An Adapted and Extended Uni Essay on the notorious Newcomb's Problem

An Atypical Defence of One-boxing in Newcomb’s Problem

I am an unashamed, incorrigible one-boxer, and I think you should be too. My arguments for this position are not typical, though they are not completely original either. All of them emerge from my reckoning that questions of free will are at the heart of the problem. For reasons that will become clear, it is the arguments of three computer scientists that form the backbone of my attack: Scott Aaronson, Radford Neal and Seth Lloyd. The first two of this trio show a way of re-imagining Newcomb’s problem that simultaneously: a) accounts for the “common cause” dilemma, b) shows why one-boxers don’t have to reject Causal Decision Theory, and c) repudiates the claim that the fully deterministic one-box argument is assuming “backwards causation”. The third of these has written a paper proving a computational/philosophical notion he calls the “Theorem of Human Unpredictability”, which I believe shows that anyone who tries to apply CDT to the imagined ‘post-Prediction’ stage of Newcomb’s problem (what Burgess calls “the second stage”), without reformulating the problem in the manner of Aaronson and Neal, cannot have the slightest clue what the requisite ‘subjunctive conditional’ probabilities actually are, and thus can’t legitimately apply CDT.[1]

In order to stimulate thoughts about free will, let us begin by means of a second-person vignette – imagining what it would be like to carry out Newcomb’s problem in real-life:

It is 11 AM on a Monday, and you are walking into a room, alone. (To add mystery and intrigue, we can say it’s a dark and moody room, with red carpet, like an old cinema.) The room is empty up until the back wall, where there sit two futuristic-looking boxes (I imagine them placed on identical and symmetrically positioned black pedestals). One is transparent and contains $1000. The other is opaque. For whatever reason, you know (for certain) that the opaque box either contains $1,000,000 or $0, and that whichever sum it contains has been chosen on the basis of a super accurate prediction of what you will choose. You know that you are only allowed to choose either to take the transparent and opaque boxes together, or the opaque box alone. You know that if the prediction has been made for the former choice, there will almost certainly be nothing in the opaque box if you take it, and you know that if the prediction has been made for the latter choice, there will almost certainly be $1,000,000 in the opaque box if you take it.  You don’t know how this super accurate prediction really works, but you know that it is super accurate. To be precise, you know that it predicts correctly which box people pick 99.9% of the time.

This is what you are thinking as you near the boxes:

“If the prediction is really 99.9% accurate, then there is only a 0.1% chance that it will not get things right, which means that if I choose both boxes, then it is 99.9% certain that there will be nothing in the opaque box and if I choose the one opaque box, then it is 99.9% certain that there will be $1,000,000 in this box, which means I should obviously choose the one opaque box. Wait no, none of this makes any sense. The sums of money that are in the boxes have already been decided; there is no backwards causation, the prediction has already been made. I can’t be 99.9% certain that if I choose both boxes there will be nothing in the opaque box and that if I choose the one opaque box there will be $1,000,000 in this box. But, hang on, if that is not how this thing is working, then how can the Predictor be 99.9% accurate? Am I really not freely deliberating right now? Are my thoughts not reflecting some kind of spontaneous process of reasoning? How can I really not be making an unpredictable choice here? I mean… I don’t know. I still don’t even know which box I’m going to pick. And I could do something really crazy. No, actually, it would probably be too predictably crazy to do something crazy. Actually, maybe it is too predictable of me to think that kind of self-reflexive thought, and devolve into ever-increasing levels of bluffing. Onebox-twoboxes-onebox-twoboxes-onebox-twoboxes.

You know, I could base my decision on some really random aspect of my personal biography – like the number of letters in the name of my year 2 teacher, Mrs Evans. It would make sense if I two-boxed for an even number of letters and one-boxed for an odd number of letters. But, actually, I don’t know which it is in her case: with the appellation, her name is 8 letters long, but without it, it’s 5. Do I say it’s 5 or 8? One box or two box? Shit. Maybe I should just use the number of letters in some other person’s name, or the name of the campsite we used to go to when I was a kid. Maybe I should use the number of letters in some species of grass – you know, the Latin name or something. Actually, no – what if I still use Mrs Evans’ name, except do two-boxing for the odd number and one-boxing for the even? Only problem is that I still don’t know whether to include the appellation or not.

You know, one good thing about thoughts is how little sense they all make – I reckon I’m being totally unpredictable in having them. They seem unpredictable, don’t they? But probably “seem” is perhaps the key word here. And maybe I’m going too far with the lack of sense. Maybe the prediction had enough information on me to know that that’s the kind of thing I would do, so maybe I should do something that just makes sense in a straightforward way – but, then again, presumably the prediction would have taken into account my agonising vacillations and recursive self-looping and meta-entanglement. So I really don’t know. I really don’t know. Ah, I guess, I’ll just take the one box – no two boxes – no one box – no two boxes – yes I’ll take two boxes – no, I’ll take one.”

