See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/373002717

What if GPT4 Became Autonomous: The Auto-GPT Project and Use Cases

Article in Journal of Emerging Computer Technologies · August 2023

DOI: 10.57020/ject.1297961

CITATIONS READS

5 240

2 authors:

Mehmet Firat Saniye Kuleli

Anadolu University Ministry of National Education Turkey
150 PUBLICATIONS 1,463 CITATIONS 5 PUBLICATIONS 7 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Mehmet Firat on 13 August 2023.

The user has requested enhancement of the downloaded file.

 Journal of Emerging Computer Technologies
Research Article
Received: 2023-05-17 | Reviewing: 2023-05-17 & 2023-08-02| Accepted: 2023-08-02 | Online: 2023-08-07 | Issue Date: 2023-12-31
Doi: 10.57020/ject.1297961

What if GPT4 Became Autonomous: The Auto-GPT

Project and Use Cases
Mehmet Fırat Saniye Kuleli
Department of Distance Education İzmir Provincial Directorate of National Education
Open Education Faculty İzmir,Türkiye
Anadolu University [email protected]
Eskişehir, Türkiye 0000-0001-7838-4997
[email protected] (Corresponding Author)
0000-0001-8707-5918

Abstract— Auto-GPT is a product of an experimental project without human intervention. Given a user query, Auto-GPT
that makes the use of GPT-4 autonomous. Notably, Auto-GPT performs step-by-step actions to realize the goal
emerged and spread rapidly, while the echo of OpenAI's autonomously. In addition to reasoning about the current
ChatGPT continues. However, there are insufficient studies on situation, Auto-GPT can also use past actions to improve
this new application in related literature. The purpose of this
exploratory case study was to explore the different use cases and
decision making [4]. Autonomous AI can learn, think and act
experiences of Auto-GPT users. For this purpose, 16 users with without continuous input from humans [5]. In this way, it
an Auto-GPT experience on the GitHub platform were aims to obtain the result by sequentially generating
interviewed. Thematic content analysis was performed on the commands from the given basic commands and their results.
qualitative data. AutoGPT experiences of users can be While there are evidence suggesting that scaling up Large
characterized by learning programs, autonomous applications, Language Models (LLM) can lead to certain level of general
conducting research, and writing reports. The results of this intelligence, there are still limitations as the lack of long-term
study showed that content creation is the most important memory, limited token length, and the lack of deterministic
purpose of using Auto-GPT. As independent research functions control over its behaviors, etc. [6].
of Auto-GPT, users also emphasize data summarization and
information organization. However, the participants also II. IS AAI POSSIBLE?
pointed out the token limit (inefficiency), forgetting generated
tools, and iteration as some prominent limitations of Auto-GPT. The ChatGPT natural language processing model was
It is possible to say that Auto-GPT has a high potential to use in publicly launched in November 2022 and scaled at an
also in educational purpose, but it is still in the development unprecedented rate, reaching 1 million users in five days
stage. compared to 300 days for Facebook, 720 days for Twitter,
and 75 days for Instagram. The most important factor
Keywords— Auto-GPT, GPT-4, ChatGPT, Use Cases underlying this rapid spread is its text production and
I. INTRODUCTION processing feature. ChatGPT can generate high-quality,
plausible, human-like written responses to statistical
What artificial intelligence does and can do has analyses, texts, computer programs, abstracts, or
undoubtedly been one of the most important topics of introductions to scientific articles[7]. ChatGPT is able to do
discussion nowadays among technology developers and this by taking advantage of its extensive data stores and
users. Natural language processing models have become efficient design to understand and interpret user requests and
increasingly popular in recent years and are recognized then generate appropriate responses in early natural human
worldwide as an important milestone in the development and language[8]. The use of ChatGPT also presents various
use of artificial intelligence [1]. A succession of fast and potential problems, such as the generation of incorrect or fake
efficient models reshapes the way we interact with AI and the information and student plagiarism, and for students it is
digital world. In particular, OpenAI's GPT series, with its crucial to introduce them to the limitations of ChatGPT, such
innovative features, has led to exciting developments in the as its reliance on biased data, limited current knowledge, and
emergence of opportunities arising from the collaboration of the potential for generating incorrect or fake information [9].
advanced language models with human and artificial Developed using the GPT-3.5 language model, an improved
intelligence. Natural language processing is a subset of version of GPT-3, ChatGPT is the result of OpenAI's ongoing
artificial intelligence which refers to the automatic work on increasingly secure and useful artificial intelligence
computational processing of human speech, and includes systems [10]. In March 2023, shortly after ChatGPT was
systems which use human-generated text as input and launched, the GPT 4 is available as final version. GPT-4
systems which produce natural-sounding text as output [2]. builds on the success of its predecessors, GPT-3 [11] and
Auto-GPT is a product of an experimental project GPT-2 [12]. These models have shown significant
developed to make the use of GPT-4 autonomous. Auto-GPT improvements in their ability to understand and generate
is a Python application that is available as an open source on natural languages. With 175 billion parameters, GPT-3 sets
GitHub [3]. Auto-GPT is powered by self-managing artificial new standards for several natural language processing tasks
intelligence agents that can self-request commands and [13]. In the GPT-4 model, the number of parameters reached
execute tasks with minimal human intervention. Self- trillions, significantly increasing its capacity compared to
triggered prompts enable complex research to be conducted previous models [14]. In addition to human feedback-
Cite (APA) : Fırat, M., Kuleli, S. (2023). What if GPT4 Became Autonomous: The Auto-GPT Project and Use Cases. Journal of Emerging Computer Technologies, 3(1), 1-6. Doi:
10.57020/ject.1297961
Volume:3, No:1, Year: 2023, Pages: 1-6, December 2023, Journal of Emerging Computer Technologies
1
 Journal of Emerging Computer Technologies
Fırat and Kuleli
