What Price Education?
Annual global expenditure on education is estimated to be of the order of $US5 trillion. The vast majority of that expenditure goes on teaching costs and the majority of that on salaries and on-costs such as pensions. There are international agreements that attempt to ensure that all countries in the world will spend at least 4% of their GDP on education by 2030, but at present fewer than 50% meet that target. In addition to government expenditure there are private/independent schools, some of which charge fees in excess of $60,000 per annum. Education is an essential human right, but it is also big business and, for some at least, very profitable, with returns on investment being reckoned in some cases as high as 30% per annum in established schools.
The short answer to the question "What Price Education?" would therefore seem to be "a very high price" because both in terms of government and family expenditure - directly in private schooling and indirectly through taxation - education absorbs a very considerable proportion of all available human economic resources.
Are things about to change? Throughout the pandemic we have seen necessity in her role as mother of invention force schools, teachers and pupils into unfamiliar territory. Online learning has been used in most countries to supplement classroom time lost because of the need to self-isolate, either because of infection or as a prophylactic measure. Much of that experience was distinctly inferior because it denied first-hand or face-to-face learning to pupils and teachers, and eliminated the collective personal learning that extends far beyond "academic" learning that takes place in the cut and thrust of classroom, playground and general school life.
This is not about the pandemic, however, although the shortcomings identified during it have something to teach us about what this blog is about.
Open AI's ChatGPT
OpenAI released the latest, fifteenth version of its chatGPT software in December, 2022, and the world has been interacting with it ever since, increasingly amazed, awed, intrigued, excited and perhaps frightened by its power, scope, sophistication, speed, wisdom and - for want of a better word - humanity. Not only can chatGTP engage in quasi-conversation-like interactions; it can also write poetry, explain mathematical proofs, compose essays and stories, and write copy that passes muster as human and is frequently better than most human authors could write.
GPT stands for "Generative Pre-Trained Transformer", a generic term applied to artificial intelligence engines that, on the basis of their pre-training on curated data, aim to answer questions using ordinary language. chatGTP does not access the Internet or any exterior source of information in order to answer questions or perform its other functions; instead, it relies entirely upon the integrated learning stored in its neural net. It is unclear how large the API is, but given its scope we may suppose that it is very large indeed.
"What Price Education?" arises as a new and much more pressing question once we take bots like chatGPT into account, for why would parents in private education or governments in public education be ready to pay eye-watering sums of money for institutional teaching if a chatbot can supply something at least as good and arguably better - we will return to this later - virtually for nothing?
Why, indeed, would anyone suppose that externally-determined, assessment-driven, impersonal, inflexible and often seemingly disconnected and irrelevant institutional teaching based upon government-approved syllabuses and examination schedules would constitute a preferable alternative to an interest-driven, tailored educational trajectory defined by the personal abilities and interests of each child?
chatGPT does not remember conversations with users. Instead, it answers each questions as best it can while remembering the content of that particular conversation until such time as the user terminates it. If the same conversation is reloaded, chatGPT does not remember what it said before so is in effect starting from scratch. That might seem to limit its educational potential, but this inference would be premature.
There is no reason why chatGTP or a successor or a specialised educational chatbot with similar or superior powers could not remember exactly what each user had asked, said, done, and then after a suitable period of time provide the user with self-assessment tasks based on what the user had decided to learn. The purpose would be to enable the user to ascertain whether or not what was required had indeed been learned, and to take remedial action accordingly as necessary. Of course, for purposes of accreditation there is a question about cheating and authenticity, but to make this a fatal objection is to presuppose that traditional assessment regimes based on individuals are optimal, so all our assumptions should at least be questioned.
My experience of chatGPT has been that I can follow almost any interest as far as I would like to go, and that because its answers are closely matched to my questions, I am in a sense learning at a level appropriate to my knowledge and so incrementally in mind-sized bites (to use Seymour Papert's happy term) that are roughly commensurate with my capacity and stage of development. This matches the admirable insight offered by DeepMind's AlphaZero as described by David Silver, the project team leader, that because AlphaZero learned entirely by self-play, it always had an opponent of roughly the right standard, namely its own current standard. Personalised learning should always follow that principle with a little leeway to ensure that "our reach always exceeds our grasp".
chatGPT already knows more than most human beings, and except in specialised areas therefore will know more than most teachers, so if I were someone embarking either on a teaching career or thinking of building a school, I'd be asking myself some pretty serious questions about where all this will end. This genie is not going back into the bottle any time soon, and probably never.
On the Sceptics
A lot of the comments about chatGPT have concentrated on its shortcomings, its capacity to be confused by complex questions, and the general limitations of an approach at present completely limited to text. This is a mistaken, even foolhardy approach to something as remarkable and innovative. We have for too long allowed ourselves to be lulled into complacency by those convinced that there will always be "things that computers can't do" out of loyalty to an antiquated estimate of human capacities. I am reminded of the famous "God of the gaps" arguments about religion and science that were often based on the "gaps" that science could supposedly never fill or never explain. Both approaches are equally misguided.
The implications of this technology are profound simply because OpenAI has achieved a "proof of concept". Even as I write I am sure countless others are trying to do better, some with malicious intent, but everyone will be encouraged and spurred on by the mere fact that what they have achieved can be done at all.
The educational world has suddenly changed.
Watch This Space!
More on this shortly. There are so many implications for education that is is hard to know where to begin.