You take one. When you leave the room, you open the opaque box: it is stuffed with wads of cash – $1,000,000. The prediction was a success, just as an outside observer would expect. You are not an outlier, you are not unusual. With a feeling of immense confusion and disquiet, you suddenly realise that if you had taken two boxes, you would have got $1,001,000. This seems deeply strange. Beating the prediction would have made you a total outlier (the first one to fail in the last 1000 enactments), and yet you feel as if you were, at multiple moments, totally on the verge of picking the two boxes. In other words, you feel that you were totally on the verge of becoming an outlier. You feel as if your decisions in that room were totally unpredictable – that your mind was running wild, in a way that was profoundly organic and spontaneous and weird. Were they not? You don’t know what to think. This is really strange.

I think that this story really brings out what I believe to be the core of Newcomb’s problem: the age-old question of ‘free will’. As this way of presenting the problem highlights, there is a deep weirdness in Newcomb’s problem. Like all people, I have a conviction that, at least when I am thinking in a very self-conscious way (listening to my own interior monologue) there is a profound spontaneity, unpredictability and – dare I say – freedom about my thoughts (and the actions they lead to). Bizarre things can happen: I can think things like, “I’m going to raise my arm, I’m going to raise my arm, I’m going to raise my arm” and then do nothing for three seconds and then lower my arm slightly, and then thrust it back upwards. These actions make no sense – not even to me. In fact, on such occasions, I feel that I can’t even predict my own actions.

Newcomb’s problem is largely complicated by the fact that we can so easily imagine someone to be thinking in just such a self-conscious way as they approach the decision between one-boxing and two-boxing. We can easily imagine someone agonising over the clash between Causal Decision Theory and Evidential Decision Theory, trying to sort out their thoughts about what has made the Predictor so successful in the past, and generally unravelling into an increasingly recursive, self-referential and involute thought-vortex (until they are at last forced to just make a choice). And yet, no matter how complicated and self-referential these thoughts get (and no matter if the decision-maker ends up making their decision based on something totally random, like the number of letters in the name of their second grade teacher) we are forced by the set-up of the problem to think that their final action will have been fully predictable in advance. This, in turn, seems to force us to regard our thoughts as we think about the problem – no matter how convoluted, self-referential and ‘meta’ they get – to still be fully predictable.

Now, some philosophers regard the bizarreness of the story above as indicating something wrong with Newcomb’s problem itself. One group believes, for various subtle reasons, that Newcomb’s problem is either fundamentally insoluble (because it’s a true paradox) or fundamentally incoherent, and should be dissolved (Levi 1975, Sorensen 1987, Priest 2002, Maitzen and Wilson 2003, Jeffrey 2004, Slezak 2005). Another, smaller group holds that, despite claims to the contrary, a Predictor that was really perfect (or even near-perfect) would have to be some kind of supernatural being relying on “backwards causation” (Mackie 1977, Kavia 1980, Bar-Hillel and Margalit 1980, Cargile 1980, Schlesinger 1980). This claim is either used to argue for two-boxing (‘let us just use Causal Decision Theory and not worry about the lack of probabilistic independence between the states and the actions’) or to argue for the dissolution of the problem (‘why discuss an absurdity?’). I shall now discuss an argument for one-boxing that simultaneously repudiates all these deflationary and ‘corrective’ positions. The argument is independently attributable to the computer scientist Scott Aaronson and to the computer scientist Radford Neal. I shall therefore propound it using quotes from both.

In chapter 19 of his book Quantum Computing Since Democritus, titled “Free Will”, Aaronson introduces Newcomb’s problem as a choice between one-boxing, two-boxing and “the boring “Wittgenstein” position” (Aaronson, 2013, p. 295). Aaronson seems to conceive of the Wittgenstein position as the recognition of the inherent contradiction in the presupposition of free will that Newcomb’s problem makes (you are making a decision), and its simultaneous requirement that we really have no choice (the Predictor knows what you will do). Evidently, this Wittgenstein position is one of those held by many of the former group of deflationists. Though Aaronson admits that he used to himself be a “Wittgenstein” on Newcomb’s problem, he now describes himself as “an intellectually fulfilled one-boxer” (Aaronson, 2013, p. 296). His and Neal’s argument explains why, since by showing that Newcomb’s problem is possible within our own universe and that one-boxing is absolutely the right way to go, Aaronson and Neal refute both the “backwards causation” latter group and the hard-deflationist former group in one fell swoop.   

