Answer Machines

Why AI Should Make Us Ask Better Questions

“The trouble with computers is that all they give you is answers.”
Pablo Picasso

¹

I would like to say upfront that ChatGPT was not involved in the making of this article. There will be no cute reveal towards the end, and no paragraphs that I’ve elicited from the bot. If you find the prose bland and the thinking derivative, I’m afraid that’s down to me.

Apparently, all the smart people now have ChatGPT as a permanently open tab. Maybe I’ll join them - it’s certainly fun to play with - but right now, I find ChatGPT useful simply as an idea. It’s like a competitor that I’ve internalised, somewhere between editor, coach, and conscience. That’s fine, it seems to say about every piece I write, but are you sure I couldn’t have written that? What are you bringing to the party?

It’s a question worth asking of any piece you read, too, including and especially op-ed columns on AI and education. In recent months there has been much commentary on how ChatGPT and its successors should change the way we think about schools, and it has had the distinct flavour of a bot trained on future-of-education boilerplate. You know the kind of thing:

Our education system is rooted in the industrial age, designed to produce pliant workers who can carry out repetitive tasks competently and joylessly. The system sees students as vessels for knowledge rather than as humans with potential. We now live in a world where anyone can access any information they need any time they want. We shouldn’t assume that the skills of the future will be anything like those which have served us over the last century or two. ChatGPT can do all the stuff we spend years teaching students to do, but in a fraction of the time. It can write perfectly good essays and even pass an MBA exam. There’s little point teaching content. We don’t need to be drilling our students with information. AI has lowered the returns to factual knowledge. Why should students be forced to memorise facts about ancient Rome or mitochondria when they can find out anything they want in an instant? Let them follow their interests and learn how they want to. In an unpredictable world, we should be teaching people skills that machines can’t replicate - problem-solving, creativity, empathy, leadership. They will figure everything else out for themselves.

This story - which has been told, in varying forms, ever since Jean-Jacques Rousseau wrote Émile - is not wholly wrong. I wrote a book about why curiosity is an increasingly valuable skill (although I think it should be inspired, not taught, by schools). I agree that the future is unpredictable and that this makes it hard to know which skills to teach. For instance, learning to code suddenly seems like a lower priority than it did, now that we have tools which can do it for us, at least when it comes to the kinds of basic tasks for which most people would be using code anyway.

But literacy and numeracy and are not skills in the same sense as coding is. They are more fundamental. They’ve proven to be of enduring worth even as society has undergone multiple transformations. They underpin cutting-edge fields of knowledge like machine learning. When educationalists or economists say that we can’t predict the skills of the future, they’re right; what they get wrong is that there is therefore no point teaching content, and that the traditional curriculum should be abolished so that we can start again from Day One.

In truth it is more important than ever to learn facts, or content, or knowledge or whatever we want to call it. Not because facts are ends in themselves, but because they help us ask better questions. If one thing is clear already from our experiments with ChatGPT and Midjourney it’s that the quality of the machine’s output depends on the quality of human input. It’s all about the prompt.

Observing some of the miracles performed by ChatGPT3, I’m reminded of something that the founder of Wired, Kevin Kelly, wrote back in 2004: “Answers are becoming cheap. In fact someday, answers (correct answers!) will be so cheap that the really valuable things will be questions. A really good question will be worth a thousand correct answers. I don’t mean that metaphorically. I mean that in the marketplace. Good questions will cost more than good answers.”

I think education pundits are right to be concerned about schools and universities sucking the joy out of learning. On the other hand, there are plenty of teachers and lecturers within the system who inspire their students, so it’s clearly not impossible to do so. My guess is that what these pedagogues have in common is a focus on helping their students ask better questions. They’re training them to become better dialogue partners. I don’t think you can teach that skill in the abstract; it has to come via immersion in the topic itself.

How do you ask a good question? Well, you can get lucky and ask one out of pure ignorance, like the little boy in the tale of the emperor’s new clothes. But to ask good questions consistently, you have to know what you’re talking about. Your question about the roots of the American civil war, or the structure of an mRNA vaccine, won’t be very interesting or generative until you know something about those topics. Deep, productive curiosity has to be rooted in knowledge - by which I mean information about a topic, authentically understood.

We tend to think of curiosity as born from ignorance - the kind of beautiful ignorance that Rousseau associated with the child in nature, picking up a leaf and wondering how it’s made. “Let us omit from our early studies such knowledge as had no natural attraction for us, and confine ourselves to such things as instinct impels us to study,” he wrote. The child should be given “no verbal lessons; he should be taught by experience alone.” The trouble with adults, Rousseau believed, is that they are too eager to force their arbitrary knowledge into young minds. “What is the use of inscribing on their brains a list of symbols which mean nothing to them?” he asked.

Multiple iterations of this argument can still be found in think tank reports, TED Talks and LinkedIn posts, by authors who think of themselves white hot futurists while relying, unwittingly, on the ideas of a three-hundred-year-old French philosopher. In coining the idea that children learn best when embarked on personal exploration, untrammelled by adult knowledge, Rousseau created a powerful inter-generational meme. But he was wrong about what inspires learning. As I explain in Curious, the empirical evidence suggests that curiosity is not a function of ignorance but of information. We need to know something about a topic in order to be aware of what we don’t know. Then the more we know, the more we understand, the more we want to know. We talk about ‘sparking’ curiosity, but sparks are useless without kindling. Rather than quenching curiosity, facts are the logs we throw on the fire. (By ‘facts’ here I mean knowledge or content more generally).

One subject that should becomes more important in the age of AI is philosophy, which is essentially the study of questions. When we are capable of asking good questions about a topic, we can add to the store of human knowledge as well as drawing down on it. Doing so requires real understanding, and real thought. It’s important to remember that the AIs, remarkable as they are, do not understand anything. They are elaborate auto-complete machines, parasitical on the corpus of existing human knowledge. They don’t ask us questions because they only have answers.

Like curiosity, creativity doesn’t happen in a void. It starts with combinations of existing knowledge. David Hume, Rousseau’s contemporary - and frenemy - wrote that there is nothing particularly interesting about the idea of gold, or about the idea of a mountain. But a gold mountain? Now you have something. Successful innovators and artists amass vast stores of knowledge from they draw on unthinkingly. Having mastered the rules of a domain, they are empowered to rewrite them. They can remix ideas, mint analogies, and import ideas from other domains. Think about your favourite artists - I bet they have a vast bank of knowledge, of songs or books or whatever it is, and that this sustains and inspires their originality.

The point of amassing knowledge isn’t so that it can sit there inert until regurgitated in pristine form at the appropriate moment. It’s so that it becomes a rich and fertile compost for deep thought, new ideas - and better questions.

1. No, really - this is that rare thing, an authentic Picasso quote.