The AI turn in education

 

An AI turn

In conversations about education these days, something about ChatGPT must be mentioned. Otherwise, the speaker appears out-of-date. Like the MOOCs a decade ago, everyone seems to have an opinion about it. Everyone thinks that this is going to change education beyond recognition.

One can guess what happens next. This is what we saw with MOOCs; It was changing all education; shortly after that, it was not. Initially, everyone wanted to show off that they have heard about Coursera; in a few months, everyone wanted to sound smart by expressing concerns about the MOOC's low completion rates. There were even reasonable-sounding debates about whether completion is the right parameter to look at in the MOOC world. However, all this is ancient history. Today, if anyone brings up MOOCs, the eyes will roll and dropping pins will be heard. Its moment has come to pass!

ChatGPT brings us to a similar moment, though it was adopted faster and therefore, opinions about it are more numerous. There is always a sense of wonder in our first contact with ChatGPT: How did it do what it did? In a way, this is partly us underestimating digital information (how much is already out there) and partly us being lenient with our expectations from a machine (some of the mistakes it makes will not be accepted if they came from a human being). If Rip Van Winkle woke up today, he might have had a similar reaction to a Google search. But it doesn't take long to know what's missing.

It is likely that we are already at the pivot point of the technology hype cycle, when all the big predictions have already been made and the reality is now catching up. 'AI hallucination' is now official, and the large language models that are hugely resource-intensive and impractically expensive to train are well understood. That these models will be hard to train on new information is obvious and we know that LLMs are not going to take over the world any time soon. The question of AI 'common sense' is being asked, with no easy answers at hand. We do indeed overestimate the impact of technologies in the short run!

However, I am not writing this post to state the obvious - that the large language models are overrated - but rather to think of their long-term impact, on education in particular. This is the other half of the insight into technology impact: We tend to underestimate their impact over the long term! I would like to claim that we are at a point of an AI turn. The LLMs and the resulting interest in AI will have a profound impact, even if these are not what we think they would be.

AI is political

But before I get into what the AI turn may bring, it is required to state the obvious:That every conversation about AI is political.

There is global excitement (and fear) about robots replacing people. This is presented as inevitable and regular reports are being churned out with ever newer estimates of how grim the situation will be. But there is nothing obvious about these developments. These are choices being made - to fund the development of labour-saving technologies, to confront the collective bargaining attempts to stop their implementation or to override ethical concerns about their unregulated use. The investment to build a large language model equals the money required to run several public school systems with well-paid teachers and well-fed students, and it is a choice the society makes, through its market-information system, that it is optimal to train a LLM than to spend the money on these schools. A lot goes into this: The relative political power of the beneficiaries, time horizons of decision-making and indeed, values! We don't speak about these factors and assume that the technologies such as AI are developing autonomously - a key reason why we fail to appreciate their trajectories.

Consider the present moment as an example. Our talk is far exceeding the real capabilities of AI, and its development is considered fâit accompli - nothing can be done to stop 'progress'! There are some spectacular attempts at automation - in car driving or coffee making - but they are still outliers. In most cases, the impact of such technologies remain at best marginal; a novel toy, but yet to propagate seriously outside the work of the laptop class! So, why invoke the 'AI spectre' when we are not yet there? This is because we are living the 1990s in the reverse: The private sector manufacturing jobs that were shipped out in that decade are now coming home. The gale of globalization experienced then has run out of strength, and value chains are now shorter. The big problem across the developed world now is the labour shortage. The political creed today is not libertinism but populism. At this time, the spectre of AI can keep the bargaining power of labour in its place. Teachers are starting to demand better pay - what is better than to talk about ChatGPT making them redundant!

The unintended consequence

But ChatGPT and enhanced language capabilities of these tools will change education, even if they don't make the teachers redundant. These developments are highlighting, yet again, that the content is just an enabler, but not education in itself.

This is an interesting parallel with MOOCs. I remember meeting an entrepreneur at the MOOC moment, when it was the key topic at every education conference. He confidently told me that the MOOCs will fail. The reason: Content doesn't equal education! He reasoned that one can have the best content, but education, at its core, is an humanistic endeavour. It is an art of connection, conversation, context. Educational success comes from application. For these reasons, the best professors at Harvard or Yale would have limited impact on the boys and girls of Manila and Kigali, if they are to educate them through impersonal, remote videos.

ChatGPT transforms the role of content, for sure, but by minimising the effort for the dissemination and production of content, it shows that the point of education is different. The teacher's job is more than just producing notes, we should now see, but to orchestrate learning among a diverse group of individuals, with different ambitions, aspirations and pre-occupations. It is a tool, still a primitive one, which can do an excellent job looking at past information, but have limited capability to process new information and ideas.

I shall argue that this is a good place to start assessing the AI turn in education: Not redundant teachers, but new ways of thinking about curriculum, delivery and assessment. The real shift is from a mindset of content scarcity (which necessitated content mastery) to embracing content abundance, something that will demand a total transformation of the apparatus we use: Classrooms, textbooks and individualised assessments! This is a moment to start seeing the individuals in their group context, curriculum in terms of not required texts but conversations about problems to solve and focusing all assessments to the assessment of the humanistic capabilities.