In order to make their case, both Aaronson and Neal pull a stunt more usually used by two-boxers seeking to overcome the “common cause” dilemma: they imagine what the Predictor would have to be like in real-life.^[2] Since they are computer scientists, both Neal and Aaronson imagine not that the Predictor is some kind of demon, alien or God, but that the Predictor is some kind of computer. However, one important difference between them and someone like Burgess (2004) is that they don’t believe that the Predictor could be as super-accurate as the problem stipulates with anything resembling current technology. To them, the Predictor’s having the ability to, say, conduct brain-scanning and to create a profile of you from your internet activity is nowhere near sufficient to explain the phenomenal powers of prediction that the problem stipulates.[3] The fundamental reason why our modern technology is so inadequate is revealed in the fact that the Predictor would have to successfully predict thought processes like those I outlined above. As Neal writes, “While some of our actions are easy to predict, other actions could only be predicted with high accuracy by a being who has knowledge of almost every detail of our memories and inclinations, and who uses this knowledge to simulate how we will act in a given situation” (Neal, 2005, p. 13)  So though we might imagine that some of the Predictor’s successes would have come on people who were pre-committed “one-boxers” or “two-boxers”, or people who don’t overthink things, we must also assume that the Predictor would have had success with highly self-conscious and recursive reasoners.

Importantly, since ‘you’ could easily be one of those tricky reasoners, the Predictor would have to know “absolutely everything about you” and would need to solve “a “you-complete problem”” (Aaronson, 2013, p. 296). And this leads to an earthshattering conclusion: “the Predictor needs to run a simulation of you that’s so accurate it would essentially bring into existence another copy of you” (Aaronson, 2013, p. 297). In Neal’s words, “you have no way of knowing that you are not [the simulated being]” (Neal, 2005, p. 13). The implications of this for one’s decisions may initially be unclear, but they are certainly not even-handed. Aaronson unpacks them nicely: “If you’re the simulation, and you choose both boxes, then that actually is going to affect the box contents: it will cause the Predictor not to put the million dollars in the box. And that’s why you should just take the one box” (2013, p. 297).

The brilliance of this line of thought should be clear. It overcomes the common cause dilemma because you yourself are the common cause. And it suggests a way that the Predictor could always (or almost always) get the right answer without the invocation of backwards causation, or indeed any supernaturalism at all. Perhaps most significantly, it dissolves the central issue of the Newcomb debate: the supposed conflict between Causal Decision Theory and Evidential Decision Theory (in particular, the supposed necessity that one-boxers have to ‘reject’ the former). If you don’t know whether you’re the decision-maker or the simulation of the decision-maker, there can be no conflict between the two decision theories at all. Both decision theories recommend one-boxing in this story, since the indicative conditional probabilities required by EDT are exactly the same as the subjunctive conditional probabilities required by CDT. To put this in formal terms:

For Evidential Decision Theory, EU(A) = ∑i P(Oi | A) x U(Oi). If we assume that utility equals monetary value, the expected utility of one-boxing in Aaronson and Neal’s story is simply 1 x 1,000,000 (+ 0 x 0) = 1,000,000. The expected utility of two-boxing, by contrast, is 1 x 1,000 (+0 x 0) = 1,000. For Causal Decision Theory, CEU(A) = ∑i P(A         Oi) x U(Oi). If we make the same assumption about utility, the expected utility of one-boxing in Aaronson and Neal’s story is once again 1 x 1,000,000 (+ 0 x 0) = 1,000,000. Just as before, the expected utility of two-boxing is 1 x 1,000 (+0 x 0) = 1,000. So one-boxing always wins, and there is no problem with CDT.[4]

I should say that I do not expect committed two-boxers to be convinced by this argument. No doubt their reply would be that this is just a cheap way of overcoming the common cause dilemma. Contra my claim in footnote 2, they would probably say that this re-imagining does destroy the structure of the problem, because it is functionally equivalent to injecting backwards causation or clairvoyance into the problem, which the original populariser of the problem, Robert Nozick, explicitly tells us not to do (Nozick, 1969, p.134). There’s a sense in which this reply shows the futility of even arguing Newcomb’s problem: if the two sides conceive of the problem in different ways, what can you do to forge agreement? Though I don’t wholly reject this nihilistic position, I hope to show why there is one right answer: namely, one-boxing. 

