makes it possible to orchestrate solutions across hybrid, multicloud,

uy
edge AI or Internet of Things.
Ng
nh
Broader use of foundational platforms provides initial success at scaling the
Bì

■
nh

production of AI initiatives with existing data, analytics and governance frameworks.

Mạ

AI engineering practices, processes and tools must be adapted to address GenAI.

ng

■
oà

GenAI specific adaptations include support for prompt engineering, vector

_H

DBs/graph KBs, architecting and deploying multiagent, and interactive deployment

MB

models.
NH
YE

■ AI engineering tools can be subdivided into model-centric and data-centric tools.

U

Terms such as DataOps, LLMOps, LangOps or FMOps, or more broader terms such
NG

as ModelOps or MLOps, are used frequently, but we believe they are all a subset of AI
24

engineering.
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 63 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Sponsorship for foundational enterprisewide AI initiatives is unclear. The

à
Ho
transformational promise of AI enablement has led executives to actively compete

_
MB
for enterprise AI responsibility.

NH
■ AI engineering needs simultaneous development of pipelines across domains and

E
UY
platform infrastructure maturity.

NG
■ AI engineering requires integrating full-featured solutions with specific tools,

4
including open-source technologies, to address enterprise architecture gaps with

02
2/2
minimal functional overlap. These include gaps around extraction, transformation

4/1
and loading stores, feature stores, model stores, model monitoring, pipeline

10
observability, and governance.

2:2
_1
■ AI engineering requires cloud maturity and possible rearchitecting, or the ability to

ên
integrate data and AI model pipelines across deployment contexts. Potential

uy
complexity and management of analytical and AI workloads alongside costs may
Ng
deter organizations that are in the initial phases of AI initiatives.
nh
Bì
nh

■ Enterprises often seek “unicorn” experts to productize AI platforms. Spot-fix vendor

Mạ

solutions will bloat costs and potentially complicate already intricate integration and
ng

model management tasks.

oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 64 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Establish a leadership mandate for enterprisewide foundational AI initiatives.

à
Ho
_
■ Maximize business value from ongoing AI initiatives by establishing AI engineering

MB
practices that streamline the data, model and implementation pipelines.

NH
E
UY
■ Simplify data and analytics pipelines by identifying the capabilities required to

NG
operationalize end-to-end AI platforms and build AI-specific toolchains.

■ Use point solutions sparingly and only to plug feature/capability gaps in fully

4
02
featured DataOps, MLOps, ModelOps and PlatformOps tools.

2/2
4/1
■ Develop AI model management and governance practices that align model

10
performance, human behavior and delivery of business value to make it easier for

2:2
users to adopt AI models.

_1
ên
Leverage cloud service provider environments as foundational to build AI

uy
■

Ng
engineering. At the same time, rationalize your data, analytics and AI portfolios as
nh
you migrate to the cloud.
Bì
nh

■ Adopt a platform approach to GenAI by investing in centralized AI engineering tools

Mạ

for automation, governance and use-case enablement across a broad set of AI

ng

models and providers.

oà
_H

Upskill data engineering and platform engineering teams to adopt tools and
MB

■
processes that drive continuous integration/continuous development for AI artifacts.
NH
UYE

Sample Vendors
NG

Amazon Web Services; Dataiku; DataRobot; Domino Data Lab; Google; Microsoft;
24

neptune.ai; OctoAI; Seldon Technologies; Weights & Biases

/20
/12

Gartner Recommended Reading

04
:21

Top Strategic Technology Trends for 2022: AI Engineering

12
n_

Demystifying XOps: DataOps, MLOps, ModelOps, AIOps and Platform Ops for AI
ê
uy
Ng

A CTO’s Guide to Top Artificial Intelligence Engineering Practices

ình
hB

Cool Vendors in AI Core Technologies — Scaling AI in the Enterprise

ạn
gM

Case Study: AI Model Operations at Scale (Fidelity)

àn
Ho
B_

Gartner, Inc. | G00809438 Page 65 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Edge AI

ng
Analysis By: Eric Goodness

à
Ho
_
MB
Benefit Rating: High

NH
E
Market Penetration: 5% to 20% of target audience

UY
NG
Maturity: Adolescent

4
02
Definition:

2/2
4/1
Edge AI refers to the use of AI techniques embedded in non-IT products

10
(consumer/commercial), IoT endpoints, gateways and edge servers. It spans use cases

2:2
for consumer, commercial and industrial applications, such as mobile devices,

_1
autonomous vehicles, enhanced capabilities of medical diagnostics and streaming video

ên
analytics. While predominantly focused on AI inference, more sophisticated systems may

uy
Ng
include a local training capability to provide optimization of the AI models at the edge.
nh
Bì

Why This Is Important

nh
Mạ

Many edge computing use cases are latency-sensitive and data-intensive and require
ng

autonomy for local decision making. This creates a need for AI-based applications in a
oà

wide range of edge computing and endpoint solutions. Examples include real-time
_H

analysis of edge data for predictive maintenance, inferences for decision support and
MB

video analytics. Increasingly, generative models (including smaller language models) have
NH

become an area of experimentation and investment.

UYE
NG

Business Impact

■ Real-time data analysis and decision intelligence.

24
/20

■ Improved operational efficiency, such as manufacturing visual inspection systems

/12

that identify defects, wasted motion, waiting and over- or underproduction.

04
:21

■ Enhanced customer experience through feedback from AI embedded within

12

products.
ên_
uy

Connectivity cost reduction with less data traffic between the edge and the cloud.
Ng

■
ình

■ Persistent functions and solution availability, irrespective of network connectivity.

hB
ạn

■ Reduced storage demand as only prioritized data is passed on to core systems.

gM
àn

■ Preserved data privacy at the endpoint.

Ho
B_

Gartner, Inc. | G00809438 Page 66 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
Overall, edge AI has benefited from improvements in the capabilities of AI. This includes:

à
Ho
_
MB
■ The maturation of MLOps and ModelOps tools and processes support ease of use

NH
across a broader set of features that span the broader MLOps functions. Initially,

E
many companies came to market with a narrowcast focus on model compression.

UY
NG
■ The improved performance of combined machine learning (ML) techniques and an
associated increase in data availability (such as time-series data from industrial

4
02
assets).

2/2
4/1
There is business demand for new and improved outcomes solely achievable from the

10
2:2
use of AI at the edge, which include:

_1
ên
■ Reducing full-time equivalents with vision-based solutions used for surveillance or

uy
inspections.
Ng
nh
Improving manufacturing production quality by automating various processes.
Bì

■
nh

Optimizing operational processes across industries.

Mạ

■
ng

■ New approaches to customer experience, such as personalization on mobile devices

oà
_H

or changes in retail from edge-based smart check-out points of sale.

MB

Growing interest in local deployments of generative AI.

NH

■
UYE

Additional drivers include:

NG
24

■ Increasing number of users upgrading legacy systems and infrastructure in

/20

“brownfield” environments. By using MLOps platforms, AI software can be hosted

/12

within an edge computer or a gateway (aggregation point) or embedded within a

04

product with the requisite compute resources.

:21
12

■ More manufacturers embedding AI in the endpoint as an element of product

n_

servitization. In this architecture, the Internet of Things (IoT) endpoints, such as in

ê
uy

automobiles, home appliances and commercial building infrastructure, are capable

Ng

of running AI models to interpret data captured by the endpoint and drive some of
ình

the endpoints’ functions.

hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 67 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Rising demand for R&D in training decentralized AI models at the edge for adaptive

ng
AI. These emerging solutions are driven by explicit needs such as privacy

à
Ho
preservation or the requirement for machines and processes to run in disconnected

_
(from the cloud) scenarios.

MB
NH
E
Obstacles

UY
NG
■ Edge AI is constrained by the application and design limitations of the equipment
deployed; this includes form factor, power budget, data volume, decision latency,

4
location and security requirements.

02
2/2
■ Systems deploying AI techniques can be nondeterministic. This will impact

4/1
applicability in certain use cases, especially where safety and security requirements

10
2:2
are important.

_1
ên
■ The autonomy of edge AI-enabled solutions, built on some ML and deep learning

uy
techniques, often presents questions of trust, especially where the inferences are not
Ng
readily interpretable or explainable. As adaptive AI solutions increase, these issues
nh
will increase if initially identical models deployed to equivalent endpoints
Bì
nh

subsequently begin to evolve diverging behaviors.

Mạ

The lack of quality and sufficient data for training is a universal challenge across AI
ng

■
oà

usage.
_H
MB

■ Deep learning in neural networks is a compute-intensive task, often requiring the use
NH

of high-performance chips with corresponding high-power budgets. This can limit

YE

deployment locations, especially where small-form factors and lower-power

U

requirements are paramount.

NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 68 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Determine whether the use of edge AI provides adequate cost-benefit improvements

à
Ho
or whether traditional centralized data analytics and AI methodologies are adequate

_
MB
and scalable.

NH
■ Evaluate when to consider AI at the edge versus a centralized solution. Good

E
UY
candidates for edge AI are applications that have high communications costs, are

NG
sensitive to latency, require real-time responses or ingest high volumes of data at the
edge.

4
02
2/2
■ Assess the different technologies available to support edge AI and the viability of the

4/1
vendors offering them. Many potential vendors are startups that may have

10
interesting products but limited support capabilities.

