Key Themes in the Free Will Debate

Reference Material:
Intuition Pumps and Other Tools for Thinking – Daniel Dennett
An Essay on Free Will – Peter van Inwagen
Freedom and Belief – Galen Strawson

        I mentioned at the end of Lotteries and Chess Engines that I’m starting to sound like a compatibilist, something I would not have predicted when That Doesn’t Follow started up more than 2 years ago. This got me thinking – how best to characterize the view (regarding free will) developed on this blog? This post is my humble attempt to address this question. I’ll do this by highlighting 5 themes (claims, distinctions, ideas, or questions) that have cropped up throughout my reading and writing on the topic, and that together nicely summarize my views.

1. The Trouble with Libertarianism
        Libertarianism doesn’t make sense. I won’t go so far as to charge it with internal logical inconsistency (gasp!), but… actually yes, that is what I’m saying. As an intro into my thinking here, start back at the beginning by asking what we mean by free will. To illustrate my point, let’s work with the ever-present freedom to do otherwise conception. The conditional formulation always seems too shallow, too… conditional. You’re free to do otherwise if, were you to choose to do otherwise, you’d in fact do otherwise. But I want to know if my choices are free, not just my actions. OK, you’re free to choose otherwise if, were you to desire to choose otherwise, you’d in fact choose otherwise. Hold on, for me to be truly free, my desires need to be free too. Free desires are what I want, so the definition of free will should capture this (more on this truly, “want”, and “should” later). Alright, you’re free to desire otherwise if, were you to desire to desire otherwise… and we’ve reached infinite regress territory. This seems to be getting us nowhere – let’s move on for now.
        How about the categorical formulation? You’re free to do otherwise at a given time if it’s possible, holding everything constant up until that time, for you to do otherwise. Now, subject to a very important caveat (the notion of “possible” being used – more on this later as well), I think it’s obvious that this formulation of free will is incompatible with determinism. The result is something like van Inwagen’s consequence argument, or Strawson’s formulation, if you prefer. The problem is this. For you to be free, the formulation requires that when you act, you somehow step outside of or get behind the infinite chain of cause and effect that determinism implies. But this is impossible, and van Inwagen’s and Strawson’s arguments follow. And note that indeterminism can’t swoop in and save the day – a randomly firing neuron gives you no more control or agency than a deterministic one (I won’t elaborate this point here – see Freedom’s Debt to Reason for a summary, or Freedom and Belief, pgs. 7-8 for an extended discussion).
        And now we can see the reason behind our failure to produce a satisfactory conditional formulation of the ability to do otherwise. By shifting the analysis from action to choice to desire to meta-desire, we were trying to formulate a conception of our freedom that places us outside the causal structure of the world. We were trying to accomplish the impossible.
        And this is the trouble with libertarianism. It seeks a conception of free will that describes human action as ex nihilo, outside the chain of cause and effect, a source of causality unto itself. Pick your poison – this is nonsensical. In The Picture So Far, I wrote – “broadly speaking, my sympathies lie with the impossibilists – at the end of the day, the [free will] concept just doesn’t seem coherent.” I still agree, but the story doesn’t end there. The libertarian notion of freedom is incoherent, but what are the alternatives? The other themes will help explore them.

