Generative Artificial Intelligence: Reviewing developments

In an article published in the June 2023 issue of the newsletter of the Caribbean Centre for Educational Planning, titled ” Chat GPT – Incremental or Exponential Revolution in Artificial Intelligence” I incipiently argue that the seeming awe with which Chat GPT was being greeted as some amazing new stuff, reflected a misreading of the evolution of artificial intelligence.  (The title of the article should have read “Chat GPT – Incremental Evolution or Exponential Revolution in Artificial Intelligence”.

My contention, in essence, was that what we now know as Generative AI is not new in substance but is in fact a new generation of an ancient application.  In my ongoing engagement with the discussion on this issue, I have seen corroboration of my position.  In a well-reasoned article in the November 19, 2024, issue of Harvard Business Review, Daily Alert, three scholars from the Rotman School of Management, at the University of Toronto, make the case.  They are Ajay Agrawal, Chair in Entrepreneurship and Innovation, Joshua Gans Chair in Technical Innovation and Entrepreneurship, and Avi Goldfarb, Chair in Artificial Intelligence and Healthcare.

Their article, which is titled, “Generative AI is still just a Prediction Machine” has two main foundations:

• AI draws on available data to do predictions, and its evolution is no different from evolutions of computers. They assert that:

“Computers are arithmetic machines; in that they take binary numbers and perform operations on them to execute complex tasks. Advances in computing can be understood as the cost of arithmetic falling”.

AI draws on available data to do predictions, and its evolution is no different from evolutions of computers.
They assert that:
“Computers are arithmetic machines; in that they take binary numbers and perform operations on them to execute complex tasks. Advances in computing can be understood as the cost of arithmetic falling”.

They illustrate this by noting that:

“As the price of machine arithmetic fell from the 1970s to the 1990s, it became clear that games, music, mail, and photography would be solved with arithmetic. At its peak in the 1990s, Kodak enabled photography primarily through chemical engineering, but when machine arithmetic became cheap enough, it was possible to reframe photography as an arithmetic problem”.

• Generative AI generates new content drawing on data but also manufactures its own “false” data.  Reliance on AI generated content, where false or fictional data are used can have massively serious implications for decision-making and can pose serious risks.

Caution & Cross-checking is the gold standard

As Generative AI becomes more dominant in the research and decision-making spaces (of education and the economy), the core skills that will be needed are the patience to cross-check and assess and the discipline to reengage the “old fashioned” (!) form of knowledge creation and parameters for decision-making which rely on comprehensive problematization of issues, the development of reliable and valid protocols for data collection, the review of prior knowledge, and the careful extraction of insights and knowledge from findings.

As Generative AI becomes more dominant in the research and decision-making spaces (of education and the economy), the core skills that will be needed are the patience to cross-check and assess and the discipline to reengage the “old fashioned” (!) form of knowledge creation and parameters for decision-making which rely on comprehensive problematization of issues, the development of reliable and valid protocols for data collection, the review of prior knowledge, and the careful extraction of insights and knowledge from findings.

I predict that there will be many versions of Generative AI over the next eighteen to thirty-six months, some of which will be seen as revolutionary. But I submit that a careful examination of each form will lead to the conclusion that “we have been here before”.

Against this background I share the unedited contents of the June 2023 article.

Chat GPT – Incremental or Exponential Revolution in Artificial Intelligence

Universities and schools are now scrambling to make sense of the latest developments in the field of artificial intelligence (AI), as reflected in what is known as Generative Pre-trained Transformer or Chat GPT.  Chat GPT is an AI-based model which reflects advances in technology based on its capacity to generate human-like conversations. Its applications are being used in customer service, sales and marketing support, human resources, law and government.

Chat GPT is an AI-based model which reflects advances in technology based on its capacity to generate human-like conversations. Its applications are being used in customer service, sales and marketing support, human resources, law and government.

The general sentiment appears to be that Chat GPT is revolutionary and exceptional in its capacities and applications. But I see it differently.  There is also the view in academia that Chat GPT poses a level of risk to the integrity of examination processes such that greater efforts will have to be made to prevent cheating.

But is Chat GPT really that revolutionary? Is it anything more than an incremental change in the landscape of artificial intelligence?

My interactions with artificial intelligence over forty years

In the mid 1980’s when I was studying in the United States of America, neither mobile phones nor smartphones existed, the main forms of commercial telecommunication were fixed landlines services or limited band small group services called CB radio. (Of course, the military community had secure communication mechanisms in the form of what is now the internet, which became commercial in the late 1990s).

During the period of my studies, being away from family, I would often get homesick. To ease my discomfort, I would use the payphone in the corridor of the dormitory and attempt to make a call to Jamaica, just to hear the recorded response which said, “we are sorry we cannot place your call at this time…this is a recording from Kingston, Jamaica”.  While just hearing the words “Kingston, Jamaica” was the reason for my placing a call I knew would not go through was my objective, the larger point is that some basic form of telephony automation existed then.

But it was more advanced than a mere recording that chipped in, in response to a prompt.  In making a collect call, I would interact with an “intelligent non-human operator”. After dialling the designated number to place the collect call, I would be asked to state my name.  The name decoding software programmed into the “intelligent operator” was not capable of decoding my pronunciation of my name and after three or four tries, with the “operator” saying, “I did not hear that” or “Repeat your name, please” or “Did you say Haanutee” or some other ‘corrupted’, if I said “yes” then the call would be placed.  At the other end, my family member would hear a record that says “This is the operator with a collect call from…. will you accept the charges?” This was in the 1980s.