supported learning, GPT-4 has a rule-based reward model However, AI agents differ from objects in object-oriented
approach compared with GPT-3 [15]. computer programs in that they are autonomous, flexible, and
have their own control structures.
Chat GPT is able to accomplish with its language
processing capability, which has been an important milestone According to Totschnig [27], "In AI and AI applications,
in the history of artificial intelligence and made it the top of autonomy refers to an artificial entity's ability to act
the agenda in a short time. The use of language models like autonomously without human guidance, assuming a fixed
ChatGPT in education appears as a potential area of interest goal or utility for evaluating its actions. However, from a
due to their rich and wide range of applications [16]. Using philosophical point of view, this concept of autonomy seems
these models, it will be possible to create personalized and unlikely to be implemented, but the rapid proliferation of
effective learning experiences for individuals at every level autonomous AI applications, such as BabyAGI and
of education, in line with each individual's unique learning AutoGPT, which emerged in 2023, proved that the pace of
preferences, abilities, and needs [17]. Going forward, we development of artificial intelligence should not be
expect ChatGPT's performance to increase through deep underestimated.
learning, a promising development educators and researchers
AutoGPT, which performs operations using the GPT-4
should pay attention to for its potential applications in
model, can work autonomously and perform the steps of
teaching and learning [18]. However, it should be noted that
thinking, planning, and taking action independently. While
ChatGPT's ability to generate highly realistic texts poses a
working with LLMs, it automatically generates prompts in
potential risk to the integrity of online exams. Precautions
line with the given command and works until it reaches the
should be taken to prevent this from happening [19,20] LLMs
result, without the need for users to add any input.
can also be used to develop more natural and sophisticated
Considering that creating the correct prompt and presenting
user interfaces by exploiting their ability to generate
it to artificial intelligence is the most important step in
contextualized, personalized, and diverse responses to natural
achieving the desired result, AutoGPT will produce effective
language questions asked by users [21]. Chat GPT performs
results by saving users from this step. While there has been
the desired commands thanks to user prompts, that is, user
growing interest in Auto-GPT-styled agents, questions
instructions. A prompt is a chat message that the user sends
remain about the effectiveness and flexibility of Auto-GPT in
to an LLM, such as ChatGPT, in a chat-based environment
solving real-world decision problems [28].
[22]. In this case, the most important step in achieving high
efficiency from artificial intelligence is to write the correct While recent studies on the use of autonomous artificial
prompt. With the introduction of AutoGPT, the automatic intelligence in areas such as health [29], energy production
prompt generation and processing promised by autonomous [30], and genetics [31] are available in the literature, there are
artificial intelligence has caused users to turn their attention no studies in the field of open and distance learning. Studying
in this direction. the Chat GPT's ability to enhance autodidactic learning is
very important for several reasons. It can identify best
The autonomous transformation of AI has been a topic of
practices and approaches for using chatbots and other
discussion since the 1960s [23]. Following the launch of
artificial intelligence (AI) tools in education as well as
GPT-4 in March 2023, studies on the use of productive
provide guidance for the future of education and the use of
artificial intelligence technologies in almost every sector
technology in learning [32]. Auto-GPT combines GPT-4's
came to the forefront and created a big agenda in the first
powerful natural language processing capabilities with the
quarter of 2023. In natural-language processing-based
ability to auto-author prompts in a single framework,
generative artificial intelligence models, directing the right
enabling a large number of tasks to be automated with a high
prompt to artificial intelligence is the key to the efficiency of
level of precision and efficiency. This development shows
autonomous artificial intelligence. AutoGPT, an open-source
that LLMs, which could be an important future solution tool
Python project, was created to fill an important gap in helping
for AI, have the potential to radically transform the way we
users reach their targeted results by generating automatic
look at business automation. Despite this, studies on Auto-
prompts from the given commands. Auto-GPT proposes the
GPT and its potential applications are very limited in the
goal that attempts to make GPT-4 fully autonomous and try
relevant literature. In particular, the experiences of
to solve the problem that let ChatGPT interact with the
developers using Auto-GPT can provide important insights to
internet [24].The emergence of autonomous systems in an
explore the potential of this remarkable application.
increasing number of domains has made it imperative for AI
Determining the potential usage areas and limitations of
agents to deal with environmental uncertainty through
AutoGPT will shed light on future research in this field.
creativity [25]. An autonomous AI application is a system
that perceives an environment as part of it and works to III. RESEARCH PURPOSE
influence what it perceives in the future in line with its goals
In the first quarter of 2023, studies on the use of productive
over time. Wang [26] presents basic research on the
artificial intelligence technologies in almost every sector came
fundamental theories, discoveries and latest developments in
to the forefront and created a significant agenda. The
AAI and Symbiotic Human-Machine Intelligence.
Autonomous Artificial Intelligence (AAI) is recognized as introduction Auto-GPT has raised questions regarding the
the general form of AI equivalent to human Natural current use of generative artificial intelligence technologies
and in which areas they can be used effectively in the future.
Intelligence (NI), supported by intelligence science, brain-
This research aims to examine various use cases and
inspired systems, cognitive computers, intelligent
experiences of Auto-GPT users. Thus, this study aims to
mathematics and systems, humans, and cybernetics.