2:2
_1
■ Use edge gateways and servers as the aggregation and filtering points to perform

ên
most of the edge AI and analytics functions. Make an exception for compute-

uy
intensive endpoints, where AI-based analytics can be performed on the devices
themselves. Ng
nh
Bì
nh

Sample Vendors
Mạ
ng

Chooch; Edge Impulse; IFS (Falkonry); Litmus Automation; MicroAI

oà
_H

Gartner Recommended Reading

MB
NH

Emerging Tech Impact Radar: Edge Artificial Intelligence

UYE

Innovation Insight for Edge AI

NG

Emerging Tech: Differentiate With an Edge AI Benchmarking Strategy

24
/20
/12

Market Guide for Edge Computing

04
:21

Emerging Tech: Empower Outcome-Centric IoT With AI

12
n_

Foundation Models
ê
uy
Ng

Analysis By: Arun Chandrasekaran

ình
hB

Benefit Rating: Transformational

ạn
gM

Market Penetration: 20% to 50% of target audience

àn
Ho
B_

Gartner, Inc. | G00809438 Page 69 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Maturity: Early mainstream

ng
à
Definition:

Ho
_
MB
Foundation models are large-parameter models that are trained on a broad gamut of

NH
datasets in a self-supervised manner. They are mostly based on transformer or diffusion

E
deep neural network architectures and are increasingly becoming multimodal. They are

UY
called foundation models because of their critical importance and applicability to a wide

NG
variety of downstream use cases. This broad applicability is due to the pretraining and
versatility of the models.

4
02
2/2
Why This Is Important

4/1
10
Foundation models are an important step forward for AI due to their massive pretraining

2:2
and wide use-case applicability. They can deliver state-of-the-art capabilities with higher

_1
efficacy than their predecessors. They’ve become the go-to architecture for natural

ên
language processing, and have also been applied to computer vision, audio and video

uy
Ng
processing, software engineering, chemistry, finance and legal use cases.
nh
Bì

Business Impact
nh
Mạ

With their potential to enhance applications across a broad range of enterprise use cases,
foundation models will have a wide impact across vertical industries and business
ng
oà

functions. Their impact has accelerated, with a growing ecosystem of startups building
_H

enterprise applications on top of them. Foundation models will advance digital

MB

transformation within the enterprise by improving workforce productivity, automating and

NH

enhancing customer experience, and enabling rapid, cost-effective creation of new

YE

products and services.

U
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 70 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Quicker time to value — Foundation models can effectively deliver value through

à
Ho
prebuilt APIs, prompt engineering, retrieval-augmented generation or further fine-

_
MB
tuning. While fine-tuning may enable more customization, the other two options are

NH
less complex, quicker and cheaper.

E
UY
■ Superior performance across multiple domains — The difference between these

NG
models and prior neural network solutions is stark. The large pretrained models can
produce coherent text, code, images, speech and video at a scale and accuracy not

4
02
possible before.

2/2
4/1
■ Fast-paced innovation — The past year has seen an influx of foundation models,

10
along with smaller, pretrained domain-specific models built from them. Most of

2:2
these are available as cloud APIs or open-source projects, further reducing the time

_1
and cost to experiment and driving quicker enterprise adoption.

ên
uy
Productivity gains — Foundation models are having an impact across broad swaths
Ng
■
of enterprise business functions as their ability to automate tasks gets wider.
nh
Bì

Business functions such as marketing, customer service and IT (especially software

nh

engineering) are areas where clients are seeking initial gains.

Mạ
ng
oà

Obstacles
_H

■ Do not deliver perfect results — Although a significant advance, foundation models

MB

still require careful training and guardrails. Because of their training methods and
NH

black-box nature, they can deliver unacceptable results or hallucinations. They also
UYE

can propagate downstream any bias or copyright issues in the datasets.

NG

■ Require appropriate skills and talent — As with all AI solutions, the end result
24

depends on the skills, knowledge and talent of the trainers and users, particularly for
/20

prompt engineering and fine-tuning.

/12
04

■ Expansion to impractical sizes — Large models are up to billions or trillions of

:21

parameters. They are impractically large to train for most organizations because of
12

the necessary compute resources, which can make them expensive and ecologically
ên_

unfriendly.
uy
Ng

■ Concentrate power — These models have been mostly built by the largest
ình

technology companies with huge R&D investments and significant AI talent,

hB

resulting in a concentration of power among a few large, deep-pocketed entities.

ạn
gM

This situation may create a significant imbalance in the future.

àn
Ho
B_

Gartner, Inc. | G00809438 Page 71 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Create a strategy document that outlines the benefits, risks, opportunities and

à
Ho
execution plans for these models in a collaborative effort.

_
MB
■ Plan to introduce foundation models into existing speech, text or coding domains.

NH
If you have any older language processing systems, moving to a transformer-based

E
UY
model could significantly improve performance. Knowledge search, summarization,

NG
content generation are popular emerging use cases across industries.

4
■ Start with models that have superior ecosystem support and adequate enterprise

02
2/2
guardrails around security and privacy, and are more widely deployed.

4/1
Be objective about the adequate balance between accuracy, costs, security and

10
■

2:2
privacy, and time to value when selecting foundation models to determine the

_1
appropriate model needed. Beware of building models from scratch, given the

ên
complexity and steep costs.

uy
■ Ng
Educate developers, data and analytics teams on prompt engineering and other
nh
advanced techniques needed to steer these models.
Bì
nh

Designate an incubation team to monitor industry developments, communicate the

Mạ

■
art of the possible, experiment with business units and share valuable lessons
ng
oà

learned companywide.
_H
MB

Sample Vendors
NH
YE

Alibaba Group; Anthropic; Cohere; Databricks; Google; Hugging Face; IBM; Microsoft;
U

Mistral AI; OpenAI

NG

Gartner Recommended Reading

24
/20

Innovation Guide for Generative AI Models

/12
04

Quick Answer: What Are the Pros and Cons of Open-Source Generative AI Models?
:21
12

Synthetic Data
ên_
uy

Analysis By: Arun Chandrasekaran, Anthony Mullen, Alys Woodward

Ng
ình

Benefit Rating: High

hB
ạn

Market Penetration: 5% to 20% of target audience

gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 72 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Maturity: Adolescent

ng
à
Definition:

Ho
_
MB
Synthetic data is a class of data that is artificially generated rather than obtained from

NH
direct observations of the real world. Synthetic data is used as a proxy for real data in a

E
wide variety of use cases, including data anonymization, AI and machine learning (ML)

UY
development, data sharing, and data monetization.

NG
Why This Is Important

4
02
A major problem with AI development today is the burden involved in obtaining real-world

2/2
4/1
data and labeling it. This time-consuming and expensive task can be remedied with

10
synthetic data, where data can be generated faster and cheaper. Additionally, for specific

2:2
use cases such as training models for autonomous vehicles, collecting real data for 100%

_1
coverage of edge cases is practically impossible. Furthermore, synthetic data can be

ên
generated without personally identifiable information (PII) or protected health information

uy
Ng
(PHI), making it a valuable technology for privacy preservation. nh
Bì

Business Impact
nh
Mạ

Adoption is increasing across various industries. Gartner predicts a massive increase in

adoption as synthetic data:
ng
oà
_H

■ Avoids using PII when training ML models via synthetic variations of original data or
MB

synthetic replacement of parts of data.

NH
YE

■ Reduces cost and saves time in ML development.

U
NG

■ Improves ML performance as more training data leads to better outcomes.

24

■ Enables organizations to pursue new use cases for which very little real data is
/20

available.
/12
04

■ Is capable of addressing fairness issues more efficiently.

:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 73 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ In healthcare and finance, buyer interest is growing as synthetic tabular data can be

à
Ho
used to preserve privacy in AI training data.

_
MB
■ To meet the increasing demand for synthetic data for natural language automation

NH
training, especially for chatbots and speech applications, new and existing vendors

E
UY
are bringing new offerings to market. This is expanding the vendor landscape and

NG
driving synthetic data adoption.

4
■ Synthetic data applications have expanded beyond automotive and computer vision

02
2/2
use cases to include data monetization, external analytics support, platform

4/1
evaluation and the development of test data.

10
2:2
■ Transformer and diffusion architectures, the architectural foundations for generative

_1
AI (GenAI), are enabling synthetic data generation at quality and precision not seen

ên
before. AI simulation techniques are improving the quality of synthetic data by better

uy
recreating real-world representations.
Ng
nh
There is an expansion to other data types. While tabular, image, video, text and
Bì

■
nh

speech applications are common, R&D labs are expanding the concept of synthetic
Mạ

data to graphs. Synthetically generated graphs will resemble, but not overlap the
ng

original. As organizations begin to use graph technology more, we expect this

oà

method to mature and drive adoption.

_H
MB

■ The growing adoption of GenAI models and future customizations of such models
NH

will drive the demand for synthetic data to pretrain these models.
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 74 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Synthetic data can have bias problems, miss natural anomalies, be complicated to

à
Ho
develop or not contribute any new information to existing, real-world data.

_
MB
■ Data quality is tied to the model that generates the data.

NH
E
UY
■ There are no clear best practices on how to combine synthetic and real data for AI

NG
development.

■ Synthetic data generation methodologies lack standardization.

4
02
2/2
■ It is difficult to validate the accuracy of synthetic data. While a synthetic dataset

4/1
may look realistic and accurate, it is difficult to know for sure if it accurately

10
captures the underlying real-world environment.

2:2
_1
■ Buyers are still confused over when and how to use the technology due to the lack of

ên
skills.

uy
Ng
Synthetic data can still reveal a lot of sensitive details about an organization, so
nh
■
Bì

security and privacy are concerns. An ML model could be reverse-engineered via

nh

active learning. With active learning, a learning algorithm can interactively query a
Mạ

user (or other information sources) to label new data points with the desired outputs,
ng

meaning learning algorithms can actively query the user or teacher for labels.
oà
_H

If fringe or edge cases are not part of the seed dataset, they will not be synthetized.
MB

■
This means the handling of such borderline cases must be carefully accommodated.
NH
YE

■ There may be a level of user skepticism as data may be perceived to be “inferior” or

U
NG

“fake.”
24
/20
/12
04
:21
12
ê n_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 75 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Identify areas in your organization where data is missing, incomplete or expensive to

à
Ho
obtain, and is thus currently blocking AI initiatives. In regulated industries, such as

_
MB
healthcare or finance, exercise caution and adhere to rules.

NH
■ Use synthetic variations of the original data, or synthetic replacement of parts of

E
UY
data, when personal data is required but data privacy is a requirement.

NG
■ Educate internal stakeholders through training programs on the benefits and

4
limitations of synthetic data. Institute guardrails to mitigate challenges such as user

02
2/2
skepticism and inadequate data validation.

