Legend has it that the wise Anand (a name suggested by ChatGPT, whom I call Skippy ^2) helped King Devendra (a name also proposed by Skippy) solve a difficult problem. In gratitude Devendra offered Anand the payment he desired. Anand asked that the king give him a grain of rice for the first square of a chessboard and that he double the amount in each successive square. The king accepted, but after a good while the king’s mathematicians informed him that there was not enough wheat in the entire kingdom to pay what Anand asked for.
Gordon Moore, one of the founders of Intel, observed in 1965 that the capacity to integrate transistors on a chip doubles every 18 months. That is to say, following an exponential progression. This observation became a predictor of industry behavior that has held true until Moore’s death in early 2023.
It seems impossible that something doubles 64 times, but since Moore’s observation, processor power has doubled 43 times. But there’s more: 20 years ago inventor and futurist Ray Kurzweil calculated the computing capacity one could acquire for 1000 dollars adjusted for inflation throughout the 20th century. The result indicated exponential growth perhaps starting before the beginning of the 20th century, if we consider the work of Blaise Pascal, Charles Babbage and Ada de Lovelace at the beginning of the 19th century. What we can say is that during the 20th century the computing capacity we can buy at a given price has doubled, regardless of the technology used, 66 times during the 20th century. More than squares on a chessboard. And this behavior has not stopped in the 23 years we’ve been into the 21st century.
Both the wise Anand and King Devendra would be dumbfounded. But the question the wise Anand would ask is: what happens when something doubles so many times and what happens when this affects so many things?
A possible answer is given by Peter Diamandis, Greek-American entrepreneur and futurist, who proposes in his book “Abundance: The Future Is Better Than You Think” (^1) that when something is digitized it becomes connected with this exponential improvement. This at first is not a big deal (it’s deceptive) because at first exponential curves are quite flat, but suddenly they make a kink upwards and grow suddenly. It is then that digitization becomes disruptive. And this disruption implies changes in the way things are done, in the artifacts we use (which become smaller or disappear), in the jobs we people have to do - the disappearance of some and the creation of others -, and in the value of things: some things cease to have value because they become abundant and other things become very valuable.
You can ask Skippy about Diamandis’s 6 D’s and he will explain it to you very well.
Another possible answer was given by Alvin Toffler, in an essay in the prestigious magazine Playboy - which is not Open Access nor JCR, but surely you’ve heard of it - in 1970 and subsequently in the book of the same title: “Future Shock” ^3. The shock of the future. Toffler makes the observation that the transformations caused by technological innovation - already in 1970 - are each time deeper and each succeed each other faster, they accelerate. So that the world changes radically several times during a person’s lifetime. And this can leave us perplexed, in the shock of the future.
If Toffler already observed the shock of the future 53 years ago, now it is something evident in our day to day. Let’s think that 25 years ago there was no web - it was there, but there were only 4 nerds -, 15 years ago there were no mobile phones connected to the internet - they were there, but we only had 4 nerds -, 6 months ago there were no generative models that seem like Artificial Intelligences. Well, they were there, but only 4 nerds were playing with them.
On November 30, 2022 ChatGPT was launched, another Chatbot on the internet like many others we had seen, which allowed talking to the generative model GPT-3 (Generative Pretrained Transformer) developed by the non-profit organization OpenAI (who have reconsidered and now they do want to make profits). In 5 days ChatGPT reached 5 million users, in two months it had reached 100 million users. ChatGPT can already enter the Guinness Book of Records as the fastest expanding product to date. In January 2023 ChatGPT was as popular in Google searches as Shakira, now - in May 2023 - it is difficult to find a more popular term anywhere.
For some reason ChatGPT is good enough that all of us who use it see that this is something different. A Chatbot worth paying attention to. Just as in their day internet users who tried the new Google search engine realized it was something different; ten times better than anything else, Google’s creators like to say.