2. Two Kinds of “Could”
        The phrase “could have done otherwise” is all over the place in the free will debate. Understanding more about what “could” means is crucial to moving beyond the libertarian quagmire. When you use the word “could”, you invoke some notion of possibility. Regarding the free will debate, I think one distinction is central – metaphysical vs. epistemic possibility. Metaphysical possibility is what you might casually call real possibility (although this wording quickly gets misleading). I’ll use it here as the possibility that exists, given the laws of nature and the initial conditions of the universe. It’s also referred to as nomological possibility, but it’s important to my conception to also fix the initial conditions. Epistemic possibility is what’s possible based on what you know. Roughly speaking, something is epistemically possible if it might happen, you just don’t know if it will.
        Stated plainly – if determinism is true and there is no strong emergence, then metaphysical possibility doesn’t exist. Again, you can throw in the indeterminism of random initial conditions, radioactive decay, wave function collapse, etc., but it’s just not going to make a difference to the free will debate (see Freedom’s Debt to Reason for an extended discussion of both these points). So I think it’s safe, and simpler, to move forward with the discussion assuming that there is no metaphysical possibility. Of course, our world is full of epistemic possibility. A nice simple example to illustrate the distinction here is just an ordinary dice throw. There is no metaphysical possibility in this situation. Knowing the initial position, momentum, and angular momentum of the dice, their precise shape and material characteristics, and the geometry and properties of the surface they’re landing on, you could (in principle) predict the result. It’s fixed once you throw the dice. Of course, you don’t know any of these quantities to anywhere near the required precision, so from your epistemic perspective, the result is random. No metaphysical possibility, maximal epistemic possibility.
        What does this distinction have to do with free will? Well, there is good reason to think that the kind of possibility relevant to the free will debate is actually epistemic possibility. Dennett has a couple good intuition pumps to motivate this idea. We think lotteries are fair, in that everyone has the same chance of winning, even when the only “chance” we have is epistemic. We can sensibly talk about whether a chess engine “could have done otherwise” in a game, where the “could” has to do with the possible worlds enumerated over our epistemic ignorance, not metaphysical possibility.
        This distinction can then be used to disarm a classic free will thought experiment. In Frankfurt cases, we are led to believe the protagonist could not have done otherwise but nonetheless is intuitively responsible for, and hence free in, their choice. The conclusion is that the freedom to do otherwise conception fails to adequately capture what we mean by free will. But this conclusion relies on interpreting the “could” in terms of metaphysical possibility. If there is no metaphysical possibility, why interpret it that way? What if we interpret it in epistemic terms? You didn’t know about the device. From an epistemic perspective, your decision was as wide open as any – you could have done otherwise. Or consider Dennett’s twist (400). You intentionally arrange for the device to be implanted, aiming to shape your behavior long-term. But you soon forget it’s in there. Again, from an epistemic perspective, you have the ability to do otherwise.

Detour – Newcomb’s Paradox
        At this point I’d like to take a detour and revisit a classic thought experiment that touches on themes 1 and 2. I’ve discussed Newcomb’s Paradox in detail previously. My basic views here haven’t changed, but I wanted to highlight how the ideas from the first two themes are involved. I was surprised recently to see that roughly 2 in 5 academic philosophers are inclined to “2-box”, while just 1 in 3 advocate “1-boxing” (see the 2020 PhilPapers survey). While I can’t say I have a complete picture of the nature of all the disagreements here, I think part of what’s going on has to do with the free will argument.
        Let’s take a look at the SEP article on causal decision theory. In section 2.2, we’re told that “the agent’s act does not cause the prediction” and “the act the agent performs does not affect any prediction’s probability because the prediction occurs prior to the act.” It’s true that the agent’s act doesn’t cause the prediction metaphysically. Metaphysical (“real”) causation only occurs in one temporal direction, forward. But there is a sense in which, from your epistemic perspective, your choice does cause the prediction. You know the prediction will be correct, so if you 2-box, the $1M won’t be there, but if you 1-box, it will be. This apparent retroactive causation, from your epistemic perspective, is possible because both the prediction and your choice have the same metaphysical cause (the predictor must be a simulation of your brain to always produce correct predictions). The failure to step out of the metaphysical framing and take the agent’s epistemic perspective results in the preemptive ruling out of (apparent) retroactive causation.
        A longer excerpt:
Some theorists hold that one-boxing is plainly rational if the prediction is completely reliable. They maintain that if the prediction is certainly accurate, then the choice reduces to taking $1M or taking $1T. This view oversimplifies. If an agent one-boxes, then that act is certain to yield $1M. However, the agent still would have done better by taking both boxes. Dominance still recommends two-boxing. Making the prediction certain to be accurate does not change the character of the problem.

How is it, exactly, that “the agent still would have done better by taking both boxes”? 
This seems to rely implicitly on the (metaphysical) possibility that two states of affairs pertain simultaneously: there being $1M in box B, and the agent 2-boxing. But this is (again, metaphysically) impossible, if the predictor is perfectly accurate. If you 1-box, you get $1M. The only way to improve upon this outcome is to “pseudo-1-box” (whatever that means), causing the predictor to predict 1-boxing and put 1$M in box B, and then, by some magic force of libertarian free will, choose both boxes. Short of this, there is no sense in which “the agent would have done better by taking both boxes”.
        Last one:
In Newcomb’s problem, assuming that the agent is rational, his beliefs and desires are a common cause of his choice and the prediction. So his choice is a sign of the prediction’s content. For the probability of a prediction of one-boxing, knowing one’s beliefs and desires makes superfluous knowing the choice that they yield. Knowledge of the common cause screens off evidence that the choice provides about the prediction. Hence, the probability of a prediction of one-boxing is constant with respect to one’s choice, and maximization of evidential expected-utility agrees with the principle of dominance. 