4/1
Measure and communicate the business value, success and failure stories of

10
■

2:2
synthetic data initiatives.

_1
ên
Sample Vendors

uy
Ng
Anonos (Statice); Gretel; Hazy; Howso; MOSTLY AI; Parallel Domain; Rendered.ai; Tonic.ai;
nh
YData
Bì
nh
Mạ

Gartner Recommended Reading

ng

Innovation Guide for Generative AI Models

oà
_H
MB

Case Study: Enable Business-Led Innovation with Synthetic Data (Fidelity International)
NH
YE

Predicts 2024: The Future of Generative AI Technologies

U
NG

ModelOps
24

Analysis By: Joe Antelmi, Soyeb Barot, Erick Brethenoux

/20
/12

Benefit Rating: High

04
:21

Market Penetration: 1% to 5% of target audience

12
ên_

Maturity: Emerging
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 76 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Definition:

ng
Model operationalization (ModelOps) is primarily focused on the end-to-end governance

à
Ho
and life cycle management of advanced analytics, AI and decision models, such as

_
MB
models based on machine learning (ML), generative AI (GenAI), knowledge graphs, rules,

NH
optimization, linguistics, agents and others.

E
UY
Why This Is Important

NG
ModelOps helps companies in standardizing, scaling and augmenting their analytics and

4
AI initiatives. It helps organizations to move their models from the lab environments into

02
production. MLOps primarily focuses on monitoring and governance of ML models, while

2/2
4/1
ModelOps assists with the operationalization and governance of all advanced analytics,

10
decision and AI models, including GenAI and retrieval-augmented generation (RAG)

2:2
systems.

_1
ên
Business Impact

uy
Ng
ModelOps, as a practice: nh
Bì

■ Provides the capability for the management and operationalization of diverse AI,
nh
Mạ

analytics and decision systems.

ng

Enables the complex subsystems required for AI, analytics and decision system
oà

■
_H

observability including versioning, monitoring, automation, data orchestration,

MB

experimentation, and explainability.

NH
YE

■ Ensures collaboration among a wider business, development and deployment

U

community, and the ability to associate AI, analytics and model outcomes with
NG

business KPIs.
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 77 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Modern AI systems are being built with a symbiotic combination of generative and

à
Ho
classic AI models, agents and intelligent software capabilities. As the number of

_
MB
advanced analytics, AI and decision models at organizations increases,

NH
organizations will have to manage different types of prepackaged or custom-made

E
models in production.

UY
NG
■ Organizations want to be more agile and responsive to changes within their
advanced analytics and AI pipelines not only with models but also with data,

4
02
application and infrastructure.

2/2
4/1
■ ModelOps provides a framework to separate responsibilities across various teams

10
for how models (including GenAI, foundational models, analytics, ML, physical,

2:2
simulation, symbolic, etc.) are built, tested, deployed and monitored across different

_1
environments (for example, development, test and production). This enables better

ên
productivity and collaboration, and it lowers failure rates.

uy
Ng
ModelOps provides tools to address model degradation via drift, and bias. In other
nh
■
Bì

scenarios, enabling model governance, explainability and integrity is paramount.

nh
Mạ

■ The operationalization challenges of ML models are not new, but the capability to
ng

enable diverse models in production at the organization level using ModelOps is still
oà

evolving.
_H
MB

■ Organizations don’t want to deploy an unlimited number of open-source offerings to

NH

manage ModelOps, but there are few comprehensive solutions that provide end-to-
YE

end capabilities in every domain of model operationalization. Moreover, not every

U
NG

capability is required immediately. Often, versioning, monitoring and model

orchestration precede the full implementation of feature stores, pipelines and
24

observability.
/20
/12

■ GenAI will require an increased focus on testing, and the introduction of capabilities
04

to version, manage and automate prompts, routers, and retrieval-augmented

:21

generation systems. Fine-tuning will also require enhanced ModelOps capabilities to

12

manage complex domain and function training datasets.

ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 78 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Organizations using different types of models often don’t build the right ops,

à
Ho
governance and management capabilities until they already have a chaotic

_
MB
landscape of unmanaged advanced analytic systems.

NH
■ Not all analytical techniques currently benefit from mature operationalization

E
UY
methods. Because the spotlight has been on ML techniques, MLOps benefits from a

NG
more evolved AI practice, but some models, like agentic modeling and optimization
techniques, require more attention in ModelOps practices and platforms.

4
02
2/2
■ ModelOps capabilities that help productionize GenAI are emerging but immature.

4/1
Moreover, organizations are struggling to get GenAI into production, due to data,

10
security and regulatory concerns.

2:2
_1
■ Organizations may adopt ModelOps platform capabilities that they don’t

ên
immediately need. At the same time, organizations that are siloed and fail to adopt a

uy
comprehensive ModelOps strategy create redundancy in effort with respect to
operationalization. Ng
nh
Bì
nh

User Recommendations
Mạ
ng

■ Buy ModelOps capabilities integrated into your primary AI platforms. Enrich these
oà

capabilities with best-of-breed open-source or proprietary ModelOps offerings where

_H

unique problems, like feature stores or observability, require enhanced solutions.

MB
NH

■ Utilize ModelOps best practices across composite AI, data, models and applications
YE

to ensure transition, reduce friction and increase value generation.

U
NG

■ Recruit/upskill additional engineers who can master ModelOps on AI systems that

utilize unstructured data, search, graph and optimization.
24
/20

Encourage collaboration between development and deployment teams, and

/12

■
04

empower teams to make decisions to automate, scale and bring stability to the
:21

analytics and AI pipeline.

12
n_

■ Collaborate with software engineering teams to scale ModelOps. Offloading

ê
uy

operationalization responsibilities to production support teams enables increased

Ng

ModelOps specialization and sophistication across the ecosystem of complex AI-

ình

enabled applications.
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 79 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sample Vendors

ng
DataRobot; IBM; ModelOp; Modzy; Neptune.ai; OctoAI; SAS; Valohai; Verta; Weights &

à
Ho
Biases

_
MB
NH
Gartner Recommended Reading

E
Launch an Effective Machine Learning Monitoring System

UY
NG
Innovation Guide for Generative AI in Trust, Risk and Security Management

4
02
The Logical Feature Store: Data Management for Machine Learning

2/2
4/1
10
Operationalize Machine Learning by Using Gartner’s MLOps Framework

2:2
_1
Toolkit: Delivery Metrics for DataOps, Self-Service Analytics, ModelOps and MLOps

ên
uy
Ng
Generative AI nh
Analysis By: Svetlana Sicular
Bì
nh
Mạ

Benefit Rating: Transformational

ng
oà

Market Penetration: 20% to 50% of target audience

_H
MB

Maturity: Adolescent
NH
YE

Definition:
U
NG

Generative AI (GenAI) technologies can generate new derived versions of content,

strategies, designs and methods by learning from large repositories of original source
24
/20

content. Generative AI has profound business impacts, including on content discovery,

/12

creation, authenticity and regulations; automation of human work; and customer and
04

employee experiences.
:21
12

Why This Is Important

ên_

GenAI exploration is widening:

uy
Ng
ình

■ End-user organizations aggressively experiment with GenAI. Early adopters in most

hB

industries have had initial success with GenAI.

ạn
gM

■ Major technology vendors prioritize delivery of GenAI-enabled applications and tools.

àn
Ho
B_

Gartner, Inc. | G00809438 Page 80 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Numerous solutions have been emerging to innovate with foundation models,

ng
hardware and data for GenAI.

à
Ho
■ Impacted by the GenAI hype, governments are introducing AI regulations and

_
MB
investing in national AI strategies.

NH
E
UY
Business Impact

NG
Business focus is shifting from excitement around foundation models to use cases that
drive ROI. Most GenAI implementations are currently low-risk and internal. With the rapid

4
02
progress of productivity tools and AI governance practices, organizations will be

2/2
deploying GenAI for more critical use cases in industry verticals and scientific discovery. In

4/1
the longer term, GenAI-enabled conversational interfaces will facilitate technology

10
2:2
commercialization, democratizing AI and other technologies.

_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 81 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ Industry applications of GenAI are growing. GenAI reached creative work in

à
Ho
entertainment, marketing, design, music, architecture and content generation.

_
MB
■ Best implementation practices are being discovered by the first enterprisewide

NH
deployments, and are fueling the top GenAI enterprise use cases: advanced chatbots,

E
UY
coding assistance and internal service desk. According to the 2023 Gartner AI in the

NG
Enterprise Survey, 18% of leaders highly involved in AI report that their organizations
are advanced in GenAI adoption.

4
02
2/2
■ GenAI is a top competitive area among major technology vendors. They compete on

4/1
foundation model offerings, their enterprise readiness, pricing, infrastructure, safety

10
and indemnification.

2:2
_1
■ New foundation models in new versions, sizes and capabilities are rapidly coming to

ên
market, making GenAI available for more use cases. Tools to improve model

uy
robustness, such as vector databases, graph technologies, LLM testing, security
Ng
protection and open-source resources are making GenAI more usable.
nh
Bì
nh

■ The progress is significant in multimodal models like Gemini or GPT4-Video, which

Mạ

are trained to take in both images and text; for example, they allow users to ask
ng

questions about images and receive answers via text. Models can combine
oà

concepts, attributes and styles to create original images, video and art, or translate
_H

audio to different voices and languages. Notably, text-to-image/video generation has

MB

advanced with the ability to create highly detailed and realistic visuals from textual
NH
YE

descriptions.
U
NG

■ Enterprises are learning to use their own data with GenAI via prompt engineering and
fine-tuning. AI-ready data and associated metadata have become central to GenAI
24

strategies.
/20
/12

■ Synthetic data helps enterprises to augment scarce data, mitigate bias, achieve
04

superresolution or preserve data privacy.

:21
12

■ GenAI disrupts software engineering. Development automation techniques are

n_

promising to automate 5% to 10% of the programmers’ work. Organizations are now

ê
uy

willing to tackle legacy modernization with GenAI.

Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 82 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ GenAI causes new ethical and societal concerns. Government regulations may

à
Ho
hinder GenAI research. Pending regulations proliferate.

_
MB
■ Hallucinations, bias, a black-box nature and inexperience with a full AI life cycle

NH
preclude the use of GenAI in critical use cases for now.

E
UY
NG
■ GenAI accountability, licensing and pricing are inconsistent among providers, and
may catch customers off-guard.

4
02
■ Reproducing results and finding references for generated information is challenging,

2/2
but some validation solutions are emerging.

4/1
10
■ Security professionals are new to certifying and protecting GenAI solutions; it will

2:2
take time for security best practices to crystallize.

_1
ên
GenAI is used for nefarious purposes. Full and accurate detection of generated

uy
■

Ng
content, such as deepfakes and disinformation, will remain challenging or
nh
impossible.
Bì
nh

■ The compute resources for training GenAI models are not affordable to most
Mạ

enterprises. Sustainability concerns about high energy consumption by GenAI are

ng

rising.
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ê n_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 83 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Identify low-risk use cases where you can improve your business with GenAI by

à
Ho
relying on purchased capabilities. Consult vendor roadmaps to avoid developing

_
MB
similar solutions in-house.

NH
■ Architect your GenAI solutions to be ready for near-future upgrades, as foundation

E
UY
models and data tooling for them are progressing swiftly.

NG
■ Pilot ML-powered coding assistants, with an eye toward fast rollouts, to boost

4
developer productivity.

02
2/2
■ Use synthetic data to accelerate the development cycle and lessen regulatory

4/1
concerns.

10
2:2
■ Quantify the advantages and limitations of GenAI. Issue GenAI policies and

_1
ên
guidelines, as it requires skills, funds and caution.

uy
Ng
■ Mitigate GenAI risks by working with legal, procurement, security and fraud experts.
nh
Technical, institutional and political interventions will be necessary to fight AI’s
Bì

adversarial impacts.
nh
Mạ

■ Optimize the cost and efficiency of AI solutions by employing composite AI

ng

approaches to combine GenAI with other AI techniques.

oà
_H
MB

Sample Vendors
NH

Alibaba Cloud; Amazon Web Services; Anthropic; Google; Hugging Face; IBM; Meta;
UYE

Microsoft; Mistral AI; OpenAI

NG

Gartner Recommended Reading

24
/20

Innovation Guide for Generative AI Technologies

/12
04

Innovation Guide for Generative AI Models

:21
12

How to Calculate Business Value and Cost for Generative AI Use Cases
ên_
uy

When Not to Use Generative AI

Ng
ình
hB

10 Best Practices for Scaling Generative AI Across the Enterprise

ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 84 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Sliding into the Trough

ng
Neuromorphic Computing

à
Ho
Analysis By: Alan Priestley

_
MB
NH
Benefit Rating: Transformational

E
UY
NG
Market Penetration: Less than 1% of target audience

4
Maturity: Embryonic

02
2/2
4/1
Definition:

10
Neuromorphic computing is a technology that provides a mechanism to more accurately

2:2
model the operation of a biological brain using digital or analog processing techniques.

_1
ên
These designs typically use spiking neural networks (SNNs) rather than the deep neural

uy
networks (DNNs), feature non-von Neumann architectures and are characterized by simple
processing elements, but very high interconnectivity. Ng
nh
Bì

Why This Is Important

nh
Mạ

Currently, most AI development uses parallel processing designs based on graphics

ng

processing units (GPUs). These are high-performance, but high-power-consuming, devices

oà

that are not applicable in many deployments. Neuromorphic computing utilizes

_H
MB

asynchronous, event-based designs that have the potential to offer extremely low power
NH

operation. This makes them uniquely suitable for edge and endpoint devices, where their
YE

ability to support object and pattern recognition can enable image, audio, and sensor
U

analytics.
NG

Business Impact
24
/20

■ Neuromorphic computing architectures have the potential to deliver extreme

/12

performance for use-cases pattern recognition and signal analysis at very low power
04

and can be trained using smaller datasets than other AI models, with the potential of
:21

in situ training.
12
n_

Neuromorphic computing designs can be implemented using low-power devices,

ê

■
uy

which brings the potential to drive the reach of AI techniques out to the edge of the
Ng

network, thereby accelerating key tasks such as image and sound recognition.
ình
hB
ạn

Drivers
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 85 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Today’s DNN algorithms and large language models (LLMs) require the use of high-

ng
performance processing devices and vast amounts of data to train these systems,

à
Ho
limiting scope of deployment.

_
MB
■ Different design approaches are being taken to implement neuromorphic computing

NH
designs — large-scale devices for use in data centers, and smaller-scale devices for

E
UY
edge computing and endpoint designs. Both these paths utilize SNNs to implement

NG
asynchronous designs that have the benefit of being extremely low power when
compared with current DNN-based designs.

4
02
Semiconductor vendors are developing chips that utilize SNNs to implement AI-

2/2
■
based solutions.

4/1
10
2:2
Obstacles

_1
ên
■ Accessibility: GPUs are more accessible and easier to program than neuromorphic

uy
computing. However, this could change when neuromorphic computing and the
supporting ecosystems mature. Ng
nh
Bì

■ Knowledge gaps: Programming neuromorphic computing will require new

nh

programming models, tools and training methodologies.

Mạ
ng

■ Scalability: The complexity of interconnection challenges the ability of

oà
_H

semiconductor manufacturers to create viable neuromorphic devices.

MB

Integration: Significant advances in architecture and implementation are required to

NH

■
YE

compete with other AI architectures. Rapid developments in DNN and LLM

U

architectures may slow advances in neuromorphic computing, but there are likely to
NG

be major leaps forward in the next decade.

24
/20

User Recommendations
/12

■ Prepare for future utilization as neuromorphic architectures have the potential to

04

become viable over the next five years.

:21
12

■ Create a roadmap plan by identifying key applications that could benefit from
ên_

neuromorphic computing.
uy
Ng

■ Partner with key industry leaders in neuromorphic computing to develop proof-of-

ình

concept projects.
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 86 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Identify new skill sets that need to be nurtured for successful development of

ng
neuromorphic initiatives, and establish a set of business outcomes or expected

à
Ho
value to set management’s long-term expectations.

_
MB
NH
Sample Vendors

E
UY
AnotherBrain; Applied Brain Research; BrainChip; GrAI Matter Labs; Intel; Natural

NG
Intelligence; SynSense

4
Gartner Recommended Reading

02
2/2
Emerging Technologies: Tech Innovators in Neuromorphic Computing

4/1
10
Emerging Technologies: Top Use Cases for Neuromorphic Computing

2:2
_1
Forecast: AI Semiconductors, Worldwide, 2022-2028, 1Q24 Update

ên
uy
Emerging Tech Impact Radar: Artificial Intelligence Ng
nh
Bì

Smart Robots
nh
Mạ

Analysis By: Dwight Klappich

ng
oà

Benefit Rating: High

_H
MB
NH

Market Penetration: 1% to 5% of target audience

UYE

Maturity: Emerging
NG

Definition:
24
/20

A smart robot is an AI-powered, often-mobile machine designed to autonomously execute

/12

one or more physical tasks. These tasks may rely on, or generate, machine learning, which
04

can be incorporated into future activities or support unprecedented conditions. Smart

:21

robots can be classified into different types based on the tasks or use cases, such as
12

personal, logistics and industrial.

ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 87 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Why This Is Important

ng
Smart robotics is an AI use case, while robotics in general does not imply AI. Smart

à
Ho
(physical) robots had less adoption compared with industrial robot counterparts but

_
MB
received great hype in the marketplace. The placement of smart robots has moved

NH
forward several positions this year. This is due to an increased interest and investment in

E
smart robots in the last 12 months, as companies look to further improve logistic

UY
operations, support automation and augment humans in various jobs.

NG
Business Impact

4
02
Smart robots will make their initial business impact across a wide spectrum of asset-,

2/2
4/1
product- and service-centric industries. Their ability to reduce physical risk to humans

10
while also doing work with greater reliability, lower costs and higher productivity is

2:2
common across these industries. Smart robots are already being deployed among

_1
humans to work in logistics, warehousing and safety applications.

ên
uy
Ng
Drivers nh
■ The market is becoming more dynamic with technical developments over the last
Bì

two years, enabling a host of new use cases that have changed how smart robots
nh
Mạ

are perceived and how they can deliver value.

ng

Robots have become more affordable as the cost of components such as

oà

■
_H

processors, cameras and sensors has decreased over time.

MB
NH

■ The physical building blocks of smart robots (motors, actuators, chassis and
YE

wheels) have incrementally improved over time. Similarly, areas such as Internet of
U

Things (IoT) integration, edge AI and conversational capabilities have seen

NG

fundamental breakthroughs. These change the paradigm for robot deployments.

24

■ Vendor specialization has increased, leading to solutions that have higher business
/20

value, since an all-purpose/multipurpose device is either not possible or is less

/12

valuable.
04
:21

■ Interest in smart robots has increased across various industries. Smart robots are
12
n_

used in performing diverse tasks across medical/healthcare, manufacturing, last-

ê

mile delivery, inspection of industrial objects or equipment, agriculture, workplace

uy
Ng

and so forth.
ình
hB

■ Smart robots remain an emerging technology but the hype and expectations will
ạn

continue to build over the next few years, as providers expand their offerings and
gM

explore new technologies. Adding capabilities like reinforcement learning will help
àn

drive a continuous learning loop for robots and swarm management.

Ho
B_

Gartner, Inc. | G00809438 Page 88 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Companies are still struggling to identify valuable business use cases and assess

à
Ho
ROI for robots, especially outside of manufacturing and transportation.

_
MB
■ Complexity and versatility of tasks require complex decision making. Current smart

NH
robots excel at repetitive and predictable tasks and they are adaptable to various

E
UY
tasks.

NG
■ The continuous evolution and lack of commonality of pricing models and buying

4
options create uncertainty for organizations. Companies struggle to compare and

02
2/2
normalize all the various buying options they encounter such as monthly leasing,

4/1
hourly charges, robot as a service or buying the robot outright.