ChatGPT marks the beginning of the kink in the progression of a new tool: generative models based on neural networks, a technology in the research field of artificial intelligence (AI) that has been running for almost 70 years. Let’s not forget that, as John McCarthy - one of the fathers of the discipline - said, when a technology works we no longer call it AI, but something else. AI’s goals move further beyond, perhaps now they will aim at the unachieved goal of Artificial General Intelligences (AGI) equivalent to human intelligence. But don’t worry, we are not there yet, nor will we be soon.
However, generative models have now become a technology good enough to become the basis of services for the general public. Computer systems now understand the language with which we humans communicate - any language or almost any - they begin to understand context and cultural subtleties, and are capable of responding giving information often accurate. More and more accurate.
Lately, I have seen archive photographs showing mathematics professors demonstrating in the street protesting against the use of calculators in the 70s. Denial or prohibition is one of the first responses we can expect, even if they serve little.
Calculators were and are fantastic, they allow doing arithmetic calculations, accessing logarithm and trigonometric tables, and even executing algorithms. With a calculator we are capable of solving very complex problems, quickly and with precision. Surely problems we could not address without. But the availability of calculators can cause us not to learn to do these operations without them. Or that we forget how to do them. “Use it or lose it” they say.
ChatGPT and the large family of generative models that is coming - go to huggingface.co where you will find the main learning community about generative models and transformers and you will see what’s cooking there - can do many things for us: Help us translate texts, summarize documents, prepare presentations, review grammar and style, answer questions we would have previously asked an encyclopedia, an expert or the internet search engine… And even more, if we are lucky and have a knack for giving instructions (what they call “prompt”, but they are “instructions”) they can help solve problems and connect with other programs to achieve even more things. They can generate code for our programs, help us improve style, find errors and document them. And many other things. Which like calculators will increase our capabilities, but we will have to be careful because using these tools we can lose the ability to do things without them.
Generative AI Models have a very interesting property: sometimes they hallucinate. In the context of AI, hallucinations are strange results - erroneous, for sure - that these systems give sometimes, in an unexpected and unpredictable way. So that whenever we use one of these models we have to think that maybe the result is accurate or maybe it is a hallucination.
When a calculator gives a result, do we have the capacity to check if the result is correct? Certainly, we assume the risk and responsibility of accepting the result of a calculator - or a spreadsheet or SAP - every time we use them for our day to day. But calculators are reliable, they are built with deterministic systems that always give the same result. And if they fail, we can look for where the error is and correct the system.
With neural network models this is not the case. We know they work phenomenologically: they work because they work and we observe a behavior. Nothing guarantees us that they will continue to work like this or that they will hallucinate or will explain to us why they work as they do. Think about this when, very soon, more and more systems are controlled by these models.
As teachers and educators the dilemma arises for us. Should we use these tools or should we not use them in our subjects? And if we have to use them: how should we use them? Regardless of what we decide our students already have these tools, they know who Shakira is and also what ChatGPT is.
Expert players of the real-time strategy video game StarCraft explain that to win one must play in Macro and in Micro. One must have a Macro strategy: master the whole map, have a strategy to win the match in the long term. But one must apply tactics and operations in Micro, pay attention and move your units where there are confrontations. And know when you have to pay attention to Macro or to Micro. Without mastering Macro-Micro one cannot win at StarCraft.
In the same way each subject has a Macro context: What are the objectives of the subject? The professionals we train, will they use generative AI tools in their day to day? How does this affect our objectives? And how will it affect in the coming months and years, given the shocking pace of technological advancement?
But there is also a Micro context: How do we teach? Do we ban AI in the classroom or assume that at any moment our students may have the ChatGPT window open or some other tool? And the projects? And the exercises they have to do at home? How do we evaluate them knowing they have AI tools that are impossible or very difficult to detect?
Every teacher, every team of teachers in each subject has the opportunity and the challenge to rethink our strategy and teaching practice. And we cannot ignore it.
^1 (Diamandis, P. H., & Kotler, S. (2012). Abundance: The Future Is Better Than You Think. Free Press." ^2 Skippy is the name of the alien AI that co-stars in the science fiction novel series Expeditionary Force by writer Craig Alanson. ^3 Toffler, Alvin. (1970). Future Shock. New York: Random House.