The standard two-boxer tack, exemplified by someone like Burgess (2004), is to seek out a common cause for the alignment between predictions and decisions that allows for some degree of freedom. The novel thing about Burgess’ argument is that he ends up breaking the problem up into two stages: stage one, the stage “we are currently in”, the stage before which the Predictor has gained all the information on which the prediction will be based; and stage two, the period, post-Prediction, when the decision has to be made (Burgess, 2004). Burgess believes that Newcomb’s problem is normally considered just in terms of the “second stage”, and believes, like all two-boxers, that in this stage “the conditional expected outcomes support two-boxing”, since “for all values of φ [P(According to your epistemic state, you will one-box)], your conditional expected outcome for two-boxing will be $1000 better than one-boxing” (Burgess, 2004, p. 280). However, in the first stage, he says, you have “an opportunity to influence your BATOB [brainstate at the time of brainscan]”. This means you can influence “the alien’s prediction” and “whether or not the $1m is placed in the opaque box” (Burgess, 2004, p. 280). This extra detail throws a spanner in the works, since it dramatically changes the conditional expected outcomes:

If your brainstate during the prediction can influence the prediction, the expected utility of absolutely and sincerely committing yourself to one-boxing in this first stage, according to Causal Decision Theory, would be something like 0.95 × 1,000,000 + 0.05 × 0 = 950,000. By contrast, the expected utility of remaining a confident two-boxer would be something like 0.8 × 1,000 + 0.2 × 1,001,000 = 201,000.[5]

This logic leads Burgess to a strange recommendation: “Decide to commit yourself to one-boxing before you even get brainscanned. Indeed, you may as well commit yourself to one-boxing right now” (2004, p. 280). At the same time, he does not at all dismiss his reasoning about the rationality of two-boxing in the second stage. Thus, his ultimate conclusion is as follows: “if you are in the first stage, you should commit yourself to one-boxing and if you are in the second stage, you should two-box” (2004, p. 280). This supposedly holds even though one’s sincere commitment to one-boxing will make it almost impossible to perform the rational act in the second stage.

I think Burgess’ argument is mostly very strong. One counterargument that has been made against it is that his temporal specification destroys the abstract structure of the problem. Peter Slezak, one of the first group of deflationists who thinks Newcomb’s problem is a true paradox (produced by a “cognitive illusion”) says the following of Burgess’ reformulation: “Burgess’s strategy attempts to avoid the demon’s prediction by assuming that a presentiment or tickle is merely evidence of the choice and not identical with the choice decision itself which is, thereby, assumed to be possible contrary to the tickle. Clearly, however, this must be ruled out since, ex hypothesi, as a reliable predictor, the demon will anticipate such a strategy” (Slezak, 2005, p. 2029). I’m not sure Slezak is right about this, since Burgess seems to concede that most likely one won’t be able to renounce one’s commitment in the second stage, and therefore the demon won’t be outwitted. (I shall later give a different argument for why Burgess is mistaken to think that Causal Decision Theory is actually rational in the second stage.)

To me, the interesting thing about Slezak’s own argument is its similarity to Aaronson’s and Neal’s. Slezak thinks what really makes Newcomb’s problem paradoxical is that it is “a way of contriving a Prisoner’s Dilemma against one’s self” (Slezak, 2005, p. 2030). As he elaborates, “The illusion of a partner, albeit supernatural, disguises the game against one’s self – the hidden self-referentiality that underlies Priest’s (2002) diagnosis of “rational dilemma,” Sorensen’s (1987) instability and Maitzen and Wilson’s (2003) vicious regress. Newcomb’s Problem is a device for externalizing and reflecting one’s own decisions” (Slezak, 2005, p. 2030). Of course, the difference between the arguments of Slezak and the two computer scientists is that Aaronson and Neal show that conceiving of the problem as a Prisoner’s Dilemma against one’s self doesn’t have to be paradoxical. But even though Slezak might be wrong to believe that conceiving of the problem as a battle against oneself is inherently paradoxical, there is, I believe, something important to Slezak’s preoccupation with the paradox of “self-reference” in Newcomb’s problem – and this insight (conveniently) takes us back to Burgess.

The reason why I don’t agree with Burgess, or indeed any two-boxers (i.e. the reason why I don’t think Causal Decision Theory can be rationally applied to “the second stage”) is actually directly related to the problem of self-reference – in particular, something called the “Theorem of Human Unpredictability”.

 In his paper, “A Turing Test for Free Will”, the quantum computing pioneer Seth Lloyd proves this Theorem of Human Unpredictability. As he describes it, this theorem boils down to the claim “that the problem of predicting the results of one’s decision making process is computationally strictly harder than simply going through the process itself” and that “the familiar feeling of not knowing what one’s decision will be beforehand is required by the nature of the decision making process” (Lloyd, 2013, p. 2-3). He says his argument (based on “uncomputability” and Turing’s Halting Problem) can be thought of as “making mathematically precise the suggestion of Mackay that free will arises from a form of intrinsic logical indeterminacy, and Popper’s suggestion that Gödelian paradoxes can prevent systems from being able to predict their own future behaviour” (Lloyd, 2013, p. 6). Lloyd’s mathematical reasoning would take up too much room to include, but, in essence, he provides what seems to me to be convincing proof of one matter that bears directly on Newcomb’s problem: that one cannot have any idea of the subjunctive conditional probabilities required to apply CDT in the second stage.

