Current location:	home
Current mood:	wordy
Entry tags:	bayes nets, church-turing thesis, counterfactuals, free-will, haphazardism, in itself, michael prospect, necessity, self-consciousness, self-representation, subjectivist probability, supervenience

Free will
DISCLAIMER: This is a disorganized ramble written over the course of several days. By the time I wrote the end, I didn't remember the beginning. I can't vouch for it's coherence as a whole. Or of any of the parts, really.


I'm hesitant to make this post for a number of reasons. One is that when I attempt to articulate what I am going to try to say to myself, I get overwhelmed by a sense that I really ought to stop and read some of the things I've been reading lately, more. Another has to do with a problem of timing; I'm worried that to some extent context might lead what I'm saying and its origin in thought to be misconceived. I need to remind myself that one of the purposes of this journal was to provide a canvas onto which I could force the expression of nascent thoughts so as to see if they survive exposure on the hill.

About two and a half weeks ago I had the privilege of eating lunch with my friend Michael Prospect. M.P. is one of the most virtuous people I've ever met, with humility chief among those virtues. He's honest, astute, hard-working, compassionate. He spends has spent most of his spare time working tirelessly for Providence's homeless community. Next year he's going to go do service work for the Jesuits. He's a pious Catholic, but with a liberation theology bent that drives him to serve the most marginalized.

I was eager to eat lunch with him for a number of reasons, but this was one of them: I was eager to explain to him how I had over the course of the past year become a moral realist. This wasn't the case a couple years ago when we talked more frequently about such things; I cynically dismissed the existence or importance or coherence of most of the ideas that he conceived of as central to his life: love, morality, social justice, etc. "What is social justice?" I'd ask, and he'd sheepishly admit that he didn't know, but still wanted to fight for it anyway. I got the impression from him that he was insecure in his virtue; he was afraid that these ideas that gave his life meaning were grounded in faith and not reason, but that he maintained his faith only to ward off the despair of meaninglessness. So although we were on good terms, I imagined that he was wary of me and the despair-threatening inquiry I represented.

I wanted to tell him that it is possible to get out from the other end of that inquiry with the meaning and value of the world intact. In fact, that inquiry reveals the world to be more richly infused with value than what one could have imagined when one began. I was pretty excited about it.

I don't want to go into fleshing out the kind of moral realism I had in mind in full right now, but it's an outgrowth of the kind of materialism described here a month ago and hinted at the end of that post. The idea (when fleshed out) is that practical reasons and practical rationality are matters of fact, that valuation is something that is inherent in practical rationality, and that by a process of self-creation we can come to realize what is objectively valuable; these values would be themselves just facts. No dichotomies or antinomies anywhere.

I'm marginally less confident about this idea now than I was a month ago, although I still intend to make a couple of posts to real_philosophy about it soon in order to complete what I intended to be a series. I expect that I will have the pleasure of enduring a new set of pejoratives--"bourgeois!", "contemplative!"--in light of recent work done there. I think these accusations would be misplaced, although I need to think about it and read some more.

Anyway, when I told M.P. about this, he had a concern (an uncanny one, in retrospect). I was saying that we could identify virtue in other people, and he was saying that we were unable to judge a person's true virtue, since their virtue depended on their context and capacities. So, one who works hard against strong malicious dispositions to avoid commiting atrocity is in some sense more virtuous than one who finds it easy to be compassionate, generous, intelligent, etc. I was exhausting myself doing mental backflips trying to figure out how this could work out in terms of the psychological models I was familiar with until he said something else which indicated part of the source of the issue, which was his commitment to a particular conception of free will. Some kind of homuncular agent is making free decisions, struggling against the dispositions of his or her psychology.

This made perfect sense coming from him, since as I've said he's a pious Catholic and hence a soulist. He also would have found the principle that man cannot judge his brother compelling. He also needed some foundation for moral responsibility, in God's eyes.