This is the most confusing one, because there’s a lot that’s correct here. It’s true that the agent’s choice and the prediction have a common cause, the precise brain state of the agent as they pass through the predictor’s brain scanner (“his beliefs and desires”). The problem is when we start talking about probabilities of prediction. I think the claim that “knowledge of the common cause screens off evidence that the choice provides about the prediction” is similar to thinking that “the act the agent performs does not affect any prediction’s probability because the prediction occurs prior to the act” (from Section 2.2). The fear of spooky retroactive causality is behind both.
        To get past this fear, approach the problem this way. For the predictor to always be correct, the common cause of the agent’s choice and the prediction has to be a deterministic process. For deterministic processes, metaphysical possibility doesn’t exist (see Freedom’s Debt to Reason). But the notion of probability must refer to some type of possibility, and the only type available is epistemic. The probability of a 1-box or 2-box prediction is an epistemic entity, not a metaphysical one. Hence, it can be subject to apparent retroactive causality without any spookiness or magic involved.
        More concretely, Newcomb’s paradox forces you to accept that your choice is already metaphysically determined. It does this by placing constraints on the metaphysical possibilities of the outcomes of your actions. What you know is that you live in 1 of 2 possible worlds: 1-box world (prediction = “1-box”, $1M in box B, you 1-box) or 2-box world (prediction = “2-box”, box B is empty, you 2-box). You know you live 1-box world or 2-box world, you just don’t know which. The upshot – you get to choose! (epistemically, of course). This is the kind of free will you have. Is it enough?

3. Free Will Worth Wanting
        So we’ve seen we don’t have libertarian free will. We can’t – it’s incoherent. But we do have some kind of freedom, something associated with epistemic possibility. Whatever allows us to take $1M (but not $1M and $1T) from Newcomb’s paradox – that’s what we have. So where do we go from here? Dennett proposes a new perspective (I’m unsure if this way of approaching the issues is originally due to him, but I first found them through his work). What kind of free will do we want? And what kind of free will should we want?
        Let’s return to Strawson’s argument for the impossibility of free will. It goes something like:
1) No one is ultimately responsible for how they are (e.g. their character and motivation).
2) People act the way they do because of how they are.
3) Therefore, no one is ultimately responsible for acting the way they do.
4) To be free is to be ultimately responsible, but we aren’t ultimately responsible, so we aren’t free. 

Now again, I find this argument persuasive, as far as it goes (and Dennett agrees below). It’s basically another way of stating that, assuming determinism, metaphysical possibility doesn’t exist. But Dennett has an interesting response – so what? Why do you want to be ultimately responsible? If ultimate responsibility, with its requirement that you be causa sui, is impossible, why do you care? He’s worth quoting at length on this (394-5):
Absolute responsibility is a red herring, a blessing nobody should hanker for. Strawson (2003) thinks otherwise, and criticizes me for neglecting it:  
He doesn’t establish the kind of absolute free will and moral responsibility that most people want to believe in and do believe in. That can’t be done, and he knows it. 
He’s exactly right: I don’t establish the kind of free will most people want to believe in, and I know it. But I think they are wrong to want to believe in it, and wrong to believe in it if they do. The burden falls on Strawson and others to show why we ought to care about ultimate responsibility – or the determinism/indeterminism issue – in our lives. They can define a variety of free will that is incompatible with determinism, and show that lots of folks think that’s important, but they also need to show that those folks aren’t deceiving themselves. Why should anybody care?

I think they are both right to observe that most people walk around with a more or less libertarian notion of free will (even some causal decision theorists do, apparently). But libertarianism is incoherent, so let’s conclude – many people, much of the time, are wrong about free will. What does this mean, and what should we do about it? That’s where Strawson and Dennett seem to disagree. Strawson seems committed to analyzing free will using the everyday, libertarian conception. Dennett is less concerned with the folk notion – he wants to move forward with an amended conception that doesn’t require our decisions to be “little miracles in the brain”. There’s an aspect of this debate that’s methodological, which is where I want to turn next.