10
2:2
User Recommendations

_1
ên
■ Evaluate smart robots as both substitutes and complements to the human

uy
workforce in manufacturing, distribution, logistics, retail, healthcare or defense.
Ng
nh
■ Begin pilots designed to assess product capability and quantify benefits, especially
Bì

as ROI is possible even with small-scale deployments.

nh
Mạ

■ Prepare yourself to adopt and evolve your processes and robotics strategy as you
ng
oà

gain more experience in this field.

_H
MB

■ Examine current business processes and redesign these as necessary to support the
NH

deployment of smart robots.

UYE

■ Consider different purchase models for smart robots such as robot as a service or
NG

hybrid capital expenditure/operating expenditure models.

24

■ Dissolve the reluctance from staff by developing training resources to introduce

/20

robots alongside humans as an assistant.

/12
04

■ Ensure sufficient cloud computing resources to support high-speed and low-latency

:21

connectivity in the next two years.

12
n_

■ Evaluate multiple global and regional providers due to fragmentation within the
ê
uy

robot landscape.
Ng
ình
hB

Sample Vendors
ạn

Ava Robotics; Geekplus; GreyOrange; HAHN Group (Rethink Robotics); iRobot; Locus
gM

Robotics; SoftBank Robotics; Symbotic; temi; UBTECH Robotics

àn
Ho
B_

Gartner, Inc. | G00809438 Page 89 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Emerging Technologies: Top Use Cases for Smart Robots to Lead the Way in Human

à
Ho
Augmentation

_
MB
NH
Emerging Technologies: Top Use Cases Where Robots Interact Directly With Humans

E
UY
Emerging Technologies: Smart Robot Adoption Generates Diverse Business Value

NG
Cloud AI Services

4
02
2/2
Analysis By: Van Baker, Bern Elliot

4/1
10
Benefit Rating: High

2:2
_1
Market Penetration: 5% to 20% of target audience

ên
uy
Ng
Maturity: Early mainstream nh
Bì

Definition:
nh
Mạ

Cloud AI services provide AI model-building tools, APIs for prebuilt services and
ng

associated middleware that enable the building/training, deployment and consumption of

oà

machine learning (ML) and generative AI models running on prebuilt infrastructure as

_H

cloud services. These services include pretrained vision, language and other generative AI
MB

services, and automated ML and fine-tuning to create new models and customize prebuilt
NH

models.
UYE
NG

Why This Is Important

The use of cloud AI services continues to increase. Vendors have introduced additional
24
/20

services including large language model (LLM) APIs and solutions with fully integrated
/12

MLOps pipelines. The addition of low-code tools has added to ease of use. Applications
04

regularly use AI cloud services in language, vision, and tabular data and code generation
:21

to automate and accelerate business processes. Developers are aware of these offerings,
12

and are increasingly using both prebuilt and customized ML models in applications.
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 90 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Business Impact

ng
Cloud AI services impact applications running the business allowing developers to

à
Ho
enhance application functions. Generative AI adds a new category to these solutions

_
MB
allowing for the fine-tuning of LLMs to tailor performance. Data-driven decisioning

NH
mandates inclusion of ML models to add application functionality. Some AI technologies

E
are maturing, but generative AI includes less mature capabilities. Cloud AI services

UY
enhance applications with models that score, forecast and generate content enabling

NG
data-driven business operations.

4
02
Drivers

2/2
4/1
■ Opportunities to capitalize on new insights: The wealth of data from both internal

10
and third-party sources delivers insights such as the incorporation of predictive ML

2:2
models that enable data-driven decision intelligence in applications.

_1
ên
■ Demand for conversational interactions: The emergence of generative AI and large

uy
language models facilitates conversationally enabled applications where users can
use LLMs with data sources to gain insights. Ng
nh
Bì

The need to meet business key performance indicators (KPIs): There is a mandate
nh

■
Mạ

for businesses to automate processes to improve accuracy, improve responsiveness

ng

and reduce costs by deploying both AI and ML models.

oà
_H

■ Reduced barriers to entry: The ability to use pretrained generative AI models and
MB

fine-tune them has reduced the need for large quantities of data to train models.
NH

Access for developers and citizen data scientists to AI and ML services due to the
YE

availability of API-callable LLMs will further expand the use of AI by development

U
NG

teams.

Automated ML as an enabler for custom development: Use of automated ML to

24

■
/20

customize packaged services to address specific needs of the business is much

/12

more accessible and doesn’t require data scientists.

04
:21

■ A wide range of cloud AI services: Cloud AI services from a range of specialized

12

providers in the market, including orchestration layers to streamline deployment of

n_

solutions, are available.

ê
uy
Ng

■ Emerging AI model marketplaces: New marketplaces should help developers adopt

ình

these techniques through cloud AI services.

hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 91 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Lack of understanding by developers and citizen data scientists about how to adapt

à
Ho
these services to specific use cases.

_
MB
■ Grounding generative AI models is a challenge, requiring well-crafted retrieval-

NH
augmented generation (RAG) solutions that often include vector embeddings and

E
UY
other capabilities to implement. Many CAIDS providers are offering these

NG
capabilities as part of their generative AI offering.

4
■ Pricing models for cloud AI services that are usage-based presents a risk for

02
2/2
businesses as the costs associated with use of these services can accrue rapidly.

4/1
There is a need for comprehensive cost modeling tools to address this issue.

10
2:2
■ Increased need for packaged solutions that utilize multiple services for developers

_1
and citizen data scientists.

ên
uy
■ Limited availability of ModelOps tools that enable integration of AI and ML models
into applications. Ng
nh
Bì

■ Lack of skills such as prompt engineering and fine-tuning for developers to

nh

effectively implement these services in a responsible manner.

Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 92 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Choose customizable cloud AI services over bespoke models to address a range of

à
Ho
use cases and for quicker deployment and scalability.

_
MB
■ Improve chances of success of your AI strategy by experimenting with AI techniques,

NH
including the use of generative AI models such as LLMs and multimodal models and

E
UY
other cloud services. Ensure that generative AI models are loosely coupled as the

NG
technology continues to evolve rapidly.

4
■ Use cloud AI services to build less complex models, giving the benefit of more

02
2/2
productive AI while freeing up your data science assets for higher-priority projects.

4/1
Empower non-data-scientists with features such as automated algorithm selection,

10
■

2:2
dataset preparation and feature engineering for project elements. Leverage existing

_1
expertise on operating cloud services to assist technical professional teams.

ên
uy
■ Utilize pretrained generative AI models to allow for rapid prototyping and deployment
of LLM-enabled solutions. Ng
nh
Bì

■ Develop cost modeling tools that allow the enterprise to effectively predict both
nh

usage and management costs as AI models are broadly deployed in applications

Mạ

across the business.

ng
oà
_H

Sample Vendors
MB
NH

Alibaba Cloud; Amazon Web Services; Baidu; Google; H2O.ai; IBM; Microsoft; NVIDIA;
YE

Oracle; Tencent
U
NG

Gartner Recommended Reading

24

Critical Capabilities for Cloud AI Developer Services

/20
/12

Magic Quadrant for Cloud AI Developer Services

04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 93 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Climbing the Slope

ng
Autonomous Vehicles

à
Ho
Analysis By: Jonathan Davenport

_
MB
NH
Benefit Rating: Transformational

E
UY
NG
Market Penetration: Less than 1% of target audience

4
Maturity: Emerging

02
2/2
4/1
Definition:

10
Autonomous vehicles use various onboard sensing and localization technologies, such as

2:2
lidar, radar, cameras, global navigation satellite system (GNSS) and map data, in

_1
ên
combination with AI-based decision making, to drive without human supervision or

uy
intervention. Autonomous vehicle technology is being applied to passenger vehicles,
Ng
buses and trucks, as well as for specific use cases such as mining and agricultural
nh
tractors.
Bì
nh
Mạ

Why This Is Important

ng

Autonomous vehicles have the potential to change transportation economics, cutting

oà

operational costs and increasing vehicle utilization. In urban areas, inexpensive fares and
_H
MB

high-quality service may reduce the need for private car ownership. Road safety will
NH

increase, as AI systems will never be distracted, drive drunk or speed. Autonomous

YE

features on privately owned vehicles will enable productivity and recreational activities to
U

be undertaken, while the vehicle handles the driving operations.

NG

Business Impact
24
/20

Autonomous vehicles open the potential to disrupt traditional automotive business

/12

models by providing a software-based driver that is sold as part of a service that will
04

generate high margin revenue. Self-driving systems will stimulate demand for onboard
:21

computation to run AI software, radically increasing the vehicle’s overall bill of materials.
12
n_

After the office and home, vehicles will become a space where digital content can be
ê

created and consumed. Over time, fleet operators will likely retrain and redeploy their
uy
Ng

human commercial drivers to other, higher-value-adding roles within the company.

ình
hB

Drivers
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 94 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ The formalization of regulations and standards for autonomous vehicles will aid

ng
implementation. Automated lane-keeping system (ALKS) technology has been

à
Ho
approved by the United Nations Economic Commission for Europe (UNECE). This is

_
the first binding international regulation for SAE Level 3 vehicle automation, with a

MB
maximum operational speed of 37 mph. With the new regulatory landscape,

NH
automakers worldwide are beginning to announce Level 3 solutions.

E
UY
NG
■ Mercedes-Benz was the first automotive manufacturer to secure internationally valid
system approval and has launched in Germany. In the U.S., its Level 3 solution has

4
secured approval in Nevada and California. BMW has announced its Personal Pilot

02
2/2
L3 function that controls the car’s speed, distance to the vehicle ahead and lane

4/1
positioning for them autonomously, which is now available on new 7 Series vehicles.

