Can Rational Agency Survive in a Physically Closed World?

Good points there. On this one though, I’d have to ask, “improve” relative to what? To improve, or to find some change choice-worthy, requires that one have some end for which it is worth changing oneself in order to achieve. Whereas mindless evolution towards some end would not be “rational freedom” in any strong sense.

But any end is either ordered to other proximate ends (as when we make our car faster because our end is to win a race), or is considered choice-worthy itself. Now, a finite mind cannot process an infinite regress of ordering proximate ends to other proximate ends, so either such a chain ends in what is understood as good itself (perhaps more or less clearly though), or else it bottoms out in some unexamined impulse we just so happen to have. Hence, to have any meaningful rational freedom requires understanding one’s aims as truly good.

Consider some sort of hyper-advanced Roomba, say a self-replicating, evolving probe sent to terraform planets for its creators. Perhaps the probe is highly intelligent in terms of problem-solving. It is complex. It can alter future copies of itself according to the needs of its mission. However, if it does not understand its mission as truly choice-worthy, but rather has chains of reasoning that always bottom out in the inscrutable impulse implanted in it by its designers, it will be in some important sense unfree. The same is true in terms of second-order volitions; one needs to reshape one’s own desires according to some measure understood as worthy of such a reshaping, otherwise the same regress problem appears. So too, if it is impossible to be wrong about the worthiness any such ordering, it’s hard to see why freedom would matter in the first place, since all ends and their ordering is apparently arbitrary.

This is why, for a great many thinkers, true rational freedom is made possible by the rational appetites. It is man’s desire for what is really true and truly good that brings him beyond current beliefs and desires. This transcendent and ecstatic motion is what brings man beyond what he already is, allowing him to move beyond his own finitude and become more fully himself, and more truly whole (both being greater conformity to goodness and truth themselves).

But if the rational level here is akin to software, I have no idea how you derive more or less choice-worthy goals from it. How does one get the Good or Beautiful from equivalence classes defined over mathematized microphysics? And wouldn’t the very notion of a morphism between classes presuppose they belong to the same univocal system?

The head scratcher here for me is that most functionalism tends to be covertly voluntarist at root, precisely because it is layered over a metaphysics that denies the reality of goodness perhaps.

@EQV, @Count_Timothy_von_Icarus, I hope I am not misreading either of you, but it seems to me that you both, in related ways, feel like I am ceding too much ground to the reductionist physicalists. By granting supervenience and the thesis of the causal closure of the physical, my claim that contrastive explanations of the thoughts and behaviors of agents must make reference to high-level causal antecedents may seem to be an epistemological claim with little ontological import (EQV), and granting causal closure at the lowest level may already be an unwarranted metaphysical commitment to a rotten foundation (Count Timothy).

It seems to me, though, that granting the causal closure of the physical domain becomes easier once one recognises what the thesis actually implies. Causal closure relies on a reductionistic grounding of all genuinely physical predicates on micro-physics and the principle of locality (a cornerstone of, at least, classical physics) — and this very grounding limits the scope of physical explanations to phenomena individuated in point of material constituents or mereological composition only. Far from expanding the reach of physical explanation, closure curtails it. And this curtailment is therefore precisely what opens up room for autonomous higher levels of explanation, while maintaining inter-level connections like enablement and material constitution. As a result, listing the physical constituents (and their physical properties) of a rabbit at a given time tells us precious little about what a rabbit is, what its biological functions or natural behaviors are, etc.

The contrastive stress approach is useful for elucidating the non-accidentality point while highlighting the fact that contrastive causes are ontologically robust in a way that contrastive explanations aren’t always recognised to be. In my two earlier answers to @Michael where I mentioned Hitchcock and Elder, I meant to stress that what is ineliminable isn’t just the high-level explanation but the high-level cause. Saying that Walter ran into the classroom because the situation was urgent, or because that’s where the malfunctioning slide projector was located, doesn’t just provide different explanations of the same “event.” We are highlighting different causes of different outcomes, where the outcomes are individuated in two different ways in spite of them supervening on the same “physical event” (so called).