2
 Journal of Emerging Computer Technologies
Fırat and Kuleli
identify and understand the experiences of Auto-GPT users 5. Identifying and Naming Themes: The final step was
and to reveal the potential projects and best use cases. to create clear and unambiguous descriptions for each theme
and associated subthemes to identify and name the themes.
IV. METHOD
V. FINDINGS
This research was designed based on the case study
methodology proposed by Yin [33]. The exploratory case Themes were derived from the codes in the thematic
study method was preferred to investigate and interpret the content analysis. Thus, 21 themes were identified. The themes
usage areas and experiences of users on the Auto-GPT project, were sorted into four categories based on their common
which is available on the GitHub platform, comprehensively characteristics. The codes, themes, and categories obtained
and deeply. In the relevant literature, researchers have various from the content analysis are presented together with the
definitions and approaches pertaining to the case study [34, frequencies in Table I.
35, 36]. A case study is a method that examines a single TABLE I. QUALITATIVE FINDINGS
situation or event longitudinally, systematically collecting Categories Themes Code Frequencies
data and examining what happens in real-world settings. Yin Sales, customer
[37] defines case study as a research method used when the E-commerce 1
journeys
research focuses on "how" and "why" questions, when the Ticketing Customer queries 1
researcher has little or no control over the events, when you Open data portal,
Data summarization 2
study the event or phenomenon within its natural life news articles
framework, when the link between the event and real life is Cocktail list,
not clear enough. According to Yin [38], one of the features Content creation Python script, 3
that distinguishes a case study from other designs is that it is podcast outline
Tool generation Programs, files 1
combined with an existing theory and, depending on its Use Cases
Party schools in
outputs, three types of case studies are defined as exploratory, (f=14) Information
the US, market 2
descriptive, and explanatory. An exploratory case study organization
research
design is conducted to investigate a new situation, a topic that Autonomously
has not yet been sufficiently researched, and to fill the gap in Investment analysis gather and analyze 1
the literature [39]. market data
Product research Best headphones 1
In the research, data analyzed were gathered from the
Generate content,
discussion thread on the GitHub platform, where users share Social media
schedule posts, 1
their experiences and usage scenarios with Auto-GPT. For the management
customer inquiries
purpose of the research the open-ended question posed in the Stuck at limit
discussion was "What use cases have you found for Auto- Token limit 1
tokens
GPT?". Sixteen users responded to this open-ended query. Scraping articles
Inefficiency 1
The participants provided voluntary consent for the use of Limitations
online
their data anonymously. These comments were provided by Forgetting generated Not using tools
(f=4) 1
16 different users, offering diverse perspectives and potential tools effectively
applications. The responses from the users were subjected to Airtight systems,
content analysis. Qualitative data took advantage of thematic Iteration progress after 1
shutdowns
content analysis, a common method used to identify, analyze,
Data chunking Bypass token limit 1
and report patterns or themes in data. The thematic content GPT-4 Better tool
analysis was conducted in five stages [40]: Workarounds 1
improvements utilization
(f=3)
1. Familiarization: Initially, researchers carefully read Return to a session
Session continuity 1
through user comments to become familiar with the data and after it closes
AI addiction,
identify patterns or themes at the outset.
AI enthusiasm learning new 1
2. Coding: Each comment was systematically coded by technologies
assigning labels to specific sections of the text that represented User Learning
Python language 1
a particular idea, application, or experience. This process Experience programming
enabled the researchers to effectively organize and categorize (f=5) Outsource tasks to
AI agents 1
other AI agents
the data.
Autonomous AI Conduct research,
2
3. Identifying themes: The codes were examined and capabilities write reports
compared. Similar codes were grouped into broader themes. The theme with the highest frequency among the 21
These themes reflect the most important ideas and experiences themes obtained in the study was "Content Creation". This
shared by users. theme is in line with the feature of producing content on
4. Review and Refine Themes: The identified themes different topics, which is considered the most important
were reviewed and refined to ensure that they accurately outputs of use of GPT models. Direct quotes from two user
represented the data. This process aims to verify that the related to this theme are provided below.
themes are supported by coded data, and that there are clear “I tried to let it make "tools" (programs and files) that it
distinctions between different themes. would use while on projects. It made a lot and even impressive