Causal Decision Theory tells us that in “stage two” of Newcomb’s problem two-boxing is the better option irrespective of the exact values of one’s subjunctive conditional probabilities (however likely one thinks it is that the Predictor will have predicted a given choice, based on introspection) because the two-boxing option always gives us an extra $1000. But Lloyd’s theorem raises the question: What if it is simply a mistake to try to find any values for these subjunctive conditional probabilities? To even make this attempt assumes, to at least some extent, the ability to know oneself. Yet, as Lloyd shows, recursive reasoners don’t know themselves.

The consequences of this are clear: if you don’t know yourself, you shouldn’t be so arrogant as to believe you can insert the probabilities that second-stage CDT requires. Even if you don’t want to re-imagine Newcomb’s problem in the manner of Aaronson and Neal (even if you want to keep the Predictor’s powers mysterious), you should still not try to apply CDT to the second stage of Newcomb’s problem. You should still not regard it as rational to try to outwit the Predictor. You should instead accept the power of the Predictor, and take the $1,000,000. In short, you should become a one-boxer, like me.[6]

Reference List

Cited:

Book:  Aaronson, S. 2013, Quantum Computing Since Democritus, Cambridge University Press, Cambridge.

Articles:  Burgess, S. 2004, “The Newcomb Problem: An Unqualified Resolution”. Synthese 138 (2). Springer: 261–87. Accessed from:

<http://www.jstor.org/stable/20118389>

Lloyd, S. 2013, “A Turing Test for Free Will”, Massachusetts Institute of Technology Department of Mechanical Engineering. Accessed from:

<http://arxiv.org/pdf/1310.3225.pdf>

Neal, R. 2006, “Puzzles of Anthropic Reasoning Solved Using Full Non-indexical Conditioning”, Technical Report No. 0607, Department of Statistics, University of Toronto. Accessed from:
<http://arxiv.org/pdf/math/0608592.pdf>                                  

Nozick, R. 1969, "Newcomb's Problem and Two Principles of Choice", Essays in Honor of Carl G Hempel, ed. Rescher, Nicholas. Accessed from:

<http://faculty.arts.ubc.ca/rjohns/nozick_newcomb.pdf>

Slezak, P. 2005, "Newcomb's Problem as Cognitive Illusion", Proceedings of the 27th Annual Conference of the Cognitive Science Society, Bruno G. Bara, Lawrence Barsalou & Monica Bucciarelli eds., Mahway, N.J.: Lawrence Erlbaum, pp. 2027-2033. Accessed from:

<http://csjarchive.cogsci.rpi.edu/proceedings/2005/docs/p2027.pdf>

Uncited:

Craig, W.L. 1987, "Divine Foreknowledge and Newcomb's Paradox," Philosophia 17: 331-350. Accessed from:

<http://www.leaderu.com/offices/billcraig/docs/newcomb.html>

Hedden, B. Lecture notes.

Lewis, D. 1981, “`why Ain'cha Rich?'”. Noûs 15 (3). Wiley: 377–80. Accessed from:

<http://www.jstor.org/stable/2215439>

Maitzen, S & Wilson, G. 2003, “Newcomb’s Hidden Regress”, Theory and Decision 54:2 151-162. Accessed from:

<http://commonweb.unifr.ch/artsdean/pub/gestens/f/as/files/4610/13602_104700.pdf>

---

[1] It should be noted that I do not aim to argue that Causal Decision Theory is generally incorrect; instead, like most philosophers, I believe that CDT is overall superior to its competitor, Evidential Decision Theory. One of the goals of my essay is to show that advocating one-boxing in Newcomb’s problem doesn’t have to entail spurning CDT (in other words, that Newcomb’s problem is not identical to those other common cause problems). The Aaronson and Neal re-imagining demonstrates clearly the non-necessity of spurning CDT, as does the temporally specified CDT-based “first-stage” argument of Simon Burgess, which (as will become clear) I more or less accept.

The problem with using CDT comes when we keep the Predictor’s powers mysterious (but without allowing backwards causation). It is its application in these contexts that I do find illegitimate, and EDT is left as the only formal alternative.

[2] While doing this does lead them away from the standard, abstract formulation of the problem, I nevertheless think it is a valid way of arguing the one-box case, because, as I shall explain, it doesn’t destroy the fundamental weird structure of the problem.

