From Plato's Academy to ChatGPT

Dan Stratan, CEO and Founder RCI Holding in dialogue with Tom Duff Gordon, Vice-President and Head of EMEA Policy at OpenAI, during the panel "From Plato's Academy to ChatGPT" at the 30th Annual Economist Government Roundtable in Lagonissi, Greece, July 10th.

Good morning,

It's a privilege to be here in Athens — the city where the Western idea of education was born.

Twenty-four centuries ago, in this city, a man with no wealth, no army, and no institution behind him sat in the shade of an olive grove with a handful of people who had the great luxury of time. It was here that Socrates walked the agora asking questions that made comfortable people deeply uncomfortable — and in doing so, taught an entire civilisation that the examined life was the only one worth living. Slowly, painstakingly, he taught them to think. That was beautiful. He changed the course of human thought, armed with nothing but a question. He left behind no textbook but he did leave behind a method. And that method outlived every empire that came after it.

You can see that man, Socrates. He is here behind me. He is wearing a green robe and as always we see him caught in earnest debate.

Socrates believed education wasn't about filling minds with answers, but training them to ask better questions. Plato's Academy aimed to educate people who could be trusted with power — not because they had the right answers, but because they'd been taught to ask the right questions.

We are meeting today because that idea is under pressure from a technology so useful that it is rapidly reshaping what it means to learn. The most powerful educational tool in history is being deployed at scale — without a clear conception of what kind of minds it is meant to produce.

Minds that are passive receivers of facts, or minds that exercise critical reasoning?

So here's the question: how do we  prevent generative AI from making us cognitively lazy? How do we use it to enhance minds rather than to diminish them?

Four points to I want to make before I answer that:

First: credit where it's due. Tom's company put a potential tutor in the pocket of any fifteen-year-old, anywhere, with no local school able to match it. That's genuinely historic. What follows isn't a critique of what OpenAI built — it's a question about what it should build next.

Second: access to information is not, of course, the same as wisdom. Socrates called himself a midwife to thought— he delivered questions, not answers, forcing people into aporia – a state of productive confusion. His argument? - that confusion was the  engine of learning. ChatGPT does the opposite: a frictionless oracle that resolves uncertainty instantly and with zero effort. Remove the friction, and you quietly remove the thing that built minds capable of governing, innovating and leading. That's not a flaw in the product — it's the product working exactly as it was designed to, optimized for satisfaction over struggle. A product tuned for effortless engagement rather than deep deliberation. The question is whether "as designed" is what education actually needs. Is it what humanity actually needs?

Third: evidence from our side of the table. We're an enterprise customer of this technology, not a critic. At RCI Holding, when our teams leaned on AI for conclusions, they  became fluent — and brittle. They couldn't defend their own reasoning under pressure. The fix wasn't less AI. It was redirecting it: instead of asking for answers, we asked it to attack our conclusions — find every weak assumption, every line a hostile counterpart would exploit. Used that way, AI becomes a tireless sparring partner. Teams that adopted that posture got sharper. Teams that didn't, became slaves of insight they never actually owned.

That's why this conversation belongs in a room with regulators — not only a room with the company building the product. I'm not asking us to slow down technology. I'm asking us to steer. Tom — should that redesign be the user's job, or the product's?

Fourth — and this is where I want to spend the rest of my time: there's a version of this technology we aren't building yet. We should demand it. And here’s the pitch to Tom!

The classical Socratic method never scaled. One philosopher, one student, for long hours — Athens only managed that by giving a tiny elite the leisure an enslaved economy made possible. Nobody can afford the model of one human tutor to one human student.

But AI may be the first technology in history that can put a Socratic teacher beside every student, simultaneously, at zero marginal cost. Imagine a system that doesn't answer your question on contact — it asks what you already believe, surfaces the contradiction in your own reasoning, and waits. Every capability this needs already exists. This isn't a technical barrier. It's a design choice. And right now, the market has chosen the answer machine, because that's what sells fastest.

The scale is staggering. Socrates spent a lifetime questioning a few hundred people, in one city, in one language. A Socratic AI can sustain that same rigorous, patient dialogue with tens of millions of students at once, in every language — and never grow impatient, never make a struggling student feel small for not yet understanding.

