I had the pleasure of talking to David Chapman at his house in Colorado a couple months ago. This is a small snippet from our conversation; we reflected on his AI research at MIT in the mid 1980s and the inspiration behind his current writing on metarationality.
We start by discussing David Chapman’s close collaborator, Phil Agre. Agre is one of the most fascinating and underrated thinkers about technology. He disappeared from public life around 2008, but the depth and breadth of his thought is difficult to capture. I’ve planned to write more on Agre, but for now this overview of his legacy does a good job.
Sections:
Phil Agre and a critical technical practice of AI
The origins of David’s Metarationality project
David’s time at Xerox PARC
Phil Agre and a Critical Technical Practice on AI
[00:00:00]
Max: Phil Agre has this paper about bringing a Derrida style literary analysis of early AI. I think specifically it's a paper by Herbert Simon that where he looks at it and he realizes that the way in which these sentences are constructed in the whole narrative is, he says, performing some kind of a hypnosis on the reader.
It basically funnels you into seeing the way that this author saw . You were talking before about Phil would just have these insights that would be very hard to express. There might be one or two sentences that no one else could understand, but you, after chewing on it, pausing on it, could come and make something of it, maybe build something.
David: Yeah I said I, I don't think I had a lot of influence on that paper. I think that's an absolutely brilliant paper, and I said I think it's his magnum opus. I was deeply embedded in the AI planning literature, which traces back to The paper that he analyzed there, and I think it was my interest in that literature, which eventually led to that paper. . He got led into continental philosophy through Lucy Suchman, who. was a student of Hubert Dreyfus. And so we both read Dreyfus's book on Heidegger, Being-in-the-World, a commentary. Yes, so that was immensely influential for both of us. And he went off and explored a whole lot more of Continental Philosophy than I had at that time. I don't know why he went so deep on that. I think he, again, this is, his genius was being enabled to see that there was something possible there that would not have ever occurred to me or anyone else.
[00:02:06]
I don't think anybody other than Phil would have gone in that direction. I don't think it was appreciated by very many people. I don't think many people have read that paper or would understand it or see that there's a point.
Max: I read it and I thought that would be so interesting to attempt a similar type of analysis on a Nick Bostrom philosophical paper.
I started to try to do that for his simulation hypothesis paper, but I'm not that well versed in this type of continental philosophy analysis, so it stopped very quickly. But I still, it still seems like that's a whole interdisciplinary inquiry that no one has taken forward.
David: Yeah. Very few have. He talks about a critical technical practice, which is more or less exactly what I mean by meta rationality. It's, asking hard questions about what are we actually even doing here? He went meta on AI in a way that nobody had done before from the inside.
You could say that Dreyfus’ book on AI and his critiques of AI were doing that, but he didn't have the technical understanding to be accurate about it. He got the broad story roughly right, but because Phil was on the inside, he could be much more precise and accurate about the critique.
The origins of David’s Metarationality project
My How to do Research at the MIT AI lab was the first metarational thing that I wrote. I think that was very influential for him and for actually quite a lot of people in AI at the time.
When it became clear that AI was at the time a degenerating research program, which was not going to go anywhere, which it didn't for 20, 30 years, my sense was actually the best thing I'd ever done was that how to do research paper.
[00:04:03]
In some sense, the Metarationality book is just trying to fulfill the promise of that thing. For a long time after I graduated, I got emails from people in all sorts of different academic disciplines saying “Somehow, I found this how to do research thing you wrote 10 years ago and it was so helpful in my work as a [something completely unrelated to AI]. Thank you for writing that.”
So I had this sense of, actually, the most important thing I can do is to, to try to do that for research in general. The Eggplant is basically that.
Max: Am I remembering there's another called How to Think Real Good?
David: Oh yeah, that was my my joking name for the project at that time. Around 1992, Rod Brooks, who was my Ph.D. supervisor had written a book about Lisp “How to become a Lisp programmer” and it got translated into Japanese and then back translated as “How to hack Lisp real good” by somebody who is proficient in Japanese and not English. And he had this book review written in kind of Japanese English posted on his door as a headline with big letters.
David: my informal name for what became the Meta Rationality book is now [00:06:00] “In the Cells of the Eggplant”, which I would like to drop now.