[3] Notwithstanding the results of the famous “Libet” experiments in the 1980s, which Aaronson mentions after he finishes his argument

[4] My repeated use of the probability value “1” here is perhaps open to question, since we are conventionally not meant to assume that the Predictor is absolutely perfect. But Aaronson and Neal have an answer to this: they say that the simulation could be slightly off sometimes, resulting in extremely rare errors, but you’d still have no way of knowing whether you are the simulation or the decision-maker. Thus the probability value of “1” seems permissible (perhaps it should be lowered to 0.99 or so, but this would make little difference).

[5] As I shall argue, it is actually very difficult to have any hold on these subjunctive conditional probabilities.

[6] Admittedly, there is still something very odd about this. A futility lingers. It remains true (if you don’t re-imagine the story in the manner of Aaronson and Neal) that if a seriously committed two-boxer walks into the room with the boxes, there will be no money in the opaque box, and therefore it wouldn’t be rational to take only this box. But I do have an answer in this case. I would say the following: Imagine if such a committed two-boxer suddenly decided to take the one box. The stipulations of the problem tell us that there would still almost certainly be $1,000,000 in it. The reason this is not absurd is that, unbeknownst to her, this committed two-boxer actually harboured signs that she wasn’t as fully committed as other two-boxers during the prediction. In other words, she didn’t know herself… And this would be true of all two-boxers who suddenly decided to take one box.

Another response is to take a step back like Burgess and say, “It would have been more rational to commit yourself to one-boxing to begin with”. 

An Essay called "My Surprisingly Novel Thoughts on the Freedom of the Will"

My Surprisingly Novel Thoughts on the Freedom of the Will

The mainstream, secular debate about free-will is a verbal dispute: it boils down to a question of framing. The debate happens under shared beliefs about the nature of the mind and the physical laws that give rise to it, as well as shared beliefs about the results of famous experiments (the Libet experiments) and recent work in neuroscience. This means there can’t be a “further fact” in dispute (indeed, a logical positivist would probably say there ought to be no debate at all). The only thing in dispute is how we should conceive of the concept of free will -- more specifically, whether there is a sensible way of modifying the traditional/religious or intuitive 'concept' such that it is compatible with the scientific facts, or whether there is no such sensible way because the idea of free will is an *inherently* superstitious concept that sits in irrevocable conflict with the facts. This is a question of framing.

Of the two framings that do exist in this debate, it seems to me that the “incompatibilist” framing is more popular among secularists than the “compatibilist” one. I think there are obvious reasons why one might expect this, despite the fact that we are all intuitive “libertarians” about the will. If you do take the word “freedom” in anything like our intuitive sense of it, the notion of free will is clearly absurd.

First of all, as soon as you dispense with the Cartesian Ego, you conclude that the mind must come from the brain (and only the brain), and that this hunk of matter must be subject to the bio-chemical-physical laws of our universe, just as all things are. This means that everything in your mind must be determined by these laws, including every aspect of your personality (whether you are good, bad, violent, non-violent, smart, dumb, rational, irrational, sane, insane, sociable, solitary) and all the decisions you make. There may well appear to be a division between the voluntary and the involuntary (we say addiction makes people zombies, for example), but it is more or less illusory. This means that when we deem murderers sane according to clinical standards – thereby implying they had a choice whether or not to kill their victim – we’re probably making some kind of error. There are, to be sure, degrees of insanity; some murderers might have genuinely weighed up the options, and experienced hesitation or regrets. However, even for these murderers, the deliberation was done according to fixed natural laws inside their brain. The reason a particular man ends up committing murder and his cousin doesn’t is almost certainly due to some slight abnormality in the man’s brain (caused by a bad set of genes, or a bad environment that led to the expression of bad phenotypes, or some kind of braindamage from physical trauma or drug abuse (probably also combined with bad genes and a bad environment)). There is no mysterious Will that somehow disrupts the natural processes going on in the brain and produces a truly spontaneous action (in any case, would the Will really be you, or just some part of you?). It may be true that quantum mechanics – the laws at the bottom of reality – are not deterministic but probabilistic, and seem to produce utterly random outcomes. While this suggests that your decisions were not predetermined from the moment of the Big Bang (which seems to be slightly more reassuring than believing they were), it still means your decisions have nothing to do with you (whatever “you” is).

Secondly, you (whatever “you” is) have absolutely no control over the vast majority of things that shaped and shape you (genetic and environmental): you don't choose your parents, your uterine environment, your general infantile environment, your city, your state or your nation, your infantile and childhood diet, the quality of the air you breathe, your parents' wealth, the size of your house, the number of books in your house, your siblings, the school you attend, your teachers at school, and so on ad infinitum.