I tried my best here. "Of course there is still choice, in the materialist view," I said. "It's just that choice is just another psychological fact among others. So, presumably, is the difference between hard and easy choices. So why couldn't we just tell whether or not somebody was making virtuous choices just by looking?" But he was not satisfied. In this conception of choice, is there any way they could have chosen otherwise? If not, isn't that an impoverished conception of choice? How could one possibly hang moral responsibility on that?

---

We parted; he had to got to a meeting with the homeless. I was left with something to puzzle over for the next couple of weeks. The problem was deeper than I thought, but I think I've got some solutions in mind. Here was the first line of thought:

"It is not a choice in the suitable sense unless one could have done otherwise." Damn it all! It's counterfactuals rearing their ugly, Lernean heads again.

I've never liked how counterfactual talk gets worked out in contemporary analytic philosophy. "Possible worlds" talk has always seemed silly to me, despite its elegance as a formalism and the fact that I occasionally find myself slipping into it. Modal realism? No way.

On the other hand, my conception of causality is at this point wrapped up heavily in notions of counterfactual dependence, and causality seems to be the thread out of which my understanding of the world is weaved, so I can't run away from it forever. In the meantime, I've been working on a kind of hand-wavy, stopgap "theory" of causality that goes something like this:

Our best models of causal reasoning say that people represent causal relations in causal Bayes nets. As a subject experiencing the world, I'm not going to be able to get beyond my capacities to represent causal relations, so things in the world have causal relations as characterized by causal Bayes nets.

Before I go on: I'm not saying that we some how have causal Bayes net "sense data" that mediate between real objects and ourselves. Rather, when I talk about mental representations (ever) I'm talking about some unconscious mental thing that characterizes how we experience the world.

So...yeah. That was that. The tricky question is, I guess, "Well, are you a realist about causality (as characterized by causal Bayes nets)?" and my answer is "Yes," and people would I guess reasonably say "You're crazy." To which I think my response is, "Actually, anything else would be Cartesian."

Honestly: I have suspected for a long time that I might actually by an "absolute idealist," although I hesitate to associate myself with a position that I know so little about. But regardless of the label: I think I'd comfortably say that the only things that exist in themselves are those that exist for us. (I see no contradiction between this and materialism, which seems to me to be the thesis that those things that exist for us are also things in themselves)

This is a digression, but I feel like saying it anyway. Here's the story I want to tell: We see a dog directly. What are experiencing? The property of being a dog (of course not a mere representation of that property, but the property itself). That's the first-person account. Is there a third-person account of that property? Of course. It is the property of causing in a suitably constituted subject (construed in a third-person way) of the "activation" of a DOG concept token.

The story doesn't have to end here. The "third-person" description is not meant to stand alone and final with any sort of metaphysical privilege. I actually think this is kind of an improper reading of the term. "Third-person" (and "first-person", etc.) are not metaphysical properties. I don't even think they are standpoints. They are, originally, voices, in the sense of ways of describing things. (Or, working back upstream away from linguistic commitments and into purely mentalistic ones, of conceptualizing things) So one could presumably go on unpacking the property into further levels of ideation. (Identifying it with the property of causing the activation of a complex conceptualization, A-PROPERTY-CAUSING-A-SUITABLY-CONSTITUTED-SUBJECT-TO-ACTIVATE-HIS-DOG-CONCEPT, in a third, suitably constituted subject.) Throughout this I'm conflating modes of presentation with properties themselves and I think this raises some problems (deep ones; I think they are crucial ones for conceptual role semantics...) but I think they are surmountable.

Where was I? Oh, causality. Well, do the same thing above with dogs and DOGs but replace it with causal relations governing the proper functioning of a 6-cylinder engine and [CAUSAL BAYES NET MENTAL REPRESENTATION OF A 6-CYLINDER ENGINE] and I think you'll get an idea of what I'm trying to get at (if any of this is clear, at all).

I'm not sure how that helps with the counterfactuals. Maybe what I want to say is this: there is a first-person notion of the counterfactual properties of some object which can be cashed out into a third-person account which does not depend on those counterfactual properties being "real" in the sense of these counterfactual situations really existing. On the other hand, this description (no doubt involving causal descriptions) will have its own counterfactual implications (which are just as eliminable).

