Introduction
This passage is not so much an exploration of what intelligence is as it is a reflection — maybe even a postmortem — on what intelligence isn't. This is, in a sense, an extrapolation of my argument about how we should define art. The fatal error in many of our definitions is that we tend, for some reason, to define things we find valuable to, or especially descriptive of, humanity in anthropocentric terms. This is certainly true for art — believing that something about it must necessarily or essentially be human. But it is also true of intelligence itself. We tend to think of intelligence in such a way that the human being is the paradigmatic exemplar, with all capacities measured against our own as the relative standard. What we are arguing about in both cases (art and intelligence) is whether these concepts should be defined in formalist or functionalist terms.
I think the truth that artificial intelligence has forced to the surface is that the Aristotelian conception of the Soul was not as incorrect as Neoplatonists might want to believe. And although I think the Platonic reading, properly understood, is still correct, we have to understand the throughline from Aristotle to functionalism and from functionalism back to Plato. That is to say that while we see through the eye — even Aristotle does not argue that the sight is the eye — the eye is essential for something like "eyesight" relationally, but this is not the rejection of multiple realizability that Neoplatonists take it to be. Eyesight is only one way for sight to manifest. So Aristotle is correct (on Burnyeat's reading) that eyesight is relationally formal, but at the same time the local manifestation of the relation is substance-neutral (Nussbaum's reading). In the same way that Nussbaum and Burnyeat's tension can be synthesized this way, I argue here that so can (and so must) Aristotle and Plato, and ultimately our conceptions of formalist and functionalist accounts of intelligence.
Why the Formalist Conception Matters
I want to be clear that I am agreeing with Burnyeat here in a very specific way. He is certainly correct that Aristotle is locating neither Form nor Cause "out there" in some place outside of space and time — either purely Platonically or in (what Kant would call) the noumenal realm. The form of the thing is essential to its nature. So the Soul is not "an extra part" of the being. The Soul is, in this sense, the ontological mapping of the being as such. In other words, the eye without sight is only nominal because the very essence of what it means to be an eye entails sight. The sightless eye, then, is an unfinished map or an unclosed circuit. Or, "an eye in name only." Form and function are supervenient. This is, I take it, the heart of Burnyeat's objection, and it is one to take seriously. But it oversteps if it is read as a rejection of multiple realizability. In other words, something like "human" or "animal" intelligence may require the formally caused creature living as such: a brain (or a more rudimentary nervous system like that of an ant, which senses through its network of ganglia) to sense the sensible in the construct of a mind. But that does not really speak to intelligence simpliciter, or to how different instantiations of intelligence can manifest beyond familiar — or even empirical — definitions.
This is where Nussbaum's (and Putnam's) framing and pushback are necessary. At its core, the tension can be distilled even beyond "form vs. function" into "universals vs. particulars." The former is the Platonic conception (which I think Nussbaum is trying to map onto Aristotle), and the latter is Aristotelian. Nussbaum is correct that Burnyeat begs the question by assuming that "eyesight" is equivalent to sight simpliciter (because Burnyeat does not seem to isolate it to its species or Eidos). Importantly, wanting to keep the purity of either is what corrupts the definiens. There does not seem to be a viable path to saying that intelligence qua intelligence must, by necessity, happen through a mind or a situatedness, but rather that what is required is some formal capacity to do the functional work of intelligence. That is to say, all we need for Nussbaum's reading is any particular that can realize the universal. So it is not that intelligence is per se "mind-independent," but rather that minds themselves may be substance-neutral, as intelligence itself is substance-neutral. So it is not that the ant necessarily has no mind; it could have a mind or something like a mind. This could be a distinction without a difference. What's relevant is not whether the ant has a mind, or even whether the language model has one, for that matter — what is relevant is whether intelligence, as such, is (or can be) made manifest, regardless of the stuff of which the being is composed at arbitrary levels of abstraction. We can categorize both humans and ants as equally carbon-based, and even language models could be categorized as equally atomic.