In the early 1990s I began working at the then, Jamaica Telephone Company, later named Telecommunications of Jamaica, then Cable & Wireless and now Flow.  Based on my work as the Company’s Counsellor, I was actively engaged in seeking to prepare the workforce for changes in the ‘world of work’ as a result of the emergence of new technologies. One of the company’s largest departments was Operator Services, and it was that department which faced the immediate greatest threat from AI when the “intelligent operator” technology was to be deployed by the company.  As was the case when I would try to make a collect call from the United States in the 1980s, the technology that the Jamaica Telephone Company would be deploying was interactive.

With the advent of smartphones, in the last two decades, if I want some information I could press a button, or type something in a search engine, and ask a question. Within seconds, I would see highly relevant and reliable information appear on my screen, and in many cases hundreds and sometimes thousands of options.

Chat GPT: Pathbreaker or a step further

A brief summary of the characteristics of Chat GPT, drawn from various open internet sources show that it can translate language, recognize images, mimic speech patterns, provide information in a format which appears reasoned (called deep learning), like a human being would present, and has a large vocabulary.

But how does this differ from earlier forms of AI?  My laptop which I had long before Chat GPT existed had a software that translated essays and articles from one language to another. I can go into Google and ask for a foreign language for almost any word and can ask for the pronunciation as well. The vocabulary capabilities of both my smartphone and laptop are enormous. I can search for and customize images on my smartphone.  My smartphone and my latest laptop both have predictive texts and store my written patterns and use of some words and phrases.

Internet sources also indicate that Chat GPT can understand poorly written communication and identify grammatical and spelling errors and can suggest structuring to improve the quality of written communication.  My laptops have had those features and capabilities for years. 

While the deep learning capabilities of Chat GPT, whereby it can construct an essay from scratch, represents an advancement over what previously exists where search engines gather already existing information, the fact is that many of those with pre-set information were used extensively by researchers and students writing term papers.

Based on the foregoing, it is my contention that Chat GPT rather than being a revolutionary development in AI, represents an incremental improvement which pulls on existing AI capabilities which go back over forty years and concentrates on capabilities developed in the last ten or so years.  In this regard I see Chat GPT, not as a pathbreaker, but merely as a step further a long road of AI inventions.

Based on the foregoing, it is my contention that Chat GPT rather than being a revolutionary development in AI, represents an incremental improvement which pulls on existing AI capabilities which go back over forty years and concentrates on capabilities developed in the last ten or so years.  In this regard I see Chat GPT, not as a pathbreaker, but merely as a step further a long road of AI inventions.

Chat GPT and the academy

Universities, colleges, and schools are expressing worry that with Chat GPT the phenomenon of students cheating will likely increase and will be harder to detect.  If my argument is plausible, that Chat GPT does not represent an exponential advancement on earlier AI technologies, then the same cures used previously to detect plagiarism and outsourced work, can be used with Chat GPT, but perhaps with greater controls.  From before the advent of Chat GPT, there was the need for new forms of assessment which force students to think critically.  Oral examinations may now need to become more widespread with students required to think on their feet in answer to previously unseen questions, as well as be subject to interrogation in the presence of peers and a panel of examiners responding to questions related to written work submitted.   In addition, the academy has been using solutions such as Turnitin for about a decade and a half to evaluate the extent to which students’ works are original.  The deep learning capabilities of Chat GPT are in my view not radically advanced over the capability to search engines to assemble information from multiple sources.

The way some people are talking about Chat GPT makes it appear as though a massive uncontrollable phenomenon has landed in the academy placing us at the mercy of the student who cheats. I do not share that anxiety.

References

Agrawal,A.,  Gans, J., and  Goldfarb, A. (2024, November 18). Generative AI Is Still Just a Prediction Machine. Harvard Business Review.  Retrieved from https://hbr.org/2024/11/generative-ai-is-still-just-a-prediction-machine?utm_medium=email&utm_source=newsletter_daily&utm_campaign=dailyalert_&deliveryName=NL_DailyAlert_20241119

Fedewa, J. (2024, May 7). What Is ChatGPT and How Does It Work? How-To Geek. Retrieved from https://www.howtogeek.com/871071/what-is-chatgpt/

Zewe, A. (2023, November 9). Explained: Generative AI – How do powerful generative AI systems likeChatGPT work, and what makes them different from other types of artificial intelligence?MIT News Retrieved from https://lids.mit.edu/news-and-events/news/explained-generative-ai-how-do-powerful-generative-ai-systems-chatgpt-work-and#:~:text=The%20base%20models%20underlying%20ChatGPT%20and%20similar%20systems,larger%20and%20more%20complex%2C%20with%20billions%20of%20parameters.

---

Canute Thompson is Professor of Educational Policy, Planning and Leadership, Pro Vice-Chancellor – Undergraduate Studies and Director of the Caribbean Centre for Educational Planning at The University of the West Indies, Mona Campus, a social activist, and author of eight books and twenty journal articles.

Professor Thompson has earned several awards. Among them, are eight UWI Mona Campus Principal’s Award – two for Best Publication (Article Category) in 2019 and in 2020 for his book, ‘Reimagining Educational Leadership in the Caribbean’; three for Most Outstanding Researcher (2020, 2021, and 2024); two in 2023 on behalf of the CCEP –  for Research Activity generating the most funds and Research with the most Development Impact, and one in 2024 for Research Activity generating the most funds.  In 2022 he was awarded a bronze medal in the Independent Publishers’ Book Awards, for his 2020 book, Education and Development: Policy Imperatives for Jamaica and the Caribbean.

---