10
In China, Changan, Great Wall Motor and Xpeng have announced Level 3 systems.

2:2
_1
■ The autonomous vehicle market is expected to evolve gradually from ADAS systems

ên
to higher levels of autonomy on passenger vehicles, rather than seeing a robotaxi-

uy
Ng
based revolution. This will require flexible vehicle operational design domains
nh
(ODDs).
Bì
nh

■ Self-driving trucks are a compelling business case. Driver pay is one of the largest
Mạ

operating costs for fleets associated with a commercial truck, plus goods can be
ng

transported much faster to their destination because breaks are no longer necessary.
oà

The Aurora Driver product is now at a “feature complete” stage, with a plan to launch
_H
MB

a “middle-mile” driverless truck service at the end of 2024.

NH

■ With the ability of GenAI to generate synthetic data, the training of AI algorithms in
YE

simulation can be accelerated.

U
NG

■ For off-road use cases, autonomous vehicles can assist, replace or redeploy human
24

workers to improve the accuracy with which work is done, lower operational costs
/20

and improve worker safety.

/12
04
:21

Obstacles
12

■ Due to the complexity of designing an autonomous vehicle, the cost to bring a

n_

commercial model to market has been greater than companies had envisioned,
ê
uy

requiring significant investments.

Ng
ình

■ When autonomous vehicles are commercially deployed, the vehicle developers, not
hB

the human occupants, will be liable for the vehicles’ autonomous operations.
ạn

Specific insurance solutions are needed to cover the vehicle should it be involved in
gM

an accident.
àn
Ho
B_

Gartner, Inc. | G00809438 Page 95 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Challenges increasingly include regulatory, legal and societal considerations, such

ng
as permits for operation and the effects of human interactions.

à
Ho
■ Automaker plans are being set back. For example, Hyundai’s Genesis G90 and the

_
MB
Kia EV9 vehicles were expected to be equipped with a Level 3 Highway Driving Pilot

NH
(HDP) function. The delay was caused by the variety of real-world driving scenarios

E
UY
that the system would need to support.

NG
■ Despite continued improvements in Level 4 perception algorithms and broader self-
driving systems used for mobility use cases (such as robotaxis), driverless

4
02
operations have not scaled to different cities quickly. Cruise’s accident in 2023

2/2
resulted in a strategy change that saw them lay off nearly 25% of its workforce.

4/1
10
2:2
User Recommendations

_1
ên
■ Governments must:

uy
Ng
■ Craft national legislation to ensure that autonomous vehicles can safely
nh
coexist with a traditional vehicle fleet as well as a framework for their approval
Bì

and registration.
nh
Mạ

■ Work closely with autonomous vehicle developers to ensure that first

ng

responders can safely respond to road traffic and other emergencies and self-
oà
_H

driving vehicles don’t obstruct or hinder activities.

MB
NH

■ Autonomous mobility operators should support consumer confidence in

YE

autonomous vehicle technology by remaining focused on safety and an accident-

U
NG

free road environment.

24

■ Traditional fleet operators looking to adopt autonomous technology into their fleets
/20

should minimize the disruptive impact on driving jobs (bus, taxi and truck drivers) by
/12

developing policies and programs to train and migrate these employees to other
04

roles.
:21
12

■ Automotive manufacturers should instigate a plan for how higher levels of

n_

autonomy can be deployed to vehicles being designed and manufactured to future-

ê
uy

proof vehicle purchases and enable future functions-as-a-service revenue streams.

Ng
ình
hB

Sample Vendors
ạn

Aurora; AutoX; Baidu; Cruise; Mobileye; NVIDIA; Oxa; Pony.ai; Waymo; Zoox
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 96 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Gartner Recommended Reading

ng
Emerging Tech Impact Radar: Autonomous Vehicles

à
Ho
_
MB
Lessons From Mining: 4 Autonomous Thing Benefit Zones for Manufacturers

NH
E
Emerging Tech: Synthetic Data Will Drive Future Autonomous Vehicles

UY
NG
Emerging Tech: Top Semiconductor Technology Trends in Autonomous Vehicles, 2023

4
02
Market Trend: Connected and Autonomous Vehicle Data Enhances Software Life Cycle

2/2
Management Transformation

4/1
10
Knowledge Graphs

2:2
_1
Analysis By: Afraz Jaffri

ên
uy
Ng
Benefit Rating: High nh
Bì

Market Penetration: 5% to 20% of target audience

nh
Mạ

Maturity: Early mainstream

ng
oà
_H

Definition:
MB

Knowledge graphs are machine-readable representations of the physical and digital

NH

worlds. They include entities (people, companies and digital assets) and their
YE

relationships, which adhere to a graph data model — a network of nodes (vertices) and
U
NG

links (edges/arcs).
24

Why This Is Important

/20
/12

Knowledge graphs capture information about the world in a visually intuitive format yet
04

are still able to represent complex relationships. Knowledge graphs act as the backbone
:21

of a number of products, including search, smart assistants and recommendation

12

engines. Knowledge graphs support collaboration and sharing, exploration and discovery,
n_

and the extraction of insights through analysis. Generative AI models can be combined
ê
uy

with knowledge graphs to provide context for more accurate outputs in a technique
Ng

becoming known as GraphRAG or G-RAG.

ình
hB

Business Impact
ạn
gM

Knowledge graphs can drive business impact in a variety of different settings, including:
àn
Ho
B_

Gartner, Inc. | G00809438 Page 97 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Digital workplace (such as collaboration, sharing and search)

ng
à
■ Automation (such as ingestion of data from content to robotic process automation)

Ho
_
MB
■ Machine learning (such as augmenting training data)

NH
Investigative analysis (such as law enforcement, cybersecurity and risk

E
■

UY
management)

NG
■ Digital commerce (such as product information management and

4
recommendations)

02
2/2
Data management (such as metadata management, data cataloging and data

4/1
■

10
fabric)

2:2
_1
Drivers

ên
uy
■ The need to complement AI and machine learning methods that detect only patterns
Ng
in data (such as the current generation of foundation models) with the explicit
nh
knowledge, rules and semantics provided by knowledge graphs.
Bì
nh

The desire to make better use of unstructured data held in documents,

Mạ

■
correspondence, images and videos, using standardized metadata that can be
ng
oà

related and managed and provide the foundation for AI-ready data.
_H
MB

■ The increased usage of knowledge graphs with large language models (LLMs) to
NH

provide enhanced contextual understanding when answering questions on large

YE

quantities of enterprise data.

U
NG

■ The increasing awareness of the use of knowledge graphs in consumer products

and services, such as smart devices and voice assistants, chatbots, search engines,
24
/20

recommendation engines and route planning.

/12

■ The emerging landscape of Web3 applications and the need for data access across
04

trust networks, leading to the creation of decentralized knowledge graphs to build

:21
12

immutable and queryable data structures.

ê n_

The need to manage the increasing number of data silos where data is often
uy

■
Ng

duplicated, and where meaning, usage and consumption patterns are not well-
ình

defined.
hB

The use of graph algorithms and machine learning to identify influencers, customer
ạn

■
gM

segments, fraudulent activity and critical bottlenecks in complex networks.

àn
Ho
B_

Gartner, Inc. | G00809438 Page 98 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Awareness of knowledge graph use cases is increasing, but business value and

à
Ho
relevance are difficult to capture in the early implementation stages.

_
MB
■ Moving knowledge graph models from prototype to production requires engineering

NH
and system integration expertise. Methods to maintain knowledge graphs as they

E
UY
scale — to ensure reliable performance, handle duplication and preserve data quality

NG
— remain immature.

4
■ The graph DBMS market is fragmented along three properties: type of data model

02
2/2
(Resource Description Framework or property), implementation architecture (native

4/1
or multimodal) and optimal workload (operational or analytical). This fragmentation

10
continues to cause confusion and hesitation among adopters.

2:2
_1
■ Organizations want to enable the ingestion, validation and sharing of ontologies and

ên
data relating to entities (such as geography, people and events). However, making

uy
internal data interoperable with external knowledge graphs is a challenge.
Ng
nh
In-house expertise, especially among subject matter experts, is lacking, and
Bì

■
nh

identifying third-party providers is difficult. Often, expertise resides with vendors of

Mạ

graph technologies. Skills in scalability and optimization are also hard to acquire.
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 99 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
User Recommendations

ng
■ Create a working group of knowledge graph practitioners and sponsors by

à
Ho
assessing the skills of data and analytics (D&A) leaders, practitioners and business

_
MB
domain experts. Factors like use case requirements, data characteristics, scalability

NH
expectations, query flexibilities and domain knowledge of knowledge graphs should

E
be addressed.

UY
NG
■ Run a pilot to identify use cases that need custom-made knowledge graphs. The
pilot should deliver not only tangible value for the business, but also learning and

4
02
development for D&A staff.

2/2
4/1
■ Create a minimum viable subset that can capture the information of a business

10
domain to decrease time to value. Assess the data, both structured and

2:2
unstructured, needed to feed a knowledge graph, and follow Agile development

_1
principles.

ên
uy
Utilize vendor and service provider expertise to validate use cases, educate
Ng
■
stakeholders and provide an initial knowledge graph implementation.
nh
Bì
nh

■ Include knowledge graphs within the scope of D&A governance and management.
Mạ

To avoid perpetuating data silos, investigate and establish ways for multiple
ng

knowledge graphs to interoperate and extend toward a data fabric.

oà
_H
MB

Sample Vendors
NH

Cambridge Semantics; Diffbot; eccenca; Fluree; Neo4j; Ontotext; Stardog; TigerGraph;

YE

TopQuadrant
U
NG

Gartner Recommended Reading

24

How to Build Knowledge Graphs That Enable AI-Driven Enterprise Applications

/20
/12

3 Ways to Enhance AI With Graph Analytics and Machine Learning

04
:21
12

How Large Language Models and Knowledge Graphs Can Transform Enterprise Search
ên_
uy

Intelligent Applications
Ng

Analysis By: Justin Tung, Stephen Emmott, Alys Woodward

ình
hB

Benefit Rating: Transformational

ạn
gM
àn

Market Penetration: 5% to 20% of target audience

Ho
B_

Gartner, Inc. | G00809438 Page 100 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Maturity: Early mainstream

ng
à
Definition:

_Ho
MB
Intelligent applications utilize learned adaptation to respond autonomously to people and

NH
machines. While applications can behave intelligently, intelligent applications are

E
inherently smart/proactive. Rule-based approaches based on conditional logic are giving

UY
way to math-based training to elicit an appropriate response across a range of

NG
circumstances, including those that are new or unique. This enables the augmentation
and automation of work across a broad variety of scenarios and use cases.