3
 Journal of Emerging Computer Technologies
Fırat and Kuleli
ones i thought. … it would be cool to have it "externalize" and it is observed that the inefficiency and some shortcomings of
therefore reduce token cost.” AutoGPT are emphasized. These limitations can be
considered natural considering that AutoGPT is still under
“I used it to come up with a pretty basic cocktail list. All
development. Participants made suggestions to overcome
the cocktails and their recipes were quite good...”
these limitations. These suggestions included session
This theme was followed by "Data Summarization", continuity, removal of the token limit, and further
"Information Organization" and "Autonomous AI development of the GPT-4 model.
Capabilities" with two frequencies for each. These three
VI. DISCUSSIONS
themes are related to the AutoGPT's ability to conduct self-
search. When prompts are given in sufficient detail and in a The rising trend in AI development has led to discussions
cyclical structure, AutoGPT's independent research feature of the reduced use of manpower in many fields. Many studies,
comes to the fore. Direct quotes from two user's view on data especially after the introduction of ChatGPT, have revealed
summarization are given below. important findings regarding the breadth and diversity of its
use. AutoGPT, which was created for the autonomous use of
“Search and summarize data works pretty well. I let it
the GPT-4 model, can perform many operations such as
write me a list of all data available in the open data portal.”
content generation, access to web content, text summarization,
“I'm trying to summarize the news. But still stuck at limit decision-making, and situation analysis with the sequential
tokens...” commands it generates to execute the given command. While
there are studies in the literature on the use of AAI agents in
AutoGPT users emphasized the potential of AutoGPT for
areas such as health [41], energy production [42], and genetics
searching and summarizing data, but criticised the token
[43], it is important to determine the potential for use in
limits. About "information organization", a direct quote from education in general, and open and distance learning in
a user is given below. particular. AutoGPT's ability to perform operations by
“I used it to find the top 10 party schools in the US... and minimizing human intervention determines the areas of use in
then write each schools online application URL to a text file open and distance learning, and reveals the need to draw a
for easier applying. If only I had this 19 years ago when I was framework on how it can be used in this direction.
applying to college...”
The high number of students and the diversity of programs
In this theme, users emphasized the potential of AutoGPT in open and distance learning bring about the need for the
to organize information with the support of AI technology. effective use of AI technologies to save labor and time at many
Finally, here is a direct quote from a user on the topic of stages of teaching. Content creation, support services,
"Autonomous AI Capabilities". assessment, and evaluation are among the areas that need the
most improvement in terms of labor and time [44]. As a result
“…chunking large data into smaller bits that don't pass the of this research, AutoGPT's ability to perform more operations
token limit. I believe Auto-GPT does this when scraping in fewer steps, to produce content, to summarize the content
articles online but its not really efficient.” produced, and to provide guidance in line with the commands
Here in this theme, users criticize the autonomous AI entered, reveals that it can be used effectively in these steps of
capabilities with token limits that cause inefficiency. Each of open and distance learning. However, the GPT-4 has some
the other 17 themes has one frequency. limitations. The first is that it is paid. The fact that GPT-4 is
available for a fee as of April 2023 can also be considered one
By examining the themes in the Workarounds and User of the limitations in the use of AutoGPT.
Experience categories, it is possible to gain a more detailed
understanding of the user experience. There is significant VII. CONCLUSIONS AND SUGGESTIONS
interest and curiosity about AI technology among users. Users This qualitative research is an exploratory case study that
are exploring the potential of AI to learn new technologies and aims to examine use-cases and experiences of Auto-GPT
expressing their reliance on AI technologies in the process. users. Data were collected from the GitHub platform. Answers
The results indicate that users are turning to learning of users to the question of “What are your experiences on the
programming languages such as Python. This will increase Auto-GPT use-cases*”. In response to this question, 16
their ability to use these tools more effectively and deepen comments were analyzed and 21 themes were identified. The
their understanding of these technologies. Users appreciate three important findings of this study are as follows.
Auto-GPT's ability to delegate tasks to other AI agents. This
allows users to focus their time and energy on more strategic • Integration of Auto-GPT into business processes:
or creative tasks, while ensuring that routine or repetitive tasks The results indicate that integrating Auto-GPT into business
are completed quickly and efficiently. Autonomous AI processes can provide significant productivity gains,
capabilities, such as conducting research and writing reports, especially in data-driven and information management jobs.
are among the most frequently cited positive aspects by users. This means that companies and organizations can use Auto-
This demonstrates AI's ability to respond to user needs and GPT to optimize their business processes and become more
add value. competitive.
In addition to the advantages of AutoGPT, some • Auto-GPT's limitations and areas for improvement:
limitations were also mentioned. These were identified as This research also reveals Auto-GPT's current limitations and
"token limit", "inefficiency", "forgetting generated tools", and areas for improvement. Limitations such as the token limit,
"iteration". When analyzing the codes related to these topics, inefficiency, forgetting of generated tools, and iteration have