I don't know if this counts as a paradox, or a duality, or a mere tension, or what.

---

That was just one way I've been thinking over the past couple of weeks. Here's another:

We (I says to myself) are comfortably subjectivist about probability, since we are good Bayesians. So we ascribe probability to "possible worlds" (ugh) based on how likely we think they are to be actual. Certainly we can get a subjectivist position about possibility out of this as well. It is possible if we assign a non-zero probability to it (under certain conditions of rationality).

I stop me! I'm talking about epistemic possibility here. Don't you need to tackle metaphysical possibility? I'll get to it.

So, non-zero subjective probability implies subjective (epistemic) possibility. As good Putnam pragmatists, we are fallibilists, and so are going to assign non-zero subjective probability to all possible worlds. So everything is (epistemically) possible!

Does that matter for any of this?

I think so. Because talk about metaphysical modality seems to me to be parasitic on the God's-eye-view. This was actually my first reaction to M.P.'s problem. In what sense can we ask "Could he have chosen any differently?" without depending on a literal God's eye that sees the agent as operating on its own independent power.

I've never really understood what "metaphysical necessity" means. The only way I can think to cash it out that makes sense is that it provides structure for the relative frequency of various hypothesized "possible worlds." So, under the condition of the identity of A and B, some hypotheses ....

mental block. I apologize. As I will, some time in the future, write as a disclaimer to this post as a preface, this is being written incautiously over the course of several days with evolving and receding thoughts on the matter. So it's unstructured, and it's blather.

Anyway, I forget what I was trying to say about metaphysical necessity. I think it hinges on a kind of metaphysical realism that I am happy to dismiss at this point. One reason why I am happy to dismiss it is because I've finally picked up Putnam's The Three-fold Cord, again, and he nails what he is calling "metaphysical realism" (which I think might be the same thing as "Trancendental Realism," he certainly believes that it is tied to Cartesianism about "appearances") a lot, and hard.

Not that that necessarily has anything to do with what I'm talking about. What was that? Free will I think.

---

Note to self: How does one reconcile Bayesian rationality with Putnam's version of James' "natural realism (of the common man)" (i.e. rebranded "direct realism")? It would seem to depend on our being able to see objects as all the things they may potentially be--probabilistically weighted.

Note to self (2): Does the causal Bayes net formalism still "work" when the probabilities are all subjective probabilities (recall that there is nothing Bayesian (in terms of Bayesian inference) about Bayes nets except, literally, the name)?

---

One way of framing the free will problem: Is our behavior driven by necessity?

No. Here the (recent, right?) innovation of probabilistic causation is important. Today's natural laws preclude causal necessity. (Quantum irrationality, rationalized by subjective probabilistic law) And probabilistic causation is within the purview of our subjective theorizing (although it takes a while to mature--children's causal reasoning is deterministic, as a rule.)

---

This isn't satisfying though. Another way to frame the question is: does our willing make a causal difference?

Yes, of course. As a component in a causal chain, the alteration of the Will (construed as a psychological object) will of course cause different effects.

But this can't be an answer to the question of free will, can it? The efficacy of the Will was never in question; it was it's freedom.

Back it up. We can ask: is the Will determined by it's antecedent causes? The answer has to be "no," given the probabilistic nature of the substrate. But "haphazardism" is just as unsettling as "determinism." To the extent that the Will is caused, is it caused by its antecedent causes? I think so, yeah. Dammit!

But wait. Here's something that's lacking here: the ability to account for strongly supervening properties within the Bayes net formalism, and the causal relations between them. In particular, if I define some event E over some of the events whose probabilities are governed by a causal Bayes net, and some other event F, I could, I suppose, get some probability relation between them. But what if I intervene on E? What happens to the states of the events in the "substrate"?

This seems to be a question that would come up here. From the first-person, the Will appears to itself as intervening on the world, but these are high-order properties that are supervenient on stuff below. What is going on here?