Thirdly, as that annoying shithead Sam Harris writes, a number of findings in neuroscience have proved that “Some moments before you are aware of what you will do next—a time in which you subjectively appear to have complete freedom to behave however you please—your brain has already determined what you will do. You then become conscious of this ‘decision’ and believe that you are in the process of making it.” This means that the subjective feeling of deliberation has no connection to the actual machinations of deliberation that are going on in your brain (to which you have no access whatsoever). Consciousness of decision-making is therefore irrelevant to the question of free-will.

It is easy to see why these truths make free-will in our intuitive sense seem absurd. Define freedom in any normal way, and you will conclude that we don’t have it.

And yet, somehow, compatibilists still exist. Even more strangely, their arguments (i.e. ways of framing) aren’t totally insane either. They somehow manage to accept all these facts and still argue that we have the freedom required for moral responsibility (the only people who don't have freedom, they say, are the manifestly deranged or impaired). So how do they do it? In order to pull off this great feat of philosophical artistry, they have four main tacks.

The first is Derek Parfit’s strategy in On What Matters: highlighting the subtle distinction between “determinism” and “fatalism”. In the relevant chapter of this book, Parfit argues (as I recall it) that those who claim that determinism contradicts moral responsibility are victims of a common confusion of determinism and fatalism. Fatalism, Parfit says, is the condition of reality for Oedipus Rex in Sophocles’ famous play. If you are fated to experience some calamity, as Oedipus is, it means that whatever decision you make, the universe will find some way of putting you back on the path towards disaster. In Oedipus Rex, no matter what Oedipus chose to do, he was always going to end up having sex with his mother – thereby fulfilling the prophecy. But, as Parfit points out, nobody thinks our universe is fatalistic, just deterministic (or perhaps probabilistic, but if that makes any difference at all, it is on the side of compatibilism). In a deterministic universe, he says, our big, complex brains mean we can genuinely “make decisions”. We genuinely have “choices”, and these choices then affect our future. Even if these choices have antecedent causes in bio-chemical goings on in our neural structure (and these bio-chemical goings on have antecedent causes in the laws of physics), they are choices nonetheless. Our language of “decisions” and “responsibility” etc is therefore not in error.

The second tack is to observe how darn complicated our brains are. In his Big Think interview on Free Will, Steven Pinker (who seems to agree with me that the debate on free-will is a matter of framing) points out that the human brain consists of “a hundred billion neurons connected by a hundred trillion synapses”. This means, he says, that choices are not possible to predict “in any simple way”. Pinker also observes that there is a significant neurological difference between reflexes (unconscious responses to stimuli like an iris contracting in strong light), and active decisions (say, deliberating over what to get for dinner or where to move a chess piece). The latter involve vast areas of the brain, particularly the frontal lobes, and incorporate a vast array of information. So even if you’re not conscious of all the deliberation that goes into these active decisions (whatever “you” means), the brain has to make far more of an effort – seemingly vindicating the idea that a choice really is being made out of the landscape of possibilities. As a compatibilist would say, Does it matter that this choice goes on under cover of darkness if it’s still the brain in your skull doing it?

The third is to focus on the mentalistic level of analysis rather than the bio-chemical-physical. John Horgan demonstrates this strategy brilliantly in a scathing critique of Sam Harris’ book on free will in an article for Scientific American. His argument is worth quoting directly:

“Harris keeps insisting that because all our choices have prior causes, they are not free; they are determined. Of course all our choices are caused. No free-will proponent I know claims otherwise. The question is how are they caused? Harris seems to think that all causes are ultimately physical, and that to hold otherwise puts you in the company of believers in ghosts, souls, gods and other supernatural nonsense.

But the strange and wonderful thing about all organisms, and especially our species, is that mechanistic physical processes somehow give rise to phenomena that are not reducible to or determined by those physical processes. Human brains, in particular, generate human minds, which while subject to physical laws are influenced by non-physical factors, including ideas produced by other minds. These ideas may cause us to change our minds and make decisions that alter the trajectory of our world.

Some of us have a greater capacity to perceive and act on choices than others. The killer with a brain tumor, the schizophrenic, the sociopath, the obsessive-compulsive do not and cannot make decisions--or change their minds--in the way that I do. When I weigh the pros and cons of writing about Harris, my chain of reasoning is determined by the substance of my thoughts, not their physical instantiation.

Consider: When I watch the video of Sam Harris talking at Caltech, is it the electrons streaming through my MacBook, the photons impinging on my eye, the sound waves entering my ear that make me want to respond to Harris? Of course not. It's the meaning of the video that stirs me, not its physical embodiment. I could have watched a DVD of Harris's talk, or read a transcript, or listened to someone summarize his lecture over the telephone. And it's possible that Harris's words, instead of provoking me to write a critical response, could have changed my mind about free will, so that I decided to write a column defending his point of view. Of course, if I thought about it for a moment, I'd realize that the fact that Harris had changed my mind and hence my actions was evidence of my free will.”