4
 Journal of Emerging Computer Technologies
Fırat and Kuleli
been identified as important areas for the development and [8] Lund, B. D., & Wang, T. (2023). Chatting about ChatGPT: how may
AI and GPT impact academia and libraries?. Library Hi Tech News,
improvement of Auto-GPT. Users' suggestions for 40(3), 26-29.
overcoming these limitations focused on session continuity, [9] Lo, C. K. (2023). What is the impact of ChatGPT on education? A rapid
removing the token limit, and further development of the review of the literature. Education Sciences, 13(4), 410.
GPT-4 model. [10] OpenAI, (2023). Chat GPT. Retrieved from
https://openai.com/blog/chatgpt/ on 2 January 2023.
• Potential for use in open and distance learning: The [11] Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal,
results of various use cases and experiences of Auto-GPT P., ... & Amodei, D. (2020). Language models are few-shot learners.
show that it can offer significant opportunities for open and Advances in neural information processing systems, 33, 1877-1901.
distance learning. The effective use of Auto-GPT in areas such [12] Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I.
as content creation, support services, assessment, and (2019). Language models are unsupervised multitask learners. OpenAI
blog, 1(8), 9.
evaluation can contribute to making such learning systems
[13] Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal,
more efficient and user friendly [44]. P., ... & Amodei, D. (2020). Language models are few-shot learners.
Advances in neural information processing systems, 33, 1877-1901.
The results of this study show that Auto-GPT can offer
[14] Firat, M. (2023). How chat GPT can transform autodidactic
successful use cases in areas such as content creation, data experiences and open education. Department of Distance Education,
summarization, knowledge organization, and AAI Open Education Faculty, Anadolu Unive.
capabilities. In addition, it has been determined that users' [15] Koubaa, A. (2023). GPT-4 vs. GPT-3.5: A concise showdown.
experiences such as learning programs, using autonomous TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.22312330.v1
applications, conducting research, and writing reports stand [16] Firat, M. (2023). How chat GPT can transform autodidactic
out in their experiences. The codes and themes derived from experiences and open education. Department of Distance Education,
Open Education Faculty, Anadolu University.
participant opinions about Auto-GPT indicate that this
[17] Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D.,
application has high usage potential in open and distance Fischer, F., ... & Kasneci, G. (2023). ChatGPT for good? On
learning, especially in mass teaching systems. However, both opportunities and challenges of large language models for education.
the lack of desired maturity in AI technologies and the fact Learning and Individual Differences, 103, 102274..
that Auto-GPT is a newly emerged tool that has not yet [18] Talan, T., & Kalınkara, Y. (2023). The role of artificial intelligence in
higher education: ChatGPT assessment for anatomy course.
evolved into a stable application pose challenges for its usage Uluslararası Yönetim Bilişim Sistemleri ve Bilgisayar Bilimleri
in sensitive areas such as education. Dergisi, 7(1), 33-40.
VIII. RECOMMENDATIONS [19] de Winter, J. C. (2023). Can ChatGPT pass high school exams on
English language comprehension. Researchgate. Preprint.
Continue the development of Auto-GPT and overcome [20] Susnjak, T. (2022). ChatGPT: The end of online exam integrity?. arXiv
existing limitations. In particular, issues such as token limits preprint arXiv:2212.09292.