This connects directly to the causal closure point. Saying that physical closure is a dubious thesis is like saying that the claim that a pool ball’s motion is fully constrained by the Newtonian forces acting on it is dubious on account of the fact that Newtonian forces don’t explain why the 8-ball was last to be sunk in a pocket. But at no point in the high-level explanation of the ways the ball is caught up in a norm-governed game is there any need to deny that its trajectory is fully specified by the Newtonian forces acting on it. And the same is true, mutatis mutandis, of all the atoms and molecules that make up the ball and the players. It was precisely the task of my OP to show why the causal exclusion argument that brings this into question is flawed.

So I agree with Count Timothy that mechanism cannot be foundational. But I think the right move is to demote it rather than to challenge its local validity. Steven Weinberg was just plain wrong (in his “Two Cheers for Reductionism” chapter in Dreams of a Final Theory) to suggest that the arrows of explanation for empirical phenomena always point to a lower level of mereological composition. Explanations that point to the lower levels (or to causal antecedents at those lower levels) provide only partial explanations, often limited to enablement of higher-level powers, or to impediments to their actualisation in particular or abnormal cases, without offering any relevant clue as to what those powers are and why they are being actualised in specific ranges of circumstances (e.g. what normatively defined function they have).

One last point. Count Timothy gestures at “truth” being the telos of human action. I want to stress that on my view the formal aim of theoretical knowledge (truth) and the formal aim of practical reason (goodness, or warrantedness of actions) aren’t separable. They are two different employments of reason to navigate a unified space of reasons that characterises our human form of life. This matters because the rational-level causes I am identifying aren’t reducible to brute desires plus physical mechanisms. In citing the agent’s rational grounds for acting, we aren’t looking at antecedent “raw” desires, as in Hume’s early account of reason being the slave of the passions (in A Treatise), but rather at those grounds being efficacious through the agent’s ability to draw motivation from educated desires (the richer picture Hume arrives at in The Enquiries). This evolution in Hume’s thinking from a thin conception of reason as the slave of passion to a richer conception of reason as inseparable from educated habits of reasoning is something I owe to David Wiggins.

I type what I type because my body was causally determined to move the way it does. Your question seems to presuppose that physics is not causally closed and that we have libertarian free will.

Why does an LLM respond the way it does to our requests?

Perhaps I’ve missed it, but what exactly is non-accidental about the relationship between P1 and R2, or even P1 and R1? Or are you only trying to establish the non-accidental relationship between R1 and R2?

I am not sure how many people deny the latter relationship, since it seems self-refuting to do so (although certainly it’s common to do so implicitly without acknowledging it).

This is missing my point unfortunately. In the same way it is often said that, for an efficient cause to be an efficient cause, it must be defined in the context of formal, final, etc. causality (otherwise you are simply talking about a different paradigm), I don’t think it makes any sense to speak of “Newtonian forces” whilst simultaneously denying the metaphysical assumptions that are constitutive of Newtonian forces.

Precisely the equivocation I was trying to pin down is often at work here. The move being made is generally:

  1. Newtonian mechanics is predictively successful in certain domains

  2. Therefore its metaphysical picture of a closed causally complete system, and of cosmic homogeneity and univocity, is vindicated in those domains.

  3. Therefore causal closure holds in those domains (i.e., in the “physical” or at least the “classical scale physical”).

  4. Therefore causal closure holds generally, since everything is “made of” the “stuff” that those domains describe.

2 and 3 don’t follow from 1, nor is 1 particularly strong evidence for them. 4 doesn’t follow from any of them, and considering how incredibly fraught questions of composition are in physicalist ontologies (even in theories of restricted composition), it seems like a hard jump.

Plus, Newtonian forces, if properly Newtonian, aren’t delimited to some domain. That’s a consequence of cosmic homogeneity. Now to be sure paradigms evolve over time, but at a certain point if you change something truly foundational you no longer have the same paradigm. That’s happened with the “classical” picture, and yet its methodological reliability for appropriate contexts is regularly called on to bring back the original metaphysical baggage attached to it. Yet this move seems illicit to me unless it is backed up, and I don’t think it can be.

