AI in Education

John Traxler, Professor of Digital Learning, University of Wolverhampton

John Traxler, Professor of Digital Learning, University of Wolverhampton

A lot has been written about AI recently, and AI has written a lot of it. I wrote this piece, and it reflects my attempts to catch up with the recent and very visible eruption of AI, meaning chatGPT from OpenAI, as well as its current and emerging competitors, Bard from Google and Bing from Microsoft, into public, professional and educational spaces. This current eruption has probably taken me by surprise. As someone working with computers for the past forty years, I had become slightly jaded, being told annually that AI was the next big thing and now, suddenly, it does really seem to be the next big thing.

Given that many readers might already be familiar with both the facts and the fantasy around AI and, specifically, chatGPT, I wondered, as an educationalist, what might be the pedagogic possibilities, especially the innovative pedagogic possibilities, hidden amongst the technological complexity.

Looking at the conversations with chatGPT, and perhaps to state the obvious, the chatGPT system seems stateless; it has no recollection of previous conversations or queries and no knowledge of any other users. It lives in the present and once a conversation is finished, it is gone, never to be recalled. Also, chapGPT is reactive, not proactive, it will respond and it will answer, but it will never initiate or follow up. You can, of course, ask chatGPT to take the initiative and set you a task or ask you a question. That seems to be telling me chatGPT will only ever be the teacher’s assistant, never the teacher.

Sadly, reading the traffic of suggestions about teaching with AI, it could end up as a fairly conservative teacher's assistant, being asked to deal with lesson plans, content, assessments and with feedback uncritically and implicitly based on a largely transmissive and didactic approach to teaching and learning.

So, suppose we are to see the pedagogic applications of AI catch up with the more progressive ideas out there in schools, colleges and universities. In that case, we need to look at approaches such as constructivism, building on each learner's previous understandings, such as social constructivism, building on learner's ability to learn from each other through discussion, and such as connective, building on massive and connected communities of learners interacting with each other.

Or perhaps we need to look at whether the pedagogies of AI can appropriate and adapt tactics such as flipped learning, open learning, blogging, citizen science and project-based learning, amongst the various tactics designed to empower and validate learners and their experiences, rather than just pushing out content and curricular.

"What is wrong today will, however, probably be true tomorrow, and we may all be running to keep up with developments that increasingly have a life of their own"

Sadly, before we can do any of this, we must deal with facts that AI chatbots can evidently fantasize, hallucinate, misinform, or go wildly off-course, and so clearly, getting the answers you need requires crafting the question you ask with considerable care. The literature is already beginning to outline what some have called prompt engineering, the skill of phrasing the queries in order to get precisely and efficiently the answer required. It turns out that this is not easy and is likely to be an art as well as a science.

In the meantime, international agencies such as UNESCO are encouraging ministries, corporations, and institutions to think about ethical and legal issues. Alongside the predictable and recurrent concerns about plagiarism, that are concerned about harm, liability, and responsibility, and about the transformation of our ideas about authorship, originality, and creativity, and about how we regulate and legislate around all of these.

At a deeper level, there must be concerns about the ways in which AI systems are trained and informed, namely from the world of digital resources with all their flaws, prejudices, biases, and fabrications, representing and reproducing the values, languages, resources, and structures of specific minorities, namely the countries, cultures and corporations with the most power. On current evidence, chaptGPT will probably be better informed about Middle Earth than about Central Africa and happy to discuss it in American English but not in Kikuyu and not discuss it with anyone actually in Central Africa owing to the problems of infrastructure and access.

There will always be concerns about consciousness, about the point at which AI can be understood to be actually aware and thinking and presumably, these concerns will only increase as the digital technology of AI merges into the digital technology of robotics and the interfaces become ever more conversational.

These are some of my concerns and given my level of expertise and experience, I’d happily be proven wrong. What is wrong today will, however probably be true tomorrow, and we may all be running to keep up with developments that increasingly have a life of their own.