Beyond Concepts

"Ruth takes no prisoners. Ruth is an indomitable force. When she wrote her first book, Language, Thought, and Other Biological Categories , I read it in draft. I recommended it be published, edited the book, and wrote the foreword for it. I realized this was an amazingly powerful mind because she went into the analytic tradition—the propositional attitude tradition, the Frege, Russell, and Wittgenstein tradition—and really understood it. And she had the strength to find a position that was critical of it in a very powerful way. She was able to pull herself out of that vortex. I tried for years to do the same. The result is a paper of mine called “Beyond Belief”, which is my attempt to naturalize the propositional attitudes as part of the naturalization of the intentional stance. But in Beyond Concepts Millikan carries it much further, and she goes all the way to the metaphysics, to the ontology. The brilliant idea in that book is she throws away the traditional philosophical concept of the concept and replaces it with the ‘unicept.’ We had some wonderful correspondence about whether she should come up with another name. When she came up with unicept, I said, ‘That’s it. Go! That’s what you mean. You want to have a term that sharply distinguishes.’ You and I can both talk about cats and dogs, and I understand you and you understand me. It’s not a hard topic to communicate about. The standard line is, that’s because we each share the concept of ‘cat’ and the concept of ‘dog.’ A lot of philosophy is involved with getting clear about what concepts are. How you grasp them, as Frege says, and so forth. Millikan says, ‘No. The mistake is right there.’ You and I don’t have the same concept of dog. We each have a unicept. You have your unicept; I have my unicept. We each have our unique ways of telling dogs from cats and tracking dogs and tracking cats, and putting new things we learn about dogs and cats in the right “places” in our minds, but not because we have each acquired or mastered a specific concept, that we then might spend our entire lives trying to define. We don’t need to share exactly the same concept (in oldspeak); thanks to the regularities of the patterns in the world—the clumpiness of the world—we can use the world to keep our mutual understanding in registration. We don’t need to have shared concepts in our heads. Large language models, in a way, have unicepts; they don’t have concepts. They don’t have concepts and neither do we. Yes, it has some similarity with Locke’s attack on innate ideas. The wonderful thing about this is that Millikan saw how to turn her back, elegantly, forcefully, and with lots of good hard argument, against the established traditions of analytic philosophy. Some people were very angry about it. Jerry Fodor, for example—my old dear friend—I was ashamed of him for the way he handled Millikan. He viewed her as a horrible blight on his whole worldview. She was right, and he could never come to grips with that. I have written some interesting stuff in my memoir about Ruth and her interactions with Jerry Fodor. He was unbelievably rude to her. I’m a pack rat. I have been all my life. I see a bit of junk and I think, ‘Ooh, I can imagine a use for that.’ I’ll pick it up and put it in my pile of odds and ends of junk. I love tools. I’m a worker with tools and have a collection of tools that I cherish. I also have a collection of thinking tools, and sometimes I make thinking tools of my own. A good term is a thinking tool, and some of them are very simple. I give them a name and now people can use them. One of my favorites is the ‘surely’ alarm. Whenever a philosopher says ‘surely,’ a little bell should ring. Ding! This is usually where the weakest point in the argument is. It’s where it doesn’t go without saying, because the author seems to have to say it. Thanks to computers and string search, it’s easy to look for the occurrence of ‘surely’ in philosophical texts. Sometimes the alarm rings and it’s a false alarm, but it’s a great way of spotting the weakest place in an argument. It’s the place where an unexamined assumption or presupposition is being allowed to enter without proper vetting, and that’s where you look for mistakes. I guess I never thought of it in those terms, but I love doing it. I’ve also loved championing terms that others have coined that I think are more important than a lot of their critics think. ‘Meme’ is the obvious choice there. I think Richard Dawkins clarified, crystallized, and brought into focus a very important idea—the idea that there are things that evolve in culture that have their own fitness. We need that idea now because we’re beginning to create new kinds of very fancy memes that are generated by large language models (LLMs) that make counterfeit people, which can then reproduce. This is the most dangerous advance of technology in the last thousand years. It’s more dangerous than gunpowder and nuclear war because it may destroy trust. If we lose trust, we’re toast. What we’re doing right now, talking to each other, depends (in ways that we have underestimated or taken for granted) on a shared ideal of ‘let’s get it right, let’s go for truth, let’s not lie, let’s not deceive.’ If you can’t trust your interlocutor not to deceive you, and if you find it almost impossible to tell whether you’re talking to a person who shares that trust, then that whole fabric of knowledge (which depends not on God, as Descartes thought, but on human unity and trust) disintegrates. There have always been liars and deceivers, but we’ve created a new way of making the millions of brooms in The Sorcerer’s Apprentice . They’re out of our control, and they reproduce. That’s the point I want people to understand. Nuclear weapons can’t reproduce; software can reproduce. We’re living in the age of high-fidelity reproduction. This is the small, warm pond that Darwin imagined was the birth of life. This is the salubrious environment where technological memes can reproduce very quickly and create a pandemic. I’m very worried about this. I think we should take steps as soon as possible to protect ourselves if it’s not too late. It bothers me that some of the most intelligent people I know, including Geoffrey Hinton and Yoshua Bengio, think that it may be too late, that we’re beyond the tipping point. I hope they’re wrong. The big difference is that LLMs aren’t persons. Persons are indefinitely reflexive. They have higher-order thoughts and higher-order desires and beliefs. An LLM is a counterfeit person because it doesn’t have any higher-order values. That’s the problem. My own thought is that there’s no magical way that we think that differs in principle from the computational processes that go on in LLMs. The underlying hardware, though, is profoundly different. Nobody’s got a good theory yet for how (in detail) the computing brain makes a human mind, although people are working on it. My colleague Michael Levin is leading the charge in some ways, in understanding how individual cells, unlike the parts of digital computers, have to work for a living, and have to have agendas. They are collaborating and competing, and that collaboration and competition is the basis for creating the persons that can collaborate and compete the way you and I do. They could. I have a draft of a paper where I talk about how we could, but I think we shouldn’t—we don’t have to go there. It’s called “We Are All Cherry-Pickers.” I’ve been sending it around. My piece ‘The Problem with Counterfeit People’ was published by The Atlantic , which has had some serious traction."