I want to walk that back, but my theory is that it's five parts, but they're really more like books or volumes, and each one will have its own silly title. So the first one, which is about rationalism, is In the Cells of the Eggplant. And the second volume, which is about reasonableness and ethnomethodology is called Aspects of the Theory of Breakfast, which is a play on Chomsky's early book called Aspects of the Theory of Syntax. And then the third volume Wielding the Power of Meaninglessness which is completely unwritten, is how rationality actually works and why, and right.
And it works because of meaninglessness. That's the trick. It's the decontextualization and the stripping of purpose. You take whatever the real world thing is and you turn it into equation, which is inherently meaningless. That's why rationality works if you can make it work.
The fourth part, which I'm struggling with now, is tentatively called Actually Caring About the Concrete Situation. I haven't got a title for the fifth part.
Max: How did you first come in contact with Lucy Suchman, you and Phil?
David: So Lucy’s PhD was notionally in anthropology, but her two main teachers were Hubert Dreyfus and Harold Garfinkel, who were respectively a philosopher and a sociologist. She was at PARC, Xerox PARC.
[00:08:00]
David: Because John Seely Brown had this insane, brilliant idea that in order to understand how computers could actually be used and useful, you needed to do anthropology on the users, and so he hired a bunch of anthropologists.
All tip top people. She was one of them. I think she was still a PhD student when she was first working there. And then after her PhD, she went full time. And I think she managed the anthropology department. Her boyfriend was Phil's best friend from high school.
Max: Wait, so she wasn't at MIT?
David: No, she had no connection with MIT at all.
So he was explaining her to Phil, who got very interested. And Phil I think he spent a summer while he was a graduate student at Xerox. I did two, not the same summers.
David’s time at Xerox PARC
Max: You were at Xerox PARC?
David: Yeah. I did two summers there, so basically everything I know about ethnomethodology was from hanging around with Lucy as much as she put up with me [at Xerox PARC].
Max: Were you helping with the anthropology?
David: No I was, I was an AI guy, but I could see that what she was doing was brilliant and important.
Max: So this was totally on the side. The main thing that they hired you to do as an intern was just to do some AI related research.
David: Yeah, I don't know how it works now, but at the time, the major industrial labs would hire graduate students from the top universities to basically just come and [00:10:00] do their thesis work and hang out.
Max: So you could hire them after they graduated?
David: That was probably a lot of the motivation. I was recruited there after I graduated I went and did a job interview there, and That was 1990, which was just before PARC got nuked. I couldn't tell what it was, but I thought, “Okay, no, I don't want to be here because something very bad is happening.”
Max: Did you interact with Varela, Roche, or Thompson?
David: No. I've never met any of them.
Max: Were you at all in dialogue or influenced by The Embodied Mind or the enactivist stuff?
David: Not very much. What that book said seemed obvious and not to lead anywhere. In a lot of ways I felt like what we were doing was making concrete their vague hand-wavy stuff.
Max: So would it be accurate to say your interactionist AI research [e.g. Pengi, Sonja] and the enactivists’ biological systems theory [The Embodied Mind, Tree of Knowledge] were both downstream of Dreyfus?
David: Yes. I'd also say all those things are downstream of Merleau Ponty, and Heidegger too. And Merleau Ponty is downstream from Heidegger, but Merleau Ponty made Heidegger a great deal more concrete than Heidegger was. Heidegger is, he worked out the ideas in a structural way, but he's [00:12:00] basically got one example, which is hammering.
There's a few other examples in passing, but none of them are actually worked out in any detail. So Merleau-Ponty worked those things out in a lot more detail, and his major work is called the psychology of perception, I think. And he recognized that perception is actually the key thing, which is not really a theme in Heidegger at all.
Heidegger was interested in language and action, but no, he didn't really think about perception that I remember.
Max: You're saying that it's obvious is, strikes an interesting note in me because it's I think incredibly obvious once you've made the shift. Like the ethnomethodological shift, but not to almost everyone else.
David: There's also the cybernetic shift, which is not quite the same. And that book is obvious if you have got the gestalt of cybernetics, which I did. I first encountered cybernetics when I was like 14 or something and got it then and had been working out the implications of that already for 10 years when I read The Embodied Mind.
Max: Could you give a concise summary of this cybernetic shift?
David: Yeah it's also interactionist. It's interactionist at a different level. The cybernetic worldview is that of the dynamic coupling between an organism and its environment or a system and its environment more generally. The idea is that you've got a very rapid flow of information perception and action going back [00:14:00] and forth such that there isn't actually a boundary there, or the boundary is nebulous and fictitious.
