Lecture 2: Fundamentals of AI & ML

Course code: DES646

Course name: AI/ML for designers

Dr Amar Behera
Email: [email protected]
Extension: 2401

Design Studio 605H-A

Diamond Jubilee Academic Complex
Learning Outcomes
• Understand the structure and types of AI systems
• Learn what Machine Learning is and how it relates to AI
• Explore examples of AI applications relevant to design
• Distinguish between different ML types: supervised, unsupervised,
reinforcement
Last class, we understood: AI vs ML vs DL

Artificial Intelligence
• AI is a broad field that focuses on creating systems capable of tasks typically requiring human intelligence—like
reasoning, learning, perception, and language understanding.

Machine Learning
• ML is a subfield of AI where systems learn patterns from data and improve performance without being explicitly
programmed.

Deep Learning
• DL is a specialized subset of ML that uses multi-layered neural networks to model complex data relationships.
• It's the backbone of tools like ChatGPT and DALL·E.
It is helpful to think
about:
AI vs Data Science vs
ML
• Computer Science > Data Science
> Artificial Intelligence > Machine
Learning
• AI: cognitive tasks
• ML: data-driven learning
• Deep Learning: layered ML
architecture using neural
networks
4 types of AI
• Reactive:
• Most basic level
• AI systems only respond to current stimuli
without the ability to learn or store
information from past experiences.
• Examples include a thermostat adjusting
temperature based on the room
temperature or a self-driving car reacting to
a red light.
• Limited Memory:
• These AI systems can store and access past
data to improve their decision-making over
time, allowing them to learn and adapt
based on previous experiences.
• Many modern AI systems fall into this
category, like virtual assistants that can
personalize responses based on past
interactions.
4 types of AI
• Theory of Mind:
• This hypothetical level of AI would
represent a significant leap in
capability, allowing machines to
understand and respond to human
emotions, intentions, and beliefs.
• This type of AI is currently the subject
of much research and speculation.
• Self-Aware:
• This is the most advanced and
currently non-existent form of AI,
envisioned as a machine that
possesses a sense of self and
consciousness, similar to humans.
• This concept remains largely within
the realm of science fiction.
Core
capabilities of AI
• Perception and sensing
• AI systems process input from the world
(e.g., images, sound, text) using sensors
and interpret it using models.
• Reasoning and decision making
• AI can simulate logical reasoning and make
decisions based on programmed rules or
learned models.
• Learning and adaptation
• Machine Learning allows AI systems to
improve their performance over time by
learning from new data.
Input command vs Input data:
Traditional programming vs ML
Traditional programming

In the ML paradigm, we can train a

device to
• Make autonomous decisions
• Adapt and modify assumptions
based on provided data and on
encountered errors
• Similar to how human beings learn
from their experience

Machine learning
The Machine Learning Pipeline
• Gathering relevant, quality data is the first and most
Data collection critical step. In design, this could be user interactions,
moodboard ratings, etc.

• Data is fed into algorithms (e.g., decision trees, neural

Model training nets) to learn patterns. This phase requires iteration
and tuning.

Evaluation and • Trained models are tested, improved, and deployed in

real environments—like UX feedback tools or visual
deployment recommender systems.
 Training data is used to develop the
model

Training data vs Test data is used to verify whether

Test data the achieved model is reliable or not

ML model is considered ready when

result is satisfactory both in training
and test phases
Types of
machine
learning
Key ML
algorithms
How neural networks work
Neural Networks & Complex Pattern Recognition
• Neural networks excel at solving
complex patterns
– Tasks hard for machines but
easy for humans
• Example: CAPTCHAs
– “Completely Automated Public
Turing test to tell Computers and
Humans Apart”
– Tasks like:
• Identifying stretched/distorted
letters
• Clicking specific image areas
Neural Networks &
Complex Pattern
Recognition
• Other Applications
– Challenge-response tests
– Pedestrian path prediction in
autonomous vehicles
• Why it works
– Deep neural networks simplify
high-dimensional, diverse input
data
– Each layer extracts increasingly
abstract features
A little bit more
on CAPTCHA
• Determines if a user is human or
a bot
• Used to manage bot activity on
websites
• Automated and embedded in
web pages
A little bit more
on CAPTCHA
• Classic CAPTCHA: distorted
letters to be identified
• Humans interpret distorted text;
bots struggle
• Common in login, signup, polls,
checkout pages
Why CAPTCHA works

• Humans can recognize patterns despite distortions

• Bots often fail or guess randomly
• More complex tests needed as bots evolve
What is
reCAPTCHA?
• Developed at Carnegie Mellon,
acquired by Google
• Uses real-world images: street
signs, books, etc.
• Advanced forms use AI and
behavior analysis
Examples of reCAPTCHA
 Image recognition: select
images with specific objects

Types of
reCAPTCHA Checkbox: 'I'm not a robot'
with cursor movement analysis
tests

No interaction: behavior-based
background checks
How Image Recognition
reCAPTCHA Works
• User selects objects (e.g., traffic lights) in blurry
images
• Human perception outperforms bots in this task
How Checkbox reCAPTCHA Works
• Movement toward checkbox is
analyzed
• Randomness in human motion
helps detect bots
• Cookies and device history also
assessed
reCAPTCHA
Without
Interaction
• Observes user's online
behavior and interaction
history
• Assigns risk score (0.0 to
1.0) for bot detection
What Triggers
a CAPTCHA?
• Automatically placed on websites
• Triggered by unusual user
behavior (e.g., rapid clicks)
Are CAPTCHAs
enough?

• Advanced bots and click farms can

bypass them
• CAPTCHA should be part of a larger bot
mitigation strategy
 Interrupts user
experience
Drawbacks
Not accessible for
of visually impaired users
CAPTCHAs
Can be bypassed by bots
Alternatives to
CAPTCHA
• Bot management tools (e.g.,
Cloudflare Bot Management)
• Can identify bad bots without
impacting the user experience,
based on the behaviour of the bot.
• Turnstile: invisible CAPTCHA
alternative by Cloudflare
• Invisible alternative to CAPTCHA
that uses a snippet of free code.
CAPTCHA,
reCAPTCHA & AI
• As millions of users identify hard-
to-read text and pick out objects
in blurry images, that data is fed
into AI computer programs so
that they become better at those
tasks as well.
• User-labeled data trains AI
models (text and image
recognition)
• Helps machines recognize objects
in diverse real-world contexts
Turing Test vs
CAPTCHA
• Turing Test: checks if machine
mimics human
• CAPTCHA: checks if user is *not*
a machine
• CAPTCHA assigns tasks humans
do well, bots don’t
 AI
Applications
Across Key
Sectors
 • Manufacturing

• Smart robots improve production speed, flexibility, and

decision-making

AI
• Design & Engineering

• AI assists in optimization, error prediction, and

Applications performance improvements

• Security
Across Key • Pattern recognition for cyber-attack prevention and

Sectors threat detection

• Transportation

• Traffic control, route optimization, autonomous

vehicles in development
 • AI enhances design by offering new
capabilities for data analysis, ideation, and
visualization
• ML is a core AI approach with powerful tools
Summary like supervised learning and deep networks
• Designers must understand how AI works to
use it ethically and effectively