Problem: much of our first-person "experience" of our own "free" willing is demonstrably wrong. People systematically deceive themselves as to what sorts of priming effects, for example, are influence their behavior. Introspection on cognition is far from infallible; rather, it is as often as speculative as inspection of the cognition of the other.

However, I think this might get to the subjective core of the free will problem, since the fear of determinism may be the fear of being predicted.

---

So let's ask: can I predict the actions of the other (up to quantum limits)? If so, can I predict the actions of myself?

If we could effectively predict all of our activity, that would be the kind of downer determinism promises to be.

Here is my reaction to this. It's convoluted:

Putnam (in Three-fold Cord) gives a nice genealogy of the idea of the "in-itself." It was alien to, say, Aristotelian thinking (he says), which was more in line with natural realism, according to which the objects of our experience are real objects themselves, not just appearances. However, when people were just starting to mathematize nature, things like colors didn't fit into the picture. So they became secondary properties, distinct from what things were in-themselves. Eventually, the in-itself split off entirely into the unknowable noumena.

But all this happened before formal psychology, which is to say, before anybody thought to mathematize the subject. Now that we have, we have a much better idea (I think) of how colors can be properties of mathematizable objects in-themselves: the property of 'red' is the property of causing the appropriate psychological events to occur. Or something. The point is, the unknowable in-itself was a wrong turn taken by philosophy when science was incomplete in an important way--it hadn't been turned onto the subject itself.

That doesn't mean we can know everything. The in-itself is knowable (it's just the for-us, after all), but it seems like it's impossible to know the totality The reason is that as subjects we are bound by computational limits.*

* I'm assuming the Strong Church-Turing thesis, and it's applicability to human subjects. I think this is safe. Although I've never seen it proven, I think it would be easy to show that Bayes nets, construed as a model of computation, would be Turing-equivalent. And since human psychology, as an object defined by its causal relations, should fit into that model....

So we can't compute everything. But can we, as subjects, compute predictions of ourselves?

I don't think so. Here's why: One of two things may hold,

(A) Subject S does not have a complete internal representation of S (characterized by its causal roles, because we're functionalists), or

(B) Subject S does have a complete internal representation of S.

If (A), it's hard to see how S could have perfect ability to predict their own actions.

If (B), well...I'm not sure it's possible. But if it is, then I think that S's ability to predicting it's own actions would be tantamount to the Halting Problem (and hence, impossible).

Here's the thing: if (A) is the case, then nothing precludes some other subject from knowing S's actions to their heart's content.

If (B), then I'm not sure about this, but I think it might be that nobody could predict S's behavior. (Because the complete self-representation would be a recursive one, with no base case. Infinite computational loops).

This leads to a claim worthy of the Continent: "Only when the subject becomes fully conscious of itself as object does it become free."

All of the above is just a sketch, of course. Incidentally, Douglas Hofstadter has recently reappeared on my consciousness. But crucial here is that there isn't just self-reference going on here--which would merely involve, like normal conceptual reference, a mental representation that only partially captures its object (n.b.: this is not to preclude a phenomenological analysis of one face [I forget the technical word] of the object being experienced directly), but rather a "complete" representation; I'm not really sure what that would entail. Objectively, it would require a recursively structure subject, I think, with the the representation being a microcosm isomorphic to the whole. I have no idea what that would be like, from a first-person perspective.

Actually, I have a guess. Not a great guess, but it seems like a plausible candidate at least. The conscious states allegedly reached through meditative practice are probably quite strange computationally. The problem though, here, is that phenomenologically they are apparently associated with the absence of a representation of the subject. However, this may be be a red herring (since in the general case, the objective computational representation of a thing plays out in the first person as just the experienced contact with the thing itself. I should not let myself be seduced by the aesthetic of this, however.

---

Ok, I was going to say more. I have a lot of reactions to the Lukacs stuff, for example, especially since I can't help but get the impression that I'm a target of it. I think much of this post might be relevant, but words fail me. I'm done now.