Some patterns of action. You can't understand perception and action other than in terms of the coupling of the environment and the system. And the mentalist, the alternative to the cybernetic view is the cognitivist, rationalist view where basically the mind is in a box and as much as possible try to ignore everything outside the box and you're looking at what's happening inside the box.
And the shift is just in getting yourself out of the box.
Max: So that, what do you feel like that point of view was present in the earliest form of cybernetics? Did you end up interacting much with Gibson or with ecological psychology?
David: We read it. Yeah, it was very influential. Though I don't think we actually interacted with anybody in that field.
Max: Was that downstream of cybernetics too? What you're describing about Interactions between the environment and its interface are very similar to how Gibson would describe affordances.
One of the curiosities I have is how interactionism, enactivist/4E cogsci and ecological psychology, relate to each other because there are people who are doing sports like MMA who are directly influenced by ecological psychology.
There's new versions, ecological dynamics, which tries to bring that into complexity theory and theories about human development. And people are actually changing how they do MMA training as a result of that. Tracing the lineages gives some sense of like where these ideas are [00:16:00] coming from.
Max: That might in some sense be like the most concrete result out of all of these intellectual projects. Better MMA. Better MMA and better sports. It seems to be working pretty well. At least the cutting edge, Brazilian Jiu Jitsu people are really excited about ecological dynamics.
David: I had no idea. That's very cool.
Max: Yeah. And active inference, which I think came about after you were less interested in this for instance, whole free energy principle stuff. Which you have not really engaged with or don't find that interesting.
David: Yeah, I'm pretty skeptical of that.
Max: You tend not to like big picture theories.
David: Yeah, so a big picture theory is valuable if it actually explains concrete things, and, I haven't seen that the Friston stuff actually explains any concrete thing in a way that gives insight or testable results.
I could be wrong and you could tell me that I'm wrong and I might be interested.
Max: So far I've only found vague hand-wavy results. Using some Bayesian math to make predictions about brain waves, but there's a lot of technical details that it's hard for me to judge so far.
But it is poetically inspiring, I will say.
David: Yeah there's no accounting for taste.
Max: Yeah. So maybe po poetically inspiring is not the right word, but it's like, it gives an account that seems persuasive to a rationally inclined person, such that they can learn to see in ways that might take them a lot longer if you tried to walk them through the cybernetic shift.
David: Oh, that's interesting.
Max: That's my hypothesis anyway.
David: That, that may have value.
Max: So all of these things we're trying to deal with [00:18:00] the separation of subject and object in their own way, and also the false attempt to separate emptiness and form,
and they have similar shapes in their responses, but we haven't been able to , do that much with that. And it's baffling.
David: Yeah it is baffling because this was a very, the cross pollination of all of these things was happening around Xerox PARC in the 1990s and it seemed very exciting and things were coming together. I guess it was 80s and the 90s. And Lucy and, was really at the center of that, and Brian Smith
Max: Brian Cantwell Smith?
David: Yeah. , he was at CSLI, which was Stanford's Center for the Study of Language and Information, which still exists, but it he and Lucy were close. They Co founded an organization called Computer Professionals for Social Responsibility with Winograd, I think.
I think they were the three founders. Cool. I read a paper, I saw a paper recently about why that got, didn't continue. But anyway, I probably should read that because it's a mystery to me why that scene, which was very exciting. And I was a junior member of why that scene came to a sudden end.
I think Xerox defunding PARC was probably a major aspect. Brian Smith had a health issue and wound up at the University of Toronto where he still is. So he left. Phil went off to do his thing. I went off to do my thing. Basically everybody, for different personal reasons, left, got dispersed, and the scene evaporated.
And so that [00:20:00] intellectual lineage didn't get carried forward, and 4E seems like it is actually exploring the same set of ideas, but they don't seem like they've made any progress, so I don't understand why that is.
Max: Yeah, maybe we'll never find out , maybe that was a small moment in the history of civilization.
David: I, I have been meaning to see if I could get Lucy to reminisce about that. Ash, she and I were on a Zoom call together for the first time about a year ago. It's the first time I had spoken to her in. 25 years. I think we both enjoyed it a lot and I've been meaning to contact her ever since and not gotten around to it.
Max: The plot thickens. Ha! To be resumed.
Share this post