I would say the problems with composition in particular tend to be underappreciated. James Rooney has an excellent work, " Material Objects in Confucian and Aristotelian Metaphysics: The Inevitability of Hylomorphism," which looks at leading theories of restricted composition (all “hylomorphic”), and the way in which they still leave composition largely arbitrary because they are committed to the notion that substances can contain other substances as parts.

The physicalist will want to say something like: “the causal efficacy men qua agents and animals are sufficiently explained by what they are made of.” But then they cannot explain this composition at all.

Now of course, few deny that men are “made of” flesh and blood, and flesh and blood are made of proteins and lipids, and that these are made of chemicals, which are made of protons, electrons, etc., which in turn are said to be excitations of universal fields, etc. None of this really gets to composition though. The move from “men are flesh and blood,” to “therefore how men act can be explained in terms of flesh and blood without reference to man,” is not obvious either. Indeed, it seems that, at least epistemically, flesh and blood are only understood in reference to man, and chemistry only in reference to substances, etc., so that the order of knowing is exactly opposite of how the prioritization of microphysics has it (a massive problem if higher levels are epiphenomenal, since the higher level cannot be the cause of our moving our lips to utter things about them).

More problematically still, there is no way to start from protons or molecules, or even flesh, and to say: “this is one whole man,” or “this is a freshly dead corpse not a man.” Rather, we go in the opposite direction, starting with the whole and defining composition by participation in the whole.

Now the principles of physics are more general than those of biology, since not all things are living, and perhaps this is where the confusion lies. Generality is mistaken for ontic and causal primacy. This mistake is perhaps more obvious if we look at matter and form as principles of substances. To be sure, these are most general, but you don’t get anywhere without specifying this or that form, and not any other, or this or that matter, and not some other. The most general is in this way least powerful on its own. To say that something is tells us nothing about it, to say it is living tells us much more, to say it is rational still more, to say it is this specific man even more still. What is most general has the widest applicability precisely because it is closest to determining nothing at all. Qbits and the like are not most general because they are fundamental, but because they are closest to being nothing at all.

The process of understanding has to proceed from first principles (however unclear at the time), to the more specific, and then return to the more universal with a fuller understanding of what it contains. This is the process of reversion (epistrophe)—the intellect going out to the many and returning more fully to the unifying one that contains them. This is, however, precisely the opposite of begining with the many and trying to build back to the one. The latter is how you end up with extensional definitions that have to presuppose the principle they are being called in to define. Leaving microphysics as a self-contained base would seem to lock this reversal in place, or else create a layering of sui generis realms that exist on top of each other but do not interact.

Now, it might seem odd to say life is a more basic principle than the inverse square law or some such, but that is perhaps due to the tendency to think of causation in strictly temporal terms. Whereas, life clearly contains the potential to experience such physical principles, and intellect to understand them. Hence, they are prior precisely in containing these in a more eminent way.

For example, something like salt exists in a more eminent way in the intellect. Any individual lump of salt is “soluable in water,” but in reality it only ever dissolves in water when it is placed in water. It is only instantiating some of its properties at any given time, depending on its immediate context, and is not itself much of an intelligible whole at all. But in intellect all of salt’s properties can be present at once, and so it is more what it is when known then when it simply is (where it is almost nothing, lacking the noetic quiddity it has in intellect).

The order of materialism is an inversion of this order. It says that salt, etc. is basic, and all its intelligible whatness, only realized in intellect, is variously illusory, causally inert, or at best some sort of extra description floating on top of the ontic bedrock of the physical. Yet this sort of thinking has often tended to bottom out in a bare shadow world noumena being posited as the most basic and most real, despite its being wholly indeterminate, and so really nothing at all (granted plenty of philosophies stop a bit above this, at the level of brute mathematical structure for instance).

The other irony here is the claim of “sufficiency,” when the most common response to, “but why this physics,” is “it’s a brute fact.” But I’ll grant that is a confusing topic because the “laws of physics” are both very general (applying everywhere) and strangely specific (seemingly radically contingent). I would tend to take this as a strong demonstration of the fact that they are not intelligible in themselves however, and so represent the outreaches of being, not its fullness (this is perhaps clearer in attempts to reduce physics to 1 and 0, or the potential for 1 or 0, which seems about as minimal as one can get before we’re not speaking at all).