4
02
2/2
Why This Is Important

4/1
10
AI is the current competitive play for enterprise applications, with many technology

2:2
providers now enabling AI in their products via inbuilt, added, proxied or custom

_1
capabilities. Recent developments in AI are continuing to enable applications to work

ên
autonomously across a wider range of scenarios with elevated quality and productivity.

uy
Ng
Integrated intelligence and AI can also support decision-making processes alongside
nh
transactional processes.
Bì
nh

Business Impact
Mạ

Three benefits to organizations buying or augmenting intelligent applications are:

ng
oà
_H

■ Automation — They increase automated and dynamic decision making, reducing the
MB

cost and unreliability of human intervention, and improving effectiveness of

NH

business processes.
UYE
NG

■ Augmentation — They increase the speed and quality of dynamic decision making
based on context and risk, whether automated or via improved decision support.
24
/20

■ Contextualization — Applications can adapt to the context of the user or process,

/12

creating personalized experiences.

04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 101 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Drivers

ng
■ The hype wave for generative AI and large language models (LLMs), and the

à
Ho
prevalence of conversational user interfaces (UIs) as a way to interact with them,

_
MB
has inspired innovation and surfaced valuable ways to add AI functionality within

NH
preexisting applications. Features such as recommendations, insights and

E
personalization are now more easily accessible via natural language prompts.

UY
Looking ahead, wider incorporation of chat-based interfaces will blur the line

NG
between interface and intelligence in an easily composable manner.

4
02
■ AI capabilities and features are increasingly integrated into ERP, CRM, digital

2/2
workplace, supply chain and knowledge management software within enterprise

4/1
application suites. Embedded generative AI (as detailed above with LLMs) and

10
traditional AI capabilities, such as predictive analytics, help to derive more insights

2:2
from data in such applications. The 2023 Gartner AI in the Enterprise Survey shows

_1
ên
that the top way to fulfill GenAI use cases is to use GenAI embedded into existing

uy
(purchased) applications (see Survey Shows How GenAI Puts Organizational AI
Maturity to the Test for more information). Ng
nh
Bì

■ Organizations are demanding more functionalities from applications, whether built

nh

or bought, expecting them to enhance current processes for both transactions and
Mạ

decision making with recommendations, insights and additional information. This in

ng
oà

turn allows vendors to deliver higher value and drives higher prices.
_H
MB

■ The trend toward composable business architectures has highlighted the

NH

possibilities for delivering advanced and flexible capabilities to support, augment

YE

and automate decisions, which have traditionally required an underlying data fabric
U

and packaged capabilities to build. However, the increased adoption of LLMs has
NG

the potential to be used as a composable interface layer, kick-starting the ability to

24

deliver on the composable architecture.

/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 102 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Obstacles

ng
■ Lack of data — Intelligent applications require access to data from a range of

à
Ho
systems, meaning application vendors need to think about data management

_
MB
technology and processes outside their own solutions.

NH
■ Adding AI adds complexity to operations — Models have to be trained and

E
UY
maintained, and users must understand what data is being used. Contextualizing

NG
insights requires business metadata.

4
■ Overuse of AI in marketing — Vendors sometimes neglect the focus on business

02
2/2
impact, which can generate a cynical response in business buyers, particularly when

4/1
AI has not delivered value in the past.

10
2:2
■ Trust in system-generated insights — It takes time for business users to see the

_1
benefit and trust such insights and some explainability is key.

ên
uy
Ng
User Recommendations nh
■ Challenge your packaged software providers to outline in their product roadmaps
Bì

and/or ecosystems how they are incorporating AI to add business value in the form
nh
Mạ

of a range of AI technologies.
ng
oà

■ Evaluate the architecture of your providers by considering that the best-in-class

_H

intelligent applications are built from the ground up to be constantly collecting data
MB

from other systems, with a solid data layer in the form of a data fabric.
NH
YE

■ Prioritize investments in specialized and domain-specific intelligent applications

U

delivered as point solutions, which help solve problem areas such as customer
NG

engagement and service, talent acquisition, collaboration, and engagement.

24

Bring AI components into your composable enterprise to innovate faster and safer, to
/20

■
reduce costs by building reusability, and to lay the foundation for business-IT
/12
04

partnerships. Remain aware of what makes AI different, particularly how to refresh

:21

ML models to avert implementation and usage challenges.

12
ên_

Sample Vendors
uy
Ng

Alkymi; ClayHR; Creatio; Eightfold AI; JAGGAER; OpenText; Prevedere; Salesforce; Sievo;
ình

SugarCRM
hB
ạn

Gartner Recommended Reading

gM

Top Tech Provider Trend for 2023: Intelligent Applications

àn
Ho
B_

Gartner, Inc. | G00809438 Page 103 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Survey Shows How Generative AI Puts Organizational AI Maturity to the Test

ng
à
Ho
Maximize Competitiveness in Banking With Behavioral and Data Science

_
MB
NH
E
UY
NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 104 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Entering the Plateau

ng
Computer Vision

à
Ho
Analysis By: Nick Ingelbrecht

_
MB
NH
Benefit Rating: Transformational

E
UY
NG
Market Penetration: 20% to 50% of target audience

4
Maturity: Early mainstream

02
2/2
4/1
Definition:

10
Computer vision is a set of technologies that involve capturing, processing and analyzing

2:2
real-world images and videos to extract meaningful, contextual information from the

_1
ên
physical world.

uy
Why This Is Important
Ng
nh
Bì

Computer vision (CV) comprises a transformational collection of technologies, including

nh

AI/generative AI, advanced sensors and analytics that are essential to sensing and
Mạ

understanding the environment. Computer vision technology is driving innovation across

ng

many industries and use cases and is creating unprecedented business applications and
oà

opportunities.
_H
MB
NH

Business Impact
YE

CV technologies are used across all industries, and address a broad and growing range of
U
NG

business applications. These include physical security, healthcare, retail, automotive,

robotics, manufacturing, supply chain/logistics, banking and finance, agriculture,
24

government, and media and entertainment. Computer vision operates in the visible and
/20

nonvisible spectrum, including infrared, hyperspectral imaging, lidar, radar and ultraviolet.
/12
04

Drivers
:21

CV adoption is driven by an increasing demand for automation to reduce costs and

12
n_

improve monitoring and response capabilities. Other drivers include:

ê
uy
Ng

■ Improvements in the availability and application of machine learning (ML) methods,

ình

tools and services, hardware processing efficiencies, and data generation and
hB

augmentation techniques.
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 105 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Multimodal training of CV with large language models (LLMs) facilitates natural

ng
language contextual search of unstructured image data and correlation of data on

à
Ho
video streams at scale.

_
MB
■ New architectures, models and algorithm enhancements steadily improve the

NH
price/performance of CV applications. Combinations of convolutional neural

E
UY
networks (CNNs) and vision transformers are delivering leading levels of

NG
performance.

Advances in image and video generation, such as Google Lumiere and OpenAI Sora,

4
■

02
have broken new ground in the sophistication and realism of text to video

2/2
generation.

4/1
10
■ The proliferation of cameras and other sensors is generating exponential increases

2:2
in image data, creating a critical and growing demand for methods to automate

_1
ên
analysis and manage and extract value from that data. Dynamic vision systems are

uy
now being integrated into smartphones and lower-cost lidar products are opening
new innovation areas. Ng
nh
Bì

■ 3D capture, modeling and editing of real-world objects and environments have been
nh
Mạ

enabled by new techniques such as Proximity Attention Point Rendering (PAPR) and
ng

applications of 3D Generative AI.

oà
_H

■ Edge-enabled frameworks, developer ecosystems, model compression and chip

MB

advancements.
NH
YE

■ New business models and applications range from smartphone cameras and filters,
U

through to global video content production and distribution, life-saving medical

NG

image diagnostics, autonomous vehicles, video surveillance for security, robotics,

and manufacturing automation.
24
/20

Sensor fusion, multispectral and hyperspectral imaging expand the range of

/12

■
applications.
04
:21

■ Improved reliability, price, performance and functionality generate compelling

12
n_

business value.
ê
uy

Open-world recognition using GenAI can identify and classify known objects, as well
Ng

■
as handle unknown/unseen classes of objects and activities in novel contexts,
ình
hB

without the need to train a model on specific examples.

ạn
gM

Obstacles
àn
Ho
B_

Gartner, Inc. | G00809438 Page 106 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ High-end systems are expensive, and building business cases with adequate ROI is

ng
challenging.

à
Ho
■ The CV market lacks independent standardization and performance

_
MB
benchmarks/KPIs, advanced solutions are far from being commoditized, and

NH
reliability remains an obstacle for mission-critical applications like autonomous

E
UY
vehicles.

NG
■ Integration is problematic due to a lack of open interfaces.

4
02
■ Enterprises struggle to activate CV models in business processes and face data

2/2
security as well as organizational challenges and user resistance to visual

4/1
monitoring.

10
2:2
■ Scaling solutions is challenging due to the hardware costs and high levels of

_1
customization and service support.

ên
uy
Ng
■ Adequate training data may be hard or expensive to acquire, especially in areas
nh
where available open-source CV datasets are declining.
Bì
nh

■ Proprietary algorithms and patent pools deter innovation.

Mạ
ng

■ Ethical, privacy and regulatory issues have emerged, including the use of deepfakes
oà

for embezzlement, misleading advertising and blackmail, as well as the capture of

_H

facial and other biometric data and the impact of new CV technologies on copyright
MB

and authenticity.
NH
UYE

User Recommendations
NG

■ Assess change management impacts of CV projects on the organization and its

24

people.
/20
/12

■ Focus initially on a few small projects, using fail-fast approaches, and scale the
04

most promising systems into production using cross-disciplinary teams.

