"I cannot help feeling, Phaedrus, that writing is unfortunately like painting; for the creations of the painter have the attitude of life, and yet if you ask them a question they preserve a solemn silence. And the same may be said of speeches. You would imagine that they had intelligence, but if you want to know anything and put a question to one of them, the speaker always gives one unvarying answer.”
Socrates was not one of those who like to problematise everything just for the sake of it. His questioning was motivated, he claimed, by love. In the Apology Socrates tells us that he did not teach the youth of Athens to ask critical questions in order to undermine order but in order to help them and the whole community to flourish. For Socrates dialogue is the best way to find the Good: a concept which, for Socrates, was inseparably bound up with the linked concepts of beauty and truth.
Generative AI in the form of ChatGPT3 and GPT-4 is now daily showing us how right Socrates was to question the educational value of writing. If students can pass their exams and score high marks by handing in essays generated by digital algorithms then clearly writing is not and should not be the main goal of education.
There is an important distinction made in the philosophy of education which might illuminate the challenge to education posed by ChatGPT, the distinction between training and education. Education, according to the liberal education tradition, should go beyond training to encourage critical thinking, creativity and freedom of thought. This distinction has sometimes been misunderstood as meaning that we should value a more academic education, studying Plato for example, over a more vocational education, perhaps studying computer engineering. But that is a mistake. You can’t think without thinking about something or be creative in a vacuum. Whatever you do in life requires learning knowledge and skills that are best acquired through imitation and training. The truly educational bit of education goes beyond that training element to teach you how to question what you have learnt in order to understand it better and so to be able to move forward creatively regardless of whether your focus is how to build a better widget or how to write a better essay.
ChatGPT is a shock to the system. If ChatGPT can do well in exams then perhaps much of what has been passed off as ‘education’, even in elite universities like Cambridge, even in ‘academic’ subjects like philosophy, is actually just training. Our rubrics when marking essays point to the liberal education ideal with phrases like ‘demonstrates critical awareness’, ‘develops a personal synthesis’ and even ‘shows signs of original thought’ but actually teachers know that all of these phrases correspond to particular ways of writing which we can train students to emulate. They send us essays, we pick at them and send them back and so on in an iterative cycle until the students, or those motivated enough to stick with the programme, are trained to write just like us and just like every other academic in our area. This is very similar to the way in which ChatGPT is trained which is perhaps why it can learn to write academic texts even better than academics.
Why we are different from AI
I am not totally disillusioned with our so-called education system. I know that there can be more to education than just teaching students how to write essays in the way that I was once taught to write essays. I see this sometimes in the shining eyes and enthusiasm of students who have just had an insight that changes the way that they see things. Such insights happen not in the written text itself but in dialogues; in the dialogic space of new possibilities that opens up in the clash between voices or the clash between texts. Real education seems to imply a change in identity, or at least in perspective and also some sort of sense of an expansion of awareness to see things clearly that previously were opaque. This breakthrough or insight experience is just as relevant to more obviously vocational subjects like media studies, medicine, nursing, law, business, engineering and education as to more apparently ‘academic’ subjects like social theory or philosophy.
AI can be trained but it cannot be ‘educated’. This is because if you look into how ChatGPT works you find a set of algorithms with no understanding whatsoever. In the face of persistent claims that computers could become intelligent the philosopher John Searle spelt out the difference between how computers process information and how humans think with a thought experiment that he called the Chinese Room. The example goes like this: Imagine that you are locked in a room with a book of rules for manipulating symbols. You do not know any Chinese, but you can follow the rules in the book. Someone outside the room slides papers with Chinese characters under the door. You look up the symbols in the book and follow the instructions to produce an appropriate response. You then slide the paper back under the door. To someone outside, it may seem that you understand Chinese, but you actually do not.
That is the difference between AI and humans: AI does not understand what it is reading because it is not conscious; humans at least have the potential to understand because they are, or at least have the potential to be, conscious.
Dialogic education with AI
But what does this little word ‘conscious’ really mean in practice? Socrates nailed it with his example of a kind of questioning that is like an electric shock, waking the student up to see things differently. To become more conscious of something, to be able to understand it, means to question it, to challenge it, to see it from an outside perspective. When you are led by a dialogue to see something from an outside perspective you become more conscious of it. Only by seeing things as if from the outside, as if from a distance, can you begin to ‘understand’ them. AI cannot do that but you can. To teach students to be able to do something different from what AI can do we need to teach them how to question and how to engage critically, creatively, empathetically and productively in dialogue.
Dialogic education does not need to be complicated, in essence it is drawing students into dialogue, asking them questions, getting them to ask questions in turn and having as your goal not just knowing about stuff but being able to ask better questions about stuff.
ChatGPT or other generative AI can help with this. In every area of study, often especially in more obviously vocational subjects, it can reset the balance between training and education. At the moment the needs of training students so that they can pass required tests often overwhelms the more educational aspects of the curriculum. Conversations with AI tutors have the potential to make the required knowledge part of any curriculum much more fluid and accelerated, liberating more time and resource to the education bit; asking questions, exploring alternatives, engaging in dialogue and being creative.
Based on web-trawling the current version of ChatGPT, ChatGPT3 is trained on 300 billion words, using a statistical model with 195 billion parameters. The "parameters" in this context refer to the weights and biases of the neural network that makes up the model. These parameters are essentially the "knowledge" that the model has acquired through its training process, and they allow the model to make predictions about which words are most likely to appear next in a given sentence. The new version out today, GPT4, is even larger and more powerful. Because GPT is trained on content gleaned from almost the whole web it represents the almost up to date ‘knowledge’ of the Internet in a form that we can engage in dialogue with. Because of its generalist nature and programming to be plausible rather than accurate it is not always reliable in every area but it is easy to see how more focussed generative AIs trained with only vetted data will arise to help tutor students in specific areas like statistics for example, or law where precision is required.
Whether used as a generalist essay writer or as an expert tutor, generative AI can be a voice in an educational dialogue. If you want to say something original in an essay why not check first what ChatGPT or other AI, has to say and then see if you can add to that or critique it and go beyond it. Then check again what the AI thinks about what you have written, ask it what questions could be raised about your text and then try again iteratively improving your text in dialogue with the AI.
I am not saying that we can simply return to the pedagogy of Socrates. 5th Century BCE Athens was a different time and place. But the basic insight of Socrates, that we should teach students how to ask better questions, seems like a timely response to the educational challenge posed by ChatGPT3.