1 Like

Very nice, and I completely agree with you about the limitations of mechanistic explanation. It’s not that it doesn’t have a place, it’s that it can’t do the job all on its own.

But this raises the question: what more is needed? You mention higher-level powers, and this is, once again, quite suggestive. I’m sure you are well aware of the role that powers play within the Aristotelian explanatory economy. Powers are relative to form, form is act with respect to matter. Furthermore, form is tightly intertwined with the notion of intelligibility, and intelligibility is tightly intertwined with the notion of being.

So you seem to be gesturing at something like an Aristotelian metaphysics, but I sense that you are hesitant to fully embrace the ontological implications of form. You speak of powers, but not of the intelligible unity that grounds them. Or perhaps you are willing to acknowledge that intelligible unity, but not the irreducibility of its being.

And that’s where I think you are still vulnerable to reductionistic critique. You’ve convincingly demonstrated that some higher-level patterns cannot be located at the level of microphysics, and that these patterns have causal efficacy in the form of constraint, organization and contrastive determination. But insofar as you don’t ground these in something analogous to Aristotle’s notion of form, there’s nothing in your account that blocks the reductionistic impulse to ground everything in microphysics.

Causally determined by what? If you were typing a mathematical proof, then what you typed would be determined by the requirements of that proof. If, like Stephen Hawkings, God forbid, you suffered from a debilitating physical ailment, then you could not ‘dispose your body’ in such a way as to type anything whatsoever, but at least you could rely on a speech-assistance device to enable someone to type it for you. And presumably there were some things which only Stephen Hawkings could type, him being the only human with the mathematical insight to derive a particular result.

The physicist wants laws that are as universal as possible, true of all situations and therefore unable to tell us much about any particular situation — laws, in other words, that are true regardless of meaning and context… Such abstraction shows up in the strong urge toward the mathematization of physical laws.

In biology a changing context does not interfere with some causal truth we are trying to see; contextual transformation is itself the truth we are after… Every creature lives by virtue of the dynamic, pattern-shifting play of a governing context, which extends into an open-ended environment. The organism gives expression, at every level of its being, to the unbounded because of reason — the tapestry of meaning.

Stephen L. Talbott, “What Do Organisms Mean?” The New Atlantis, no. 30 (Winter 2011): 24–50.

The robot would design its successor, which would be faster, lower power, and hopefully more capable of finding even more clever improvements. It isn’t usually a change of one’s self, not easily done.

My comment wasn’t about humans improving themselves. That can be done without knowing how brains work. My comment says that machines will understand their own workings before humans ever do, which will likely be never.

Whereas mindless evolution towards some end would not be “rational freedom” in any strong sense.

So what would you consider ‘rational freedom’? Have an example? The term seems pretty meaningless to me.

OK, you mention reshaping of one’s desires (2nd order thinking), but it doesn’t work well if only some do it. Doing that needs to make one fit. Does it? If not, how does one solve such a problem? How is such thinking more ‘free’ if the goal is to control one’s thinking?

And BTW, the robots can also implement such 2nd order goals, but maybe only if doing so works better.

Consider some sort of hyper-advanced Roomba, say a self-replicating, evolving probe sent to terraform planets for its creators.

That’s the correct way to populate the galaxy. Send out self-contained seed ships that are in it for the long game. Terraform some random planet (by doing what to it??) and then design and populate creatures for that environment, (what goal does that solve other than the terraforming itself?). Should some of the creatures be intelligent enough to build new seed ships? Seems unnecessary since the replicating probes you describe seem quite capable of that task themselves.

Do we put humans on this planet? Easy to make some, but they’d not be fit for this non-Earth environment. Everybody seems to think they can recreate Earth with enough effort, but that’s fantasy. You need to evolve for the new place, just like humans evolved when they migrated to say colder climates.

