Socrates, Plato, and the AI in my terminal

English is my second language. I don't have the vocabulary to be elegant. For years this was enough reason to not publish anything.

Then I spent an afternoon arguing with an AI about whether I should write at all. Somewhere in that conversation I realized I was already doing it. Not writing. Thinking. Through dialogue.

The Socrates problem

Socrates never wrote a single word. Everything we know about him comes from Plato, who captured his conversations. Socrates actually thought writing was dangerous.

In the Phaedrus, he tells a story about the Egyptian god Theuth presenting writing to a king. The king's answer: "What you have discovered is a recipe not for memory, but for reminder. And it is no true wisdom that you offer your disciples, but only the semblance of wisdom."

Writing gives people the appearance of understanding without actual understanding. You write something down, you think you know it. But you don't. You just have a record of it. Unlike a person, the text "continues to signify just that very same thing forever." You can't argue with it. You can't poke holes. It just sits there, looking wise and saying nothing.

His method was dialogue. Ask questions, get answers, find the contradictions, adjust, repeat. Twenty-four centuries later, programmers call this rubber duck debugging. Explain your problem to anything and the act of explaining forces clarity. Chi et al. confirmed this in 1994: students who explained things out loud performed better than students who just re-read material.

Paul Graham says the opposite. Writing is thinking. Leslie Lamport put it blunter: "If you're thinking without writing, you only think you're thinking." Writing forces precision. You can hand-wave in conversation, nod along, skip the hard parts. Writing doesn't let you.

So who is right?

Both, and that's the problem

Writing and dialogue are different kinds of thinking.

Writing is linear. You start with nothing and build sentence by sentence. It forces structure, exposes gaps. When I write architecture documents at work, I catch problems I never noticed while talking about them. Flower and Hayes showed this in 1981. Writing isn't transcription of prior thought. You discover things in the process of writing itself.

Dialogue is reactive. Someone says something and it triggers a thought you wouldn't have had alone.

Vygotsky had a theory about this. All thought originates in social dialogue. First you talk to others. Then you talk to yourself out loud. Then you internalize it as silent thought. Dialogue isn't a byproduct of thinking. It's the original form.

Bakhtin went further. All cognition is dialogical. Even when you write alone, you're arguing with an imagined reader.

Replace the other person with an AI

Is it still thinking?

Clark and Chalmers wrote a paper in 1998 called "The Extended Mind." If a tool helps you think, it's part of your thinking. A notebook that stores your memories is as much "your mind" as the neurons in your head. Earlier this year, Clark himself co-authored a new paper extending this to LLMs.

If I think better in dialogue with an AI than staring at a blank page, the AI is part of my thinking process. Not doing the thinking for me. Helping me think.

But there's a real risk. Gerlich found in 2025 that frequent passive AI use correlates with worse critical thinking. The more people used AI without intention, the worse they got at reasoning.

The follow-up study is what matters though. Structured, deliberate AI dialogue actually enhanced cognition. The difference is asking the AI to challenge you versus asking it to agree with you. Socrates challenged assumptions. A badly used AI confirms them. Singh's 2025 paper nailed it: "AI's tendency to align with preexisting beliefs can reinforce existing perspectives rather than challenging them."

When AI confirms rather than challenges, it's the opposite of Socratic method.

How this article happened

I opened a terminal to update my site's dependencies. Routine maintenance. While doing that, I mentioned to the AI that my site was abandoned because I didn't know how to write. The AI said I should try. I said everything sounds AI-generated now, so what's the point. It proposed five article ideas. They all sounded like AI garbage. I told it that. I said I was paralyzed.

Then I asked a question I didn't plan to ask: "Did Socrates write? I bet he didn't."

That question came from me, not the machine. In the middle of a conversation about package updates. That kind of jump is what dialogue makes possible. No outline would have led me there so naturally.

From there the AI researched Plato's Phaedrus, found the actual quotes about writing as "semblance of wisdom." I said that connects to Paul Graham. The AI helped me find Vygotsky, Bakhtin, Clark and Chalmers. I didn't know those names before today. Now I understand why they matter to what I'm trying to say.

The AI wrote a first draft. I rejected it. Too clean, too structured, too much like a blog post. It wrote a second. The title was wrong and the date was from six months ago. I asked a different AI to review the third version. It said the title was weak, the detector section was too defensive, the ending sounded like an apology. The AI applied those changes. I kept pushing. This is the seventh or eighth version.

Who wrote this

An AI typed these words. I didn't type most of them. But the jump from package updates to Socrates was mine. The rejection of three drafts was mine. The decision to ask a different AI for a second opinion was mine. That question, "whose article is this?", was mine.

Socrates thought through dialogue with other people. Plato wrote it down, but the animating force of those dialogues was often Socrates. We've lived with that kind of mediated authorship for centuries.

I don't know what to call what I'm doing. It's not "AI-generated content" in the way people mean it. But it's not "human-written" either. Maybe the distinction matters less than whether someone actually struggled with the ideas, changed their mind, arrived somewhere they didn't expect.

Or maybe I'm just rationalizing.

The wrong metric

Stanford found that AI detectors flag 61% of essays written by non-native English speakers as AI-generated. My second-language English looks like machine text to an algorithm. Run this article through a detector. It will probably flag it. It would have flagged me anyway.

The question everyone asks is "did AI write this?" The better question is "did someone think this through?"

This is how I think now

Dialogue to find the ideas. Writing to pin them down. An AI that helps me find research I wouldn't have found alone, and challenges assumptions I don't see. Not because it's the right method. Because it's mine.

Socrates was right that dialogue creates thoughts you can't reach alone. Graham is right that writing forces precision. Neither of them imagined the version where both happen at once in a terminal window.