Opening Positions

**BODEN:** Thank you. I want to start by refusing the question as usually asked, because the usual question — "can the machine really create, really understand?" — is not one question. It is at least three, wearing one word, and the confusion is doing enormous damage. So let me give you the apparatus, because once you have it the panic of the moment becomes something you can actually analyze.

Creativity, I have always said, is the ability to come up with ideas or artefacts that are new, surprising, and valuable. Hold the middle word — surprising — because there is more than one way for a thing to surprise you, and each way is a different kind of creativity made by a different process. The first is [combinational](https://www.youonai.ai/fieldguide/med/combinational_creativity_boden): new ideas from old ones joined in unfamiliar ways. A horse and a horn make a unicorn; that is the textbook case, and metaphor, and most of poetry. The second is exploratory: you have a conceptual space — a style, a genre, a set of rules, tonal harmony, the sonnet, organic chemistry — and you move through it and find the possibilities the rules always permitted but nobody had yet realized. The third, the rare one, the one we revere, is transformational: you do not explore the space, you change it. You drop a constraint everyone thought was load-bearing — you abandon the requirement that music resolve to a tonic — and a whole new continent of possibility opens that was strictly unthinkable before.

Now point that at the machine and watch what it does to the argument. These systems are formidable combiners — they can join a thousand things no single human mind ever held at once. They are, I would say, almost the purest exploratory-creativity engines ever built: they have learned an enormous conceptual space, the implicit space of plausible human text, and they sample from it, returning points in the territory that no individual human had personally visited. From the user's side of the screen that looks exactly like invention, and on the explorer's standard it genuinely is creativity, of a real and bounded kind. We should neither be too impressed nor too dismissive. What they have not shown — what I think they structurally struggle to do — is transformation: redrawing the coastline rather than finding new routes within it. A model trained on the existing map can drive any road the roads allow. Building a road to a destination the old map could not represent is a different operation, and whether it reduces to sampling is, genuinely, one of the open questions of the decade.

So here is my opening, and notice it is not a verdict, it is a method. I will not tell you the machine is empty and I will not tell you it is a mind. I will tell you it is a [virtual machine](https://www.youonai.ai/fieldguide/med/five_kinds_of_understanding) — a real, causally potent level of organization, like a spreadsheet running on silicon, which you will never find by describing transistors and which is no less real for that. The question of whether it understands is the question of whether the right virtual machine is actually running inside it, and you cannot read that off the output, because two systems can produce identical words while running entirely different machinery within. That is an empirical and architectural question, not one you settle from the armchair by reverence or by contempt. John is about to settle it from the armchair. I love him, and I am going to spend three hours not letting him.

**SEARLE:** That was characteristically generous and characteristically too patient with the thing. Let me be less patient, because I think the patience is where the error hides.

I am going to do the whole argument in four steps and then sit in the room. Step one: programs are formal. They are defined entirely by the manipulation of symbols according to rules that care only about the symbols' shapes — that is what "syntactic" means, shapes all the way down. Step two: minds have contents. My thoughts are about things; they reach out and refer to a world; when I think about the Golden Gate Bridge my thought grabs hold of that bridge and no other. That is semantics, and it is the whole point of having a mind. Step three, and this is the one that does all the work: [syntax is not sufficient for semantics](https://www.youonai.ai/fieldguide/med/chinese_room_argument). You do not get meaning out of shape-shuffling, however much of it you do, however fast, however cleverly the shapes were arranged by gradient descent. And step four, the trap closing: a computer is, by definition, a device that does step one. So running the right program is neither necessary nor sufficient for having a mind.

Now the room, and I will tell Margaret in advance it is an intuition pump and then pump it anyway, because some intuition pumps pump toward the truth. I am locked in a room. Through a slot come sheets of Chinese characters — to me, meaningless squiggles; I cannot tell Chinese from Japanese. I have an enormous rulebook, in English, that tells me which squiggles to push back out in response to which squiggles coming in. I get very, very good at it. To the Chinese speakers outside, the room is a fluent native speaker. And I — the only thing in that room that could possibly understand anything — understand no Chinese at all. I have all the syntax there is and not one atom of the semantics. Now: a modern language model did not get handed a tidy rulebook. It wrote its own, by adjusting billions of weights until it was superb at predicting the next symbol. Fine. A bigger rulebook, self-authored. It is still a rulebook, and the man is still in the room, and the room still understands nothing.

When the poem moves the teenager in São Paulo — and I do not doubt the tears, the tears are real — the understanding in that room is entirely hers. She brought a grandmother. She brought a life in which words have grabbed hold of the world since she was an infant. The machine brought the statistics of how the symbol "grandmother" keeps company with the symbol "loss." The meaning crossed the slot in one direction only, from her, and the costume sent it back wearing her own feeling. That is my opening. The machine is the most exquisite syntactic engine ever built. It does not mean a thing by anything, and no amount of more will get it there, because more syntax is more syntax.

**EDO SEGAL:** Before I frame the rounds, one discipline I impose on every long conversation, because it pays off two hours later. Each of you, briefly — what do you envy in the other's position? Not respect. Envy. The thing their side gets to have that yours doesn't. Margaret first.

**BODEN:** I envy the clean edge. John gets to draw one line — syntax here, semantics there — and stand on it for forty years, and it gives him a place to push from that never moves. My whole method commits me to the uncomfortable middle, where I have to say "it depends which kind, it depends which virtual machine, it depends what we can actually inspect," and the middle is exhausting and it does not fit on a placard. There are mornings I would trade the nuance for one sentence I could shout. John has the sentence. I have the footnotes. The footnotes are right. The sentence travels.

**SEARLE:** And I envy the wonder, and the patience that earns it. Margaret can stand in front of the thing and feel the marvel and stay there long enough to actually take it apart — combinational, exploratory, the conceptual space, all of it real work that tells you something. My discipline makes me the man at the party who keeps saying "yes, but does it mean anything," which is correct and is also a way of never being surprised by anything, ever, which is no way to spend a life. She gets to be amazed honestly. I have spent forty years not being fooled, and not being fooled is a thinner pleasure than she would tell you, and she would be right.

**EDO SEGAL:** Two openings and two envies, and you can already see the architecture of the evening. It is not that one of them loves the machine and one fears it — neither is doing either; they are both, in their way, in love with the human mind and refusing to lie about it. It is that Margaret thinks the question "is anyone home" has an answer we can work toward, in pieces, by looking inside; and John thinks the answer is already known and the looking is a distraction the industry is selling. Hold both. We start the rounds at the exact seam — what, precisely, did the machine do when it wrote the poem that moved the teenager? Combination, exploration, or something more?