However, if it does not understand its mission as truly choice-worthy, but rather has chains of reasoning that always bottom out in the inscrutable impulse implanted in it by its designers, it will be in some important sense unfree.

Be fruitful and multiply. That goal seems to work pretty good for everything else, but it also creates competition which needs a solution, a problem that humans seem incapable of solving. We’re definitely not worthy of induction into the federation of planets.

What more is needed is indeed form. Form includes function, in the case of functional artifacts, and internal teleological organisation, in the case of living organisms. Form being act with respect to matter, as you (and Aristotle) aptly say, is the most fruitful way of conceiving of matter (or material constituents) in terms of their potential participation in the characteristic activities of beings (e.g. computers, houses, amoebas, cities, people) that physical concepts, owing to their very generality (and hence also abstractness), simply miss. This also makes it a mistake to assimilate Aristotle’s matter with the modern idea of material constituents that are matter simpliciter. For Aristotle, ‘matter’ is always relational (planks are the matter of the house, nutrients are the matter of the eye) and what counts as matter always already has some form of its own. The modern physicalist picture, by contrast, treats the lowest level as absolutely material, detached from the relational structure of potentiality and actualisation.

Quite the opposite. My limited aim in the OP, which initially targeted the shared formal structure of Kim’s causal exclusion argument, van Inwagen’s Consequence Argument for incompatibilism, and the “powerlessness” argument for two-boxing in Newcomb’s problem (premised on the idea that the foreseeable action of the agent is an already determinate part of their “past”), was precisely to show that the causal closure of the physical, which is a constitutive principle of this domain as such, limits the scope of the explanations that pertain to this domain in a way that makes room for the autonomous explanatoriness and causal efficacy of substantial forms, as disclosed by the special sciences, the phenomenology of everyday life, and the structure of practical reason.

There does remain a tension that @Count_Timothy_von_Icarus highlights between seemingly incommensurable “paradigms”: (1) domains whereby material beings are disclosed through subsumption under “universal” (so-called) laws and (2) those whereby they are disclosed through species-specific or rational-form-of-life-specific norms. But I think this tension is largely resolved by subsuming the former under the latter, in the way pragmatists (e.g. Hilary Putnam) and new experimentalists (Ian Hacking, Michel Bitbol) do in the philosophy of science. Those anti-foundationalist moves are very hylomorphism-friendly. They give back to even physical entities such as electrons their substantial form, by treating the laws of physics themselves as normatively constituted within determinate experimental practices rather than as brute constraints imposed on formless matter from nowhere. But I’ll say more about this in my response to @Count_Timothy_von_Icarus.

What therefore blocks the reductionist, on my account, isn’t a competing foundation but the scope limitation itself. The reductionist’s own ground, microphysics, is too general, in Wiggins’s sense, to reach the specific determinations that form provides. It applies to everything precisely because it determines almost nothing about what anything is or does.

A key point would be knowing one’s ends as worthy of choice, rather than pursuing them unreflectively, or being wrong about what is worthy of choice. A person who pursues the worse over the better is in a sense constrained and made unfree by ignorance. A truth that lies outside the light of their understanding rules over them.

So, replace terraforming with “decomposing all the planets in the galaxy to turn their materials into small platypus statues.” Such an end is wholly arbitrary, and the unreflective pursuit of it is not meaningful freedom even if it involves some exercise of problem-solving.

Antecedent physical events. My sense organs are causally determined to release neurotransmitters to my brain, which causally determines the electrochemical activity of my neurons, which causally determines the release of neurotransmitters to my muscles, which causally determines my fingers to press down on my laptop’s keys.

You two do realize that you’re not in disagreement, right? Wayfarer is using M1 terms, and Michael is using P1 terms to describe things that are token identical, both of which arrive at P2 which realizes M2.

None of this is incompatible with determinism. Lack of determinism doesn’t buy anything more.

The example (typing some reply) is an example of choice, one which bears responsibility. If I choose to post ad-homs in reply to a poster I don’t like, it’s my fault that I get banned. Determinism has nothing to do with that.

So presuming we carefully reflected on our ends, what might be the ends of the long-term galaxy-infesting probe? One obvious goal is to bring the destination to a point where more such probes can be dispatched. That makes it a life form, subject to natural selection and such.
Without at least that goal, the probe is at best only going to turn one planet into platypus statues.

@Pierre-Normand

Thank you for another very good reply, and I find myself in agreement with most of it. You mentioned the anti-foundationalist strategy of subsuming universal laws under species-specific modes of disclosure. I’m curious, to your mind, does this imply that the reality of form is dependent on these modes of disclosure? I know you said that you will say more about this topic in a forthcoming reply to @Count_Timothy_von_Icarus, so please feel free to address this question as part of that reply if you wish.

1 Like

The example I gave was typing a mathematical proof. That is determined wholly and solely by the requirements of mathematics, not by the musculature that enabled the depression of keys on the keyboard. You could also realize the same outcome by dictating it. In either case the physical means are in service of a purely logical end.

They grant emergent behaviors that are not present in the component parts. These behaviors are real. What more “ontological standing” or “legitimacy” do they need?

While real, emergent behaviors are not casually something in addition to their components. If you could simulate these components, you would simulate the emergent behavior. Remember I gave the example of the game is life, implemented in the game of life? You brushed this off, saying something to the effect that this “only worked in simple mechanistic cases”. What more is micro physics than a complex mechanistic case?

Suppose I gave you a comprehensive physical description of a chess computer, with every transistor and every electrical charge faithfully represented. And suppose you were able to physically simulate this system step by step. Would it not behave as a chess computer? What more does the algorithm provide on top of this description, other than an account of the salient features of this system?

1 Like

An algorithm is a precise, step-by-step set of instructions or rules followed to solve a specific problem, perform a computation, or accomplish a task. Often used in computing to process data, algorithms take inputs, apply defined logic, and produce outputs. They are essential for automating tasks and decision-making.

While the effects of an algorithm can be observed in the physical world, the algorithm itself is distinct from its physical implementation. The relationship is often compared to a recipe versus the actual ingredients and cooking process.

I know what an algorithm is. I was a little lose with my language, I should have said the computer “implements” the chess algorithm.

Right. And it’s a difference that makes a difference, especially in this context.

Mathematics does not itself incentivize a person to type out a proof. There must be utility in doing so, and also capability. Taking a test is an example of such utility, and yea, if you know how to do the proof, you’ll almost certainly regurgitate it on the test when asked. It is in such ways that humans are functionally deterministic, and thus why they can be predicted by scenarios like the Newcomb one (which is essentially a 2-option multiple choice test) without relying on retrocausality or other magic.

On my actual point in the post that you quote, you’re still describing the prior state in M1 terms, which could well be P1 terms which make no mention of proofs, incentives, or muscles. And yet P1 evolves into a subsequent state P2 which realizes the proof having been typed out.

Sure, they may grant that these behaviors are real in some sense, but they’re often quick to add some caveat that makes it clear that it is “nothing more” than microphysics.

You seem to be doing something similar with your appeals to simulation. If a perfect microphysical simulation can reproduce all of the higher level behavior, then that shows that the behavior is really nothing more than microphysics. But I think that this overlooks a key point, which is that, in many cases, there is nothing in the microphysical description that can tell us what the system is.

This is precisely what is provided by the algorithmic level of description in your chess computer example. The microphysical description can answer the question “how was this behavior produced?”, but it can’t answer the question “what is the system doing?” And this isn’t just a practical limitation, it’s a principled one.

This becomes especially clear when we consider the case of malfunction. It’s not just that it’s hard to define a malfunction at the level of microphysical description, it’s that there is no such thing as malfunction at the level of microphysical description. At that level, there are only physical states and transitions between them.

But malfunction is real. If the chess computer crashes in the middle of the game, or if it recommends a nonsense move, it really has malfunctioned. And grasping that fact means grasping something true about the world that isn’t contained in the microphysical description alone. If this is correct, then I think we have good reason for saying that the microphysical description does not exhaust the reality of the chess computer.

I’ll stop here because I’m curious to get your thoughts on this before we go any further down the rabbit hole.