The fourth is simply to draw attention to the fact that determinists are far too demanding. Horgan partly uses this strategy in the argument quoted above when he points out Harris’ obsession with “prior causes”. Determinists like Harris say that “prior causes” are all we need to refute free will. But, if you think about it, this creates a kind of absurdity. It means that anyone who denies the existence of a Cartesian ego – which makes no sense anyway – is led to believe that a human is no freer than a toad or an amoeba. In effect, therefore, if you accept the formula prior cause = no freedom, then free-will isn’t even a question. This seems obviously unacceptable: if free-will is not even a possibility on a given formula, then it’s clearly the wrong formula. Although this objection might seem like a mere bromide, it is actually an important semantic point. After all, it clearly makes sense to say that a human is more free than a chimpanzee, or that a highly intelligent, highly rational person is more free than some paranoid schizophrenic. Indeed, this brings us to one of Dan Dennett's main arguments: Dennett claims that, though we should (gleefully) abandon the intuitive or religious conception of free will, we can legitimately adopt the engineer's concept of "freedom" instead -- the view of freedom as a kind of continuum, varying in terms of degrees of 'movement' and extent of complexity. Crucially, this way of thinking allows us to say basically what we want to say: that a rock has no degrees of freedom, an ant has slightly more, a frog has more than an ant, a cat has more than a frog, a chimp has more than a cat, a mentally ill human has more than a chimp, and a mentally healthy, temperate, rational human has more than a mentally ill human. And once you accept this way of talking as legitimate (and it seems hard to reject), the shoe suddenly flips to the other foot: instead of being on the defensive, compatibilists can now accuse the incompatibilists of a kind of absurdity. How is it reasonable to deny the property of free will to the most intelligent and most complex organism in the known universe? How is it reasonable to deny free will to the animal that is the best candidate for it?

   I’m not really sure that one way of framing the debate is better than the other, although I am sympathetic to both sides. My favourite perspective on free will is that of Noam Chomsky, who frames the problem in a third way – one that commits him to neither side while also not committing him to mysticism.

Before I explain what this brilliant third perspective is, I should point out that I don’t agree with all of Chomsky’s comments about free will. For example, I know that he often makes a quip about free will that I believe to be facile: the classic meme that people who write “learned tomes” denying the existence of free will are refuting their arguments by the very act of writing a book (why would they bother making the effort if they really believed their actions were inevitable?). I know Chomsky makes this quip in jest (that’s why it’s called a “quip”), but even so, I think the quip itself is the result of a common misconception about free-will: namely, the Parfit-identified confusion of “determinism” and “fatalism”. It is not a primitive mistake to think that decisions exist in an important sense (even if you deny that they exist from the eye of the universe). It is undoubtedly true to say that people put effort into certain activities. Our language isn’t totally misleading us.

Now onto Chomsky’s perspective.

To put it briefly, Chomsky is sceptical that we even have the right categories to answer whether we have free-will or not – or perhaps even to phrase the question in the right way. He suggests that the main reason why our intuitions of free-will seem so out of whack with our best science is that we only have two scientific concepts to bring to bear on the problem – randomness and determinacy – and both of them seem utterly inappropriate. As he remarks, an intelligent Martian might be “looking at us and thinking how stupid we are – why do we keep to determinacy and randomness when there’s obviously that thing out there (that I can’t point to ‘cause I’m a human)?”

Even though Chomsky doesn’t suggest this, I can’t help thinking that what we should do to overcome this conundrum is invent some new words ourselves. They probably wouldn’t catch on (since they presumably would have been invented before if the concepts they express were congenial to our way of thinking), but it would be an interesting thing to do nonetheless. In fact, I've even made a start myself:

Perhaps we can say that brains have the property of deterministic spontaneity (only an apparent oxymoron), bio-spontaneity or neuro-spontaneity. Perhaps we can argue that bio-freedom or weak-freedom is an emergent property of neural complexity (as people say about consciousness). Perhaps we can say that humans have, instead of free-will, deterministic freedom (another apparent oxymoron), weak-freedom, bio-freedom or neuro-freedom. Perhaps we don’t even need compound words, but can just invest an entirely new word: we can say that brains have the property of spontaneation or that humans have libertaneity. Perhaps we can use the Old English word for freedom, freodom.

If you accept this new usage, it immediately solves the philosophical problem of free-will. The answer to the question, “Do humans have free-will?” goes from “It depends on how you frame it” to “That’s the wrong language; we have libertaneity” or “That’s the wrong language; we have freodom”. The answer to the question, “How does a physical brain, subject to natural laws, give us this “libertaneity” or “freodom”?” is “The brain is sufficiently complex for the emergence of the property of “spontaