The quantum of intelligence

What might a long-running debate in physics suggest about human cognition and AI?

Today’s soundtrack

Well, I’m back. I’ll make this brief: I arrived in the United States over the past weekend to the sounds of a woman screaming in the airport as she was hauled away by government agents. If you live in America and are horrified by the ever-ratcheting fascism surrounding us, please turn out to the “No Kings” protests planned on October 18. Tell others, harass them to join you, and please make it evident that We the People do not want what is happening.

---

This week’s essay may be a disaster. I’m going to try to sketch out some thoughts ranging across philosophy, quantum mechanics, behavioral genetics, and what we mean by “intelligence.” These are huge subjects that I approach as a thoughtful amateur (or so I’d like to believe); as a result, none of this may make any sense. But that’s the fun of writing a free newsletter, I get to play around.

I’ve written previously about the influence of Richard Rorty on my understanding of the world. One of Rorty’s major aims was to undermine the entire project of philosophy as it was traditionally conceived of in the analytical tradition, ranging from Descartes to Kant. In Philosophy and the Mirror of Nature, he contended that philosophers working within this tradition want to develop a special understanding of the “nature of knowledge and of mind” that is not bounded by place or time. They want to articulate a transcendent set of capital-T Truths that forever endure. On this view, to know something is to “represent accurately what is outside the mind.” Thus, analytical philosophy succeeds when it accurately holds up a mirror to the world around us.

All this, Rorty believed, was an intellectually interesting but ultimately fruitless endeavor, and one that had outlived its usefulness. Many a PhD thesis has been written exploring exactly why that’s so, but to summarize briefly: No description of the world, made by humans, can ever be freed of the circumstances in which they arise, and truth is better understood as a product of what it is better for us to believe, rather than an accurate representation of reality. As such, Rorty argued, we’d be better off embracing instead the idea that it is “pictures rather than propositions, metaphors rather than statements, which determine most of our philosophical convictions.” On this pragmatic view, what matters most is whether an idea is useful; whether it’s “True” is a label we attach later.

Pin that thought as we will now pivot from one incredibly wonky topic to another: quantum mechanics. Don’t ask why, but I’ve been curious of late to learn more about Neils Bohr, so before my recent trip abroad I visited my local book store hoping to find something on him. After some quick searching, the helpful employee at Good Reads pulled What is Real? from the shelf, a history of quantum-related science that’s written by one Adam Becker—that’s right, the same Adam Becker interviewed for this very Substack only a few months ago!

If you don’t know anything about quantum mechanics, don’t worry, we can briefly get into it here without making our minds melt. The basic idea is to describe how the world operates at the tiniest of levels—the atomic and subatomic, or “quantum,” world. Roughly 100 years ago, scientists developed a series of mathematical equations that, to a stunning degree of accuracy, allow us to predict how things operate in this ultratiny space. As Becker notes, massive technological developments followed, ranging from silicon transistors to, uh, nuclear bombs. The scientific hangup, however, is that the equations and descriptions of this quantum world only go so far—once we start dealing with larger-sized things, we have to use a different set of scientific ideas (such as gravity and relativity and the like) to describe the world.

Is this lack of scientific harmony a problem? It depends on who you ask, and this is the core debate covered in What is Real?. For Neils Bohr and many others, the answer is: Not really. And the reason this isn’t a problem is that, under their theory of quantum mechanics (known as the “Copenhagen interpretation”), quantum mechanics is not trying to mirror some underlying reality of the world that exists apart from our study and interaction with it. On this view, we can and should accept that the usefulness of our mechanistic descriptions is enough. As Werner Heisenberg put it, “the atoms or the elementary particles are not as real [as phenomena in daily life]; they form a world of potentialities and possibilities rather than one of things or facts.” Or from Bohr himself: “Isolated material [quantum] particles are abstractions.”

This view, suffice to say, troubled many a scientist, most famously Einstein—blah blah blah, God doesn’t play dice with the Universe, you’ve might have heard that before, but Becker finds a more delicious quote in one of Einstein’s letters wherein he (Einstein) described Bohr as a “Talmudic philosopher [who] doesn’t give a hoot for ‘reality,’ which he regards as a hobgoblin of the naïve.” For Einstein and countless others, science must be—wait for it—a mirror of nature, something that describes the things that are “out there.” As Becker puts it, “physics is about the world around us. It aims to understand the fundamental constituents of the universe and how they behave.”