4. Reconciling Manifest and Scientific Images
        What should philosophers do with terms like “free will”? Should we (casually, hubristically including myself) take these concepts, as they are commonly used, at face value? Or should we feel free to tweak and mold the definitions as we see fit? I must admit I’d never thought about this issue before encountering it in Dennett. I like his approach. The following won’t be fair to Strawson – I don’t know how he thinks about these kinds of methodological questions. But I really like one of the examples Dennett used to motivate his position.
        First, an important distinction (of course), originally due to Sellars – the manifest image vs. the scientific image. The manifest image, according to Dennett, is “the world as it seems to us in everyday life, full of solid objects, colors and smells and tastes, voices and shadows, plants and animals, and people and all their stuff”, but also intangible and abstract entities like money, love, and yes, free will. By contrast, the scientific image contains “molecules, atoms, electrons, and quarks and their ilk” – basically, the scientific description of the world (69). The classic framing of free will vs. determinism is one of those philosophical issues that arises when we try to reconcile the manifest image (we all have this sense that we’re free) with the scientific image (chains of deterministic causality). How might we go about this?
        To orient our thinking here, let’s bring in the example, offered as an argument by analogy. Consider the property of solidity. What is solidity, according to the manifest image? Well, an object or surface is solid if it’s strong (not flimsy), firm or stable (not fluid or deformable), and importantly, not hollow, and without holes or gaps. In particular, you would never consider an object solid if it was mostly empty space. But solid objects, when described by the scientific image, seem to be just that. They’re all made of atoms, and atoms are mostly empty space (the nucleus and electrons are much, much smaller than the atom as a whole). What a discovery! Everything you think is solid is not actually, really solid.
        But this is, of course, quite absurd. If you think nothing is ever really solid, try running through that wall over there. The issue arises because a part of the manifest image of solidity, the absence of holes, gaps, or empty space, seems incompatible with the scientific image of solidity. This portion of the manifest image is what Dennett calls a folk ideology of the manifest image. Now consider what we actually care about with respect to this folk ideology. The reason I want solid things to lack holes, gaps, or empty space is because when I call, for example, a wall solid, part of what I mean is that: if I throw a ball at it, it bounces back; if I dowse it with water, it splashes off; if I throw a dart at it, it doesn’t pass straight through. If the wall has holes on length scales of mm, cm, or meters, some or all of these conditions won’t be met. But notice that my functional conditions only require the wall to lack holes, gaps, or empty space on more or less human length scales. This neither implies nor requires anything at the length scales characteristic of the scientific image (nm and smaller). Instead, what we see is that, at the microscopic level, solidity just isn’t what you thought it was. It’s not that nothing is ever really solid. It’s just that the microscopic constituents of solid objects aren’t what you might naively think they must be.
        Let’s return to free will, and just make the analogy explicit. The manifest image of free will is simply that we all feel free (solid objects seem solid). The folk ideology of this manifest image is commonplace libertarianism – our freedom must involve metaphysical possibility, human action must be causa sui, our decisions must be “little miracles in the brain”, etc. (surely for an object to be solid, it must have no holes, gaps, or empty spaces). The relevant scientific image is that at a fundamental level, everything in the world happens according to deterministic causal laws, with no metaphysical possibility and surely no room for miracles (solid objects are made of atoms, and atoms are, in some sense, mostly empty space). So it seems the scientific image has shown us that, deep down, we can’t really be free (nothing’s ever really solid). Now, the resolution of this apparent inconsistency, in the case of solidity, is the following. We think about what we really want solidity to mean, or what it does mean practically or functionally (w.r.t. holes, gaps, and empty space). When we recast the definition to reflect this, we see that it does not contradict the scientific image after all. Rather, we learn that solidity, at the fundamental, microscopic level, just isn’t what we naively thought it had to be. Resolving the tension in the free will case means we need to return to theme 3 and think more carefully about what we want.