and inefficiency must be addressed. [21] Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D.,
Fischer, F., ... & Kasneci, G. (2023). ChatGPT for Good? On
The use of Auto-GPT in the field of education and learning Opportunities and Challenges of Large Language Models for
Education. 2023-01-15 A Position Paper.
technologies should be encouraged, especially considering https://www.edu.sot.tum.de/fileadmin/w00bed/hctl/_my_direct_uploa
that it can be used effectively in open- and distance-learning ds/ChatGPT_for_Good_.pdf
processes. [22] White, J., Hays, S., Fu, Q., Spencer-Smith, J., & Schmidt, D. C. (2023).
Chatgpt prompt patterns for improving code quality, refactoring,
Platforms should be created to share experiences and use requirements elicitation, and software design. arXiv preprint
cases related to Auto-GPT among users, and users should be arXiv:2303.07839.
able to interact on these platforms. In this way, knowledge [23] Radanliev, P., & De Roure, D. (2023). Review of the state of the art in
sharing regarding the potential applications and effective autonomous artificial intelligence. AI and Ethics, 3(2), 497-504.
usage methods of Auto-GPT can be increased. [24] Guo, T. (2023). Auto Self-Correct ChatGPT. TechRxiv. Preprint.
https://doi.org/10.36227/techrxiv.22688317.v1
REFERENCES [25] Gizzi, E., Nair, L., Chernova, S., & Sinapov, J. (2022). Creative
problem solving in artificially intelligent agents: A survey and
[1] Galassi, A., Lippi, M., & Torroni, P. (2020). Attention in natural framework. Journal of Artificial Intelligence Research, 75, 857-911.
language processing. IEEE transactions on neural networks and
learning systems, 32(10), 4291-4308. [26] Wang, Y. (2023). Autonomous AI (AAI) and Symbiotic Human-
Machine Intelligence Systems. In Plenary talk, IEEE International
[2] Bahja, M. (2020). Natural language processing applications in Conference on Systems, Man, and Cybernetics (pp. 3-1).
business. E-Business-higher education and intelligence applications.
[27] Totschnig, W. (2020). Fully autonomous AI. Science and Engineering
[3] Lu, P., Peng, B., Cheng, H., Galley, M., Chang, K. W., Wu, Y. N., ... Ethics, 26, 2473-2485.
& Gao, J. (2023). Chameleon: Plug-and-play compositional reasoning
with large language models. arXiv preprint arXiv:2304.09842. [28] Yang, H., Yue, S., & He, Y. (2023). Auto-GPT for Online Decision
Making: Benchmarks and Additional Opinions. arXiv preprint
[4] Qin, Y., Hu, S., Lin, Y., Chen, W., Ding, N., Cui, G., ... & Sun, M. arXiv:2306.02224.
(2023). Tool learning with foundation models. arXiv preprint
arXiv:2304.08354. [29] Thodberg, H. H., Thodberg, B., Ahlkvist, J., & Offiah, A. C. (2022).
Autonomous artificial intelligence in pediatric radiology: the use and
[5] Palace, V. M. (2019). What if artificial intelligence wrote this: artificial perception of BoneXpert for bone age assessment. Pediatric Radiology,
intelligence and copyright law. Fla. L. Rev., 71, 217. 52(7), 1338-1346.
[6] Yang, H., Yue, S., & He, Y. (2023). Auto-GPT for Online Decision [30] Noye, S., Martinez, R. M., Carnieletto, L., De Carli, M., & Aguirre, A.
Making: Benchmarks and Additional Opinions. arXiv preprint C. (2022). A review of advanced ground source heat pump control:
arXiv:2306.02224. Artificial intelligence for autonomous and adaptive control. Renewable
[7] Rahimi, F., & Abadi, A. T. B. (2023). ChatGPT and publication ethics. and Sustainable Energy Reviews, 153, 111685.
Archives of medical research, 54(3), 272-274. [31] Omelianenko, I. (2022). Autonomous Artificial Intelligent Agents.
Machine Learning and the City: Applications in Architecture and
Urban Design, 263-285.