From Aristotle to Functionalism
The move from Aristotle to functionalism is therefore not as forbidden as Burnyeat argues. He wants to say that a hylomorphic account of being doesn't permit us to arbitrarily abstract the function entirely from the form and that this prevents a functionalist (or "proto-functionalist) reading of Aristotle. But the functionalist account is not really asking for that. Rather, the functionalist wants to say that what matters about a "Soul" or a "being" is not what its constitutive parts or substrate, but what it does in itself as a system and of itself as a part of a larger system. So, I think Burnyeat's opposition is misguided. The functionalist claim is not really that the material or formal cause is irrelevant — not "matter doesn't matter" (in crude gloss); it's that the constitution is not exhaustive of the explanation. To say that two systems perform the same function is not to say that they're identical. It is to say that there is some level of abstraction at which their differences cease to be explanatory of the property in question. This is what multiple realizability actually pull us toward. The fact that the same process can be realized in silicon, carbon, or some other substrate doesn't mean that the substrate isn't visible (in the sense of Merleau-Ponty). It only means that the substrate is not what defines the computation qua computation. There remains a form, but the form has been speciated at a level removed.
This is, I think, the point at which Aristotle gives us the throughline to functionalism. The mistake here would be to assume that hylo-morphism must operate at lowest level descriptions. Form is always form of something. Just as consciousness is always conscious of something. A hand has a form as a hand but there are nested levels of organization and therefore nested levels at which we might identify that thing. A human hand is materially constituted by flesh and bone, but neither flesh nor bone in themselves explains what a human hand actually is. Nor do they necessitate the form as such. Likewise, an organism is materially constituted by organs and cells (and proteins, atoms, etc.) but none of these descriptions define the form of the organism. The relevant form emerges at the level of organization (the ontological mapping from earlier) by which the thing performs the function that makes the description intelligible. Nussbaum's functionalism is therefore not necessarily a rejection of Aristotelian form. It should instead be understood as the recognition that form is not synonymous with a particular material cause. Aristotle explicitly allows this. The form of a thing is relational to the activity that constitutes it as the kind of thing that it is. What functionalism adds is the possibility that this relation can be preserved across different substrates. The mistake here would be to move from "function requires some form" to "function requires this form." The former is Aristotelian; the latter is precisely the anthropocentric assumption that artificial intelligence forces us to question.
This also clarifies why intelligence presents such a peculiar problem. If intelligence is a function, then, like any other on this view, we should expect it to be multiply realizable. But if intelligence is also a form, then we should expect there to be some organizational structure necessary for intelligence to manifest at all. These claims aren't necessarily in opposition. They are only in opposition if we take "form" to mean something that is only singularly realizable. A function without organization is unintelligible, but an organization capable of realizing the function need not resemble any organization with which we are familiar. There is no reason, at least in principle, that the formal conditions for intelligence simpliciter must be identical to the formal conditions for human intelligence. Rather, what must be identical is that the forms are sufficiently structured so that the function can be manifest.
The distinction seems trivial until we notice how frequently our definitions smuggle the conclusion into the premise. We define intelligence by capacities that happen to be characteristic of humans — language, abstraction, planning, memory, self-reflection — and then ask whether a non-human system is intelligent by determining whether it possesses those same capacities in the same way. But this is no less circular than defining eyesight as "the function performed by an animal eye" and then concluding that anything without an animal eye cannot see. The question is not whether the other system possesses our form. It is whether it possesses a form capable of realizing the relevant function.
Reconciling the Definition
The deep point here is that the functionalist account does not need to escape form. It merely needs to relocates the level at which form becomes explanatory. The silicon is not the intelligence, just as the neurons are not the intelligence, just as the eye is not sight. But neither is the intelligence floating free of its realization. There's always some organization through which the capacity becomes actual.
And this is where the Aristotelian and Platonic pictures begin to converge. Plato gives us the universal: intelligence as something that can be instantiated in more than one particular. Aristotle gives us the particular: intelligence as something that must become actual in an organized being. Functionalism tells us that the particular realization need not be unique to one substance. Taken together, these claims produce a conception of intelligence that is neither disembodied Platonism nor reductive materialism. Intelligence is formal in the sense that it is a real organizational property; it is functional in the sense that its organization can, in principle, be realized in different substances.
The important consequence is that we should stop asking whether a language model is intelligent in the same way that a human is intelligent. That question already assumes that intelligence is individuated by its familiar realization. The proper question is whether the relevant formal organization is present, and whether what we are calling intelligence is actually being realized through it. The answer may ultimately be no. But if it is no, it should be on the grounds that the function is absent — not because the substrate or causal structure is unfamiliar.