:21
12

■ Test production systems early in the real-world environment because lighting, color,
n_

object disposition and movement can break CV solutions that worked well in the
ê
uy

development cycle.
Ng
ình

■ Build internal CV competencies and processes for exploiting image and video
hB

assets.
ạn
gM

■ Exploit third-party CV tooling and services to accelerate data preparation and reduce
àn

costs.
Ho
B_

Gartner, Inc. | G00809438 Page 107 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
■ Evaluate legal, regulatory, commercial and reputational risks associated with CV

ng
projects at the outset.

à
Ho
■ Reduce the barrier to CV adoption by addressing two of the main challenges — lack

_
MB
of training data, and costly and constrained hardware — by investing in synthetic

NH
and augmented data solutions and model compression to improve model

E
UY
performance and expand the range of more valuable use cases.

NG
Sample Vendors

4
02
Adobe; Amazon Web Services; Baidu; Clarifai; Dragonfruit AI; Landing AI; Matroid;

2/2
Microsoft; Prophesee; Tencent

4/1
10
Gartner Recommended Reading

2:2
_1
Emerging Tech Impact Radar: Computer Vision

ên
uy
Ng
Emerging Tech: Revenue Opportunity Projection of Computer Vision nh
Bì

Innovation Guide for Generative AI in Computer Vision

nh
Mạ

Emerging Tech: Revenue Opportunity Projection of Computer Vision: Growth Markets

ng
oà
_H

Emerging Technologies: Emergence Cycle for Computer Vision

MB
NH

Appendixes
UYE

See the previous Hype Cycle: Hype Cycle for Artificial Intelligence, 2023
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 108 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Hype Cycle Phases, Benefit Ratings and Maturity Levels

ng
Table 2: Hype Cycle Phases

à
Ho
(Enlarged table in Appendix)

_
MB
NH
E
UY
NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà
_H
MB
NH
UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 109 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Table 3: Benefit Ratings

à ng
Benefit Rating Definition

_ Ho
MB
Transformational Enables new ways of doing business across

NH
industries that will result in major shifts in

E
UY
industry dynamics

NG
High Enables new ways of performing horizontal

4
or vertical processes that will result in

02
2/2
significantly increased revenue or cost
savings for an enterprise

4/1
10
2:2
Moderate Provides incremental improvements to

_1
established processes that will result in

ên
increased revenue or cost savings for an

uy
enterprise
Ng
nh
Bì

Low Slightly improves processes (for example,

nh

improved user experience) that will be

Mạ

difficult to translate into increased revenue

ng

or cost savings.
oà
_H
MB
NH

Source: Gartner (June 2024)

UYE
NG
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 110 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Table 4: Maturity Levels

ng
(Enlarged table in Appendix)

à
_Ho
MB
NH
E
UY
NG
4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
nh
Bì
nh
Mạ
ng
oà

Document Revision History

_H
MB

Hype Cycle for Artificial Intelligence, 2023 - 19 July 2023

NH
YE

Hype Cycle for Artificial Intelligence, 2022 - 8 July 2022

U
NG

Hype Cycle for Artificial Intelligence, 2021 - 29 July 2021

Hype Cycle for Artificial Intelligence, 2020 - 27 July 2020

24
/20

Hype Cycle for Artificial Intelligence, 2019 - 25 July 2019

/12

Hype Cycle for Artificial Intelligence, 2018 - 24 July 2018

04
:21

Hype Cycle for Artificial Intelligence, 2017 - 24 July 2017

12
n_

Hype Cycle for Smart Machines, 2016 - 21 July 2016

ê
uy

Hype Cycle for Smart Machines, 2015 - 24 July 2015

Ng
ình

Hype Cycle for Smart Machines, 2014 - 18 July 2014

hB
ạn
gM

Recommended by the Authors

àn
Ho
B_

Gartner, Inc. | G00809438 Page 111 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 N
ình
n hB
Mạ
Some documents may not be available as part of your current Gartner subscription.

ng
Understanding Gartner’s Hype Cycles

à
Ho
Tool: Gartner’s Hype Cycle Builder

_
MB
AI Design Patterns for Large Language Models

NH
E
10 Best Practices for Scaling Generative AI Across the Enterprise

UY
NG
When Not to Use Generative AI
How to Build Knowledge Graphs That Enable AI-Driven Enterprise Applications

4
02
2/2
4/1
10
2:2
_1
ên
uy
Ng
© 2024 Gartner, Inc. and/or its affiliates. All rights reserved. Gartner is a registered trademark of
nh
Gartner, Inc. and its affiliates. This publication may not be reproduced or distributed in any form
Bì

without Gartner's prior written permission. It consists of the opinions of Gartner's research
nh
Mạ

organization, which should not be construed as statements of fact. While the information contained in
this publication has been obtained from sources believed to be reliable, Gartner disclaims all warranties
ng
oà

as to the accuracy, completeness or adequacy of such information. Although Gartner research may
_H

address legal and financial issues, Gartner does not provide legal or investment advice and its research
MB

should not be construed or used as such. Your access and use of this publication are governed by
NH

Gartner's Usage Policy. Gartner prides itself on its reputation for independence and objectivity. Its
YE

research is produced independently by its research organization without input or influence from any
U

third party. For further information, see "Guiding Principles on Independence and Objectivity." Gartner
NG

research may not be used as input into or for the training or development of generative artificial
intelligence, machine learning, algorithms, software, or related technologies.
24
/20
/12
04
:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 112 of 112

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 4
10
2:2
_1
ên
uy
Table 1: Priority Matrix for Artificial Intelligence, 2024

Ng
nh
Bì
Benefit Years to Mainstream Adoption

nh
Mạ
Less Than 2 Years 2 - 5 Years 5 - 10 Years More Than 10 Years

ng
oà
Transformational Composite AI Decision Intelligence Autonomic Systems Artificial General Intelligence

_H
Computer Vision Embodied AI Autonomous Vehicles

MB
Foundation Models First-Principles AI

NH
Generative AI Neuromorphic Computing

E
UY Intelligent Applications
Responsible AI
NG

High Edge AI AI Engineering AI Simulation

24

AI-Ready Data ModelOps

/20

AI TRiSM Multiagent Systems

/12

Causal AI Neuro-Symbolic AI
04

Cloud AI Services Smart Robots

Knowledge Graphs
:21

Prompt Engineering
12
n_

Sovereign AI
ê

Synthetic Data
uy
Ng

Moderate
ình

Low Quantum AI
hB
ạn
gM

Source: Gartner (June 2024)

àn
Ho
B_

Gartner, Inc. | G00809438 Page 1A of 4A

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 4
10
2:2
_1
ên
uy
Table 2: Hype Cycle Phases

Ng
nh
Bì
Phase Definition

nh
Mạ
Innovation Trigger A breakthrough, public demonstration, product launch or other event
generates significant media and industry interest.

ng
oà
Peak of Inflated Expectations During this phase of overenthusiasm and unrealistic projections, a flurry of

_H
well-publicized activity by technology leaders results in some successes, but

MB
more failures, as the innovation is pushed to its limits. The only enterprises

NH
making money are conference organizers and content publishers.
E
UY
Trough of Disillusionment Because the innovation does not live up to its overinflated expectations, it
NG

rapidly becomes unfashionable. Media interest wanes, except for a few

cautionary tales.
24
/20

Slope of Enlightenment Focused experimentation and solid hard work by an increasingly diverse
range of organizations lead to a true understanding of the innovation’s
/12

applicability, risks and benefits. Commercial off-the-shelf methodologies and

04

tools ease the development process.

:21
12

Plateau of Productivity The real-world benefits of the innovation are demonstrated and accepted.
n_

Tools and methodologies are increasingly stable as they enter their second
ê
uy

and third generations. Growing numbers of organizations feel comfortable

Ng

with the reduced level of risk; the rapid growth phase of adoption begins.
ình

Approximately 20% of the technology’s target audience has adopted or is

hB

adopting it as it enters this phase.

ạn

Years to Mainstream Adoption The time required for the innovation to reach the Plateau of Productivity.
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 2A of 4A

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 4
10
2:2
_1
ên
uy
Source: Gartner (June 2024)

Ng
nh
Table 3: Benefit Ratings

Bì
nh
Mạ
Benefit Rating Definition

ng
Transformational Enables new ways of doing business across industries that will result in

oà
_H
major shifts in industry dynamics

MB
High Enables new ways of performing horizontal or vertical processes that will

NH
result in significantly increased revenue or cost savings for an enterprise

E
UY
Moderate Provides incremental improvements to established processes that will result
NG

in increased revenue or cost savings for an enterprise

Low Slightly improves processes (for example, improved user experience) that will
24

be difficult to translate into increased revenue or cost savings.

/20
/12
04

Source: Gartner (June 2024)

:21
12
ên_
uy
Ng
ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 3A of 4A

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG
 4
10
2:2
_1
ên
uy
Table 4: Maturity Levels

Ng
nh
Bì
Maturity Levels Status Products/Vendors

nh
Mạ
Embryonic In labs None

ng
Emerging Commercialization by vendors First generation

oà
Pilots and deployments by industry leaders High price

_H
Much customization

MB
NH
Adolescent Maturing technology capabilities and process Second generation

E
UY understanding Less customization
Uptake beyond early adopters
NG

Early mainstream Proven technology Third generation

Vendors, technology and adoption rapidly evolving More out-of-box methodologies
24
/20

Mature mainstream Robust technology Several dominant vendors

/12

Not much evolution in vendors or technology

04

Legacy Not appropriate for new developments Maintenance revenue focus

:21

Cost of migration constraints replacement

12
n_

Obsolete Rarely used Used/resale market only

ê
uy
Ng

Source: Gartner (June 2024)

ình
hB
ạn
gM
àn
Ho
B_

Gartner, Inc. | G00809438 Page 4A of 4A

HM
EN

This research note is restricted to the personal use of [email protected].

UY
NG