5
 Journal of Emerging Computer Technologies
Fırat and Kuleli
[32] Firat, M. (2023). How chat GPT can transform autodidactic [40] Vaismoradi, M., Hannele T., Terese B. (2013) Content analysis and
experiences and open education. Department of Distance Education, thematic analysis: Implications for conducting a qualitative descriptive
Open Education Faculty, Anadolu Unive. study. Nursing & health sciences, 15(3), 398-405.
[33] Yin, R. K. (2009). Case study research: Design and methods (Vol. 5). [41] Thodberg, H.H., Thodberg, B., Ahlkvist, J. et al. (2022) Autonomous
sage. artificial intelligence in pediatric radiology: the use and perception of
[34] Creswell, J. W., & Poth, C. N. (2016). Qualitative inquiry and research BoneXpert for bone age assessment. Pediatr Radiol 52(7), 1338–1346.
design: Choosing among five approaches. Sage publications. https://doi.org/10.1007/s00247-022-05295-w.
[35] Chmiliar, I. (2010). Multiple-case designs. J. Mills, G. Eurepas & E. [42] Noye, S., Martinez, R. M., Carnieletto, L., De Carli, M., & Aguirre, A.
Wiebe (Eds.), In: Encyclopedia of case study research (pp. 582-584). C. (2022). A review of advanced ground source heat pump control:
Artificial intelligence for autonomous and adaptive control. Renewable
[36] Merriam, S. B. (2013). Nitel araştırma: Desen ve uygulama için bir
and Sustainable Energy Reviews, 153, 111685.
rehber. Çev.Ed. Sellahattin Turan. Ankara: Nobel Yayın Dağıtım.
[43] Omelianenko, I. (2022). Autonomous Artificial Intelligent Agents.
[37] Yin, R. K. (1984) Case study research.
Machine Learning and the City: Applications in Architecture and
[38] Yin, R. K. (2009) Case study research: Design and methods. Vol. 5. Urban Design, 263-285.
Sage. [44] Firat, M. (2023). What ChatGPT means for universities: Perceptions of
[39] Swedberg, R. (2020) Exploratory research. The production of scholars and students. Journal of Applied Learning and Teaching, 6(1).
knowledge: Enhancing progress in social science, 17-41.

View publication stats