That last part matters more than it sounds. In a classroom, confusion has a social cost. The student who says "I don't understand" in front of thirty peers is taking a real risk — so most stop taking it. Most of us here will have suffered that feeling to some small degree. But few of us, I doubt, are young enough to have experienced as our children do, the fear of a moment of revealed ignorance going viral. And so, they nod along. They copy the method without grasping the principle. They learn to perform understanding instead of achieving it in reality. An AI tutor removes that cost entirely. A student can be as confused, as wrong, as slow as they actually are — with no audience and no judgment. Which means the productive confusion Socrates depended on becomes available to exactly the students who currently avoid it most.

That's what I mean by scaling Socrates: not a tool that hands out wisdom on demand, but one that recreates, for every student everywhere, a patient interlocutor who believes they're capable of finding the answer themselves — and simply won't let them stop thinking until they do.

It's buildable today. I know, because I have built a version of it and will share it with you in a minute. The only question is whether we have the will to demand it before the market finishes building something else in its place. And by the way, I plan to get Tom on my side in this venture. I am proposing a use for AI that is more engaging than any I have previously seen!

Closing statement

We've built the most powerful tool in history for gathering information. The question now is whether we'll also build the most powerful tool for cultivating judgment. Those are not the same thing — and the difference will be decided not by the technology, but by us.

That choice can't be left to markets, algorithms, or some future government. It has to be made now, while an entire generation's habits are still being formed. The defining question of our time isn't what AI can do. It's what we want it to do for people.

The problem is that we've raised a generation conditioned for immediate gratification and that we can see the quiet collapse of critical thinking. It's the story of a generation that's learned to outsource thinking, and a culture that largely let them.

And here's who pays the price if we get it wrong: a generation of passive thinkers. For the tech companies, that's a policy problem, a regulatory problem — someone else's problem. In this room, it's ours.

I don't want to end this introduction as a list of worries. So I brought something tangible as a gift for you all. On the screen is, what I believe the young people call a ‘QR code’. Scan it — it's a prompt, nothing more. No app, no subscription. Just a page of text.

Paste it into whatever AI you use — ChatGPT, Claude, anything with a chat window — and for that conversation, it stops being an oracle and starts being a midwife to thought. It won't answer your questions. It will question your answers.

Try it tonight. Some of you will find it irritating within three exchanges. Good — that's aporia doing its job. Stay a little longer than feels comfortable, and see what's on the other side.

And maybe we can persuade Tom to include Socratic AI directly into ChatGPT — 'thinking mode'.

I'm fully aware there've been attempts before to use Socratic tutors in universities, without success, as students simply defaulted back to the faster answer. Goodwill alone doesn't work. So here's my ask to Tom and every company that builds AI models: build a reward system into the product itself (extra tokens, extended free access, priority usage) for anyone who chooses Socratic mode over the instant answer. 

If the product currently rewards speed, build a version that rewards depth too. Call it a Socratic incentive, if you like — one that pays for itself in users who come back smarter, not just users who come back faster. Do that, and this stops being a gift I handed a room in Athens for one evening. It becomes a feature every model in the world ships with. Then carry Socrates in your pocket — into your classrooms, your products, your companies, your policy.

Plato built his Academy on a wager: a society that teaches its citizens only to know things, and never to question things, will end up governed by people who are well informed and very poorly equipped to lead.

We are sitting where that wager was made — at the exact moment history's most powerful educational tool is being deployed without an answer to what kind of minds it should produce.

That is the work ahead of us. And it starts here. In Athens. Where it always begins.

INSTRUCTIONS TO USE

AI School of Athens

Sharpen how you think about AI, not just what you know about it.

What it does

System prompt turns an AI assistant into Socrates, a relentless questioning partner for AI ethics. It does not lecture or hand you conclusions. You bring a position, it tests it, finds the case you missed, and holds your own beliefs against each other until you see where they collide.

Sound reasoning earns obols, the silver coin of Athens. Clear answers and rebuilt arguments earn them; hand-waving earns nothing. A running tally follows you through the session. It is built for people who already reason well and want to reason better.

How to use it

1. Download the file below and copy all of its text.

2. Paste it as your first message in a fresh chat (Claude, ChatGPT, Gemini, or similar).

3. Send it. Socrates opens the dialogue from there.

For repeat use, paste it into a Claude Project or Custom GPT instead, so you don’t re-paste each time.

Socrates asks if you’re ready, then what to call you, then poses his first question. 

From there:

• Commit to a real position. He cannot work with hedges.

• Answer one question at a time.

• Expect to be wrong sometimes. Rebuilding your argument stronger is the point, and where the most obols are earned.

• Push back. The session succeeds when you leave with a clearer, better-defended view.

Importance notice: No software is installed onto you phone. We do not collect or store your personal data.