We’re about to pivot again, but here’s the takeaway point: there is a central and enduring tension around whether science and philosophy are attempting to mirror reality and capture an essence of something capital-T True that exists beyond ourselves, OR whether they instead employ metaphors and abstractions that are demonstrably useful but not necessarily indicative of some “thing” that exists beyond the descriptions themselves. That’s a lot. I suggest reading that sentence a few times, ruminating upon it, then grabbing a beverage before proceeding to the next section.

---

What is your mental model of intelligence? For many people, perhaps even most, there’s a belief that intelligence is something real that exists within us. The source of intelligence, be it through genetic endowment or social intervention (such as education), is of course the subject of heated debate that I’ve written about previously, but the idea that we ultimately possess some thing (or trait) we call intelligence is broadly accepted, just as we possess, say, something we call height. We are smart or not in the same way we are tall or not.

This common-sense understanding of intelligence, however, gets complicated quickly when we unpack it scientifically. Consider the way in which we purport to measure intelligence through the use of “intelligence quotient” tests aka IQ. In his excellent book Understanding the Nature-Nurture Debate, the behavioral geneticist Erik Turkheimer—who I consider a friend—explains how these tests really work, which I’ll now quote at length:

Suppose you are running a company that provides proofreading services to publishers. You employ people who sit in front of screens, correcting written text. Spelling errors are the most frequent problem, so you are motivated to hire proofreaders who are excellent spellers. Therefore, you decide to give your job applicants a spelling test. It isn’t hard: throw together 25 words, and score everyone on a scale of 0–25. You are now a social scientist, a specialist called a psychometrician, measuring “spelling ability.”

Now, having developed your spelling test, say you become interested in ability testing more generally, and decide to develop a math test to go along with it. You ask people to compute 13 times 17 in their heads, and whether the square root of 8 is greater than or less than 3. Those math items are positively correlated with each other just like the spelling items were, so you can use factor analysis to estimate “math ability” in the same way. The math score that results may be useful in predicting other things you are interested in predicting.

Now comes the interesting part: it turns out that math scores are also positively correlated with spelling scores. As you go on to fill out your battery with tests of practical information, abstract reasoning, visuospatial skills, and whatever else you can think of, they will all be positively intercorrelated. You can therefore do another factor analysis, this time based on the test scores instead of the individual items, and estimate a “general ability” factor that describes the tendency to get items correct across domains. You could, if you wanted, call that score an Intelligence Quotient, IQ for short, and use it to predict things that interest you.

The belief that we possess some broad capacity of “general intelligence” thus arises, at least in part, from a statistical construct. Put another way, it is because there are correlations across sets of various cognitive tasks that some scientists infer there must be some “real thing” within us, something we can call intelligence, that is being described by these statistical measures. Surely if we look hard enough we can isolate and find this thing we call intelligence—and perhaps even recreate it artificially.

We can mirror our inner nature.

I hope with that last sentence you can begin to see the connection I’m drawing across these seemingly disparate scientific and philosophical realms, and the debate I’m trying to center. On the one hand, we have Rorty and Bohr suggesting we let go of the idea that scientific descriptions necessarily represent some underlying reality of the broader world. Intelligence, on this view, might simply be a useful abstraction (or maybe not so useful, considering the horrors that unfolded from eugenics). On the other hand, perhaps intelligence is real, and its only a matter of time before we identify its constituent components, using the tools of science. The debate that has long raged in the world of physics appears to be running in parallel track in the world of the cognitive.

I’m still working through all my thoughts around this, so I’ll close with the observation that in a weird way, AI as manifested through large-language models seem to straddle the line between these two perspectives. Those who believe “artificial general intelligence” is achievable are in many ways banking on the realist view of intelligence, that intelligence is something “out there” that we can recreate. Yet, at the same time, these tools are entirely probabilistic, in ways not entirely dissimilar from the probabilistic equations used in quantum mechanics. They simply work, even if we do not fully understand why they work. They present a world of linguistic potentialities and possibilities, you might even say.

I don’t know what it all means. But I find it interesting to think about.