What happens when conversation stops disappearing?
I’ve been thinking about what Socrates could have done with a Dictaphone and a PA. He could have sat in that square in Athens having his conversations, recording every single one, then handed the recordings over at the end of the day and got his secretary to, in inverted commas, type them up. Imagine what we’d have.
Socrates didn’t leave books behind. Most of what we know about him and the way he thought comes from other people recording those conversations, most famously Plato. Given how important they became to Western thought, I’ve started wondering how much more we might have learned if somebody had been able to capture all of them rather than the tiny fraction that happened to survive. Not just the famous conversations either, but the ordinary ones, the ones that went nowhere, the questions he kept returning to because he still wasn’t satisfied with the answer, the conversation that seemed trivial at the time but connected with something else months later. What we’d have would not simply be more Socrates. We’d have a much better record of him figuring things out.
That interests me because I’ve realised recently that quite a lot of my own thinking happens in conversation. I go out every day and talk to people. Sometimes they’re friends, sometimes people I’ve just met, sometimes people whose lives or views are very different from mine. We talk about work, relationships, childhood, money, power, organisations, whatever happens to come up, and quite often I find myself trying to understand what their reality looks like and why it looks that way.
I’m not generally interviewing them. We’re just talking, but somewhere in those conversations something normally catches. Somebody says something that doesn’t fit how I thought a situation worked, or describes an experience in a way I hadn’t considered, and I start asking questions. One answer changes the question, which changes the next one, and by the end I can be thinking about something quite different from where we started.
Looking back at the pieces I’ve been writing recently, I’m increasingly aware that this is how quite a lot of them begin. I don’t tend to sit down with a completed idea and write it out. I encounter something, talk about it, worry at it for a bit, compare it with something else I’ve seen, then gradually work out what I think. Sometimes the useful bit comes from the person I’m talking to. Sometimes it comes from realising that their experience doesn’t fit mine. Sometimes it is simply the point where I notice I have been asking the wrong question.
The obvious problem is that most of that process disappears. I normally remember the conclusion, or at least the shape of it. I remember that somebody said something interesting, or that I walked away thinking differently about something, but I don’t necessarily retain the route I took to get there, and increasingly I think the route is where a lot of the useful material lives. The question before the useful answer, the thing I dismissed at first and came back to twenty minutes later, the contradiction that didn’t make sense at the time but suddenly connects with something somebody else says three months afterwards. Conversation is a pretty terrible storage medium.
I’ve already started compensating for that through dictation. If something interesting happens I can talk it through afterwards while it’s still fresh, what was said, what struck me, what I think it meant, what I’m not sure about. I can then put that alongside other things I’ve been thinking about and use AI to help me find connections or compare it with other people’s research.
But I’m now wondering whether I should take that a stage further and actually record some of the conversations themselves, where the other person is happy with that, partly because memory compresses things very quickly. What somebody actually said and what I thought they meant can become the same thing surprisingly fast. I leave with an interpretation and, after a while, that interpretation becomes most of what I remember of the conversation.
If I retain the original as well, I get to test myself. Was that actually what they said? Was that really where the conversation changed? Did I notice the important thing, or did I notice the thing that most conveniently fitted whatever I already happened to believe? I might well discover that I was wrong, which is rather the point.
Recording obviously isn’t new, and neither is transcription. What feels different now is what becomes possible once you have enough of the material. A thousand hours of recorded conversation is useless if the only way to make sense of it is to listen to a thousand hours of recorded conversation. Even transcripts eventually become another form of forgetting because nobody can hold that much text in their head or remember where the interesting bits were.
AI changes that particular constraint because I can ask whether the same idea has turned up in several completely unrelated conversations. I can compare what somebody told me last night with something another person said six months ago. I can put an observation alongside published research and ask whether they genuinely support each other, whether they contradict each other, or whether I’ve simply become attached to an explanation because I like it.
The bit I’m perhaps most interested in, though, is that I don’t have to resolve everything immediately. Writing tends to force some sort of conclusion because eventually the article has to end. Conversation doesn’t really work like that. Sometimes you get halfway towards understanding something and then the person goes home, the subject changes, or you just realise you don’t know enough yet. Normally those unfinished questions disappear pretty quickly.
If the conversation is captured, they don’t have to. Something I don’t understand in August might suddenly make sense because of a completely unrelated conversation in November. Something that initially looks like an exception might slowly become a pattern. Or something I was becoming increasingly confident about might start falling apart because too many people describe a reality that doesn’t fit it.
That is much closer to how I think research actually works, and it is also why I’m slightly uncomfortable with the way we tend to talk about AI as primarily a machine for producing finished things. The interesting part, for me, still happens outside. It happens because somebody tells me something I hadn’t considered, or because I ask another question, or because their version of reality doesn’t fit mine. It happens because two things that looked unrelated start to resemble each other, or because a tidy explanation falls apart once another person’s experience is placed next to it.
AI comes afterwards and helps stop me losing the process. In that sense it is much closer to Socrates’ imaginary PA than it is to an author. Keep this, type it up, find the other conversation where something similar came up, show me where these accounts don’t agree, put this observation next to the research and tell me whether I’m stretching it. Except this particular PA can eventually hold years of conversations at once.
I don’t really know what happens when you do that for long enough, which is partly why I want to find out. We spend a lot of time talking about AI generating things, generating an image, generating some code, generating an article, generating an answer, and I’m increasingly interested in almost the opposite use.
Not generating the thought, but keeping enough of it around that the next thought has something to build on.
Most of what any of us thinks disappears. Quite a lot of the good stuff probably disappears because it happened aloud, between people, and nobody thought to preserve it.
Socrates needed Plato.
I have a Dictaphone.