5. The Moral Dog and His Free Tail
        Why does free will matter? What has this endless back-and-forth argument been for anyway? Why do we care? For any single person, there might be a variety of answers. But I think the reason for the sheer persistence of the debate boils down to one thing – we care about free will because we care about moral responsibility and all the related issues (blame/praise, punishment/reward, desert, and the like). Without the (alleged) connection between free will and these basic normative concepts, the free will debate might not matter at all. Though I’m open to alternative explanations for why free will matters, I find this claim safe enough to move forward provisionally.
        The centrality of moral responsibility to the importance of free will has immediate and direct implications for the discussions above, and the free will debate in general. Returning to the discussion in theme 1 – let’s reconsider the definition of free will, except this time, let’s heed the lessons from theme 5. The solidity example suggests that we recast the definition of free will to reflect what we really care about, moral responsibility. The provisional result is something akin to Strawson’s circular definition – free will is whatever freedom is required for moral responsibility (see Freedom and Belief). In suggesting this definitional move as perhaps the best path forward, I’ve come full circle after questioning its legitimacy (Frankfurt Cases), questioning my questioning (Recurring Themes), and teetering on acceptance (Freedom and Belief).
        Now, what freedom is required for moral responsibility? How should we characterize this freedom? Note that these are inherently normative questions, or at least questions with an inherently normative aspect. To answer them in any complete or satisfying way requires metaethical, ethical, and political principles. But that is exactly my point. Free will should be defined in terms of moral responsibility, because moral responsibility is ultimately what we care about. Who has moral responsibility? In what situations? The answers to these questions will constrain your account of free will. The moral dog should wag his free tail, not the other way around.
        Here’s an example we can use to help illustrate my point. Consider two cases. Case 1 – your ex-wife is happily remarried, while you’re bitter and lonely. You murder her out of jealousy in a calculated, premeditated attack. Case 2 – an (as yet undetected) brain tumor critically impairs your judgement and makes you impulsive, careless, and quick to anger. You mow over a pedestrian on the sidewalk while driving, killing him. The brain tumor is discovered and removed. Your behavior returns to normal. Scientists assure us its effects are gone, and it will not grow back.
        From a moral perspective, do we want to differentiate between these two cases in any meaningful sense? You can answer “no”, but then you’re going to have to address a lot of very thorny pragmatic issues about how to order society. Best of luck. I think most people do want to draw some sort of moral distinction here. For both cases, consider – should you go to prison? I’m guessing almost everyone answers yes for case 1, and I’m betting many, perhaps most, would vote no for case 2 (though some sort of restorative justice process might be appropriate). Our justification of such a distinction is going to end up leaning on, and hence constraining, a concept of human freedom. For example – the ex-wife murderer was free, or as free as any human ever gets. His action is at least somewhat representative of who he is, and is indicative of his capacity to attempt a similar act in the future. The manslaughterer beset by a brain tumor was not free, at least not in the normal way most everyone is. Her normal human faculties of reason, judgement, restraint, etc. were critically impaired. Her action was completely unrepresentative of who she is, and she will almost assuredly not reoffend.
        I’ll make a couple observations at this point. First, note that in both cases, each person was metaphysically determined to do what they did (assuming determinism, no strong emergence, etc.). So if we tie the concept of freedom to metaphysical possibility, it provides no leverage to differentiate between the cases. If we want freedom to be a helpful concept in this case, it must be defined in some other way, referencing some other kind of possibility. Theme 2 points us toward epistemic possibility as a, uh… possibility.
        Second, from a metaethical point of view, note that the approach I’ve outlined is entirely compatible with the error theoretic perspective I’ve espoused before. We’re admitting that moral principles can’t be (entirely, at least) delineated through some empirical investigation of the metaphysics of human freedom. They don’t exist “out there” in the world, waiting for us to discover them – we have to make them. In this way, my approach actually might be more suggestive of a kind of constructivism, though this is something for future investigation.
        And third, the view I’ve developed here probably plays well with the free will skeptics’ push to rid our justice system of its more retributive aspects and base punishment on more forward-looking, consequentialist considerations (see this commentary). Retributive punishment is often inextricably tangled up with libertarian notions of freedom and basic or ultimate responsibility. Theme 1 suggests these notions are incoherent, impugning retributive punishment as a practice.

        Let’s return to the original question at the top of this post – how should I characterize the view of free will I’ve developed? There’s a way in which these discussions can deteriorate into purely linguistic inquests (which -ism do you hold? which -ist are you?), but I think there’s some value in roughly locating yourself on the philosophical landscape. The views on this blog lie somewhere near the nexus of:
  • incompatibilism (the common, everyday libertarian notion of free will is incompatible with determinism),
  • skepticism/impossibilism (more fundamentally, libertarian free will is incoherent),
  • pragmatism (there are alternative conceptions of freedom that we probably want to preserve), and
  • compatibilism (such conceptions can be independent of determinism by being directly tied to moral responsibility).
At the risk of a linguistic inquest, I’ll resist the urge to further categorize. To emphasize the pragmatic side of my view, let’s close with two of my favorite quotes in this area:

“One cannot decide not to decide anything on the grounds that one cannot decide anything.” – Galen Strawson

“Sometimes we’re responsible for things not because they’re our fault, but because we’re the only ones who can change them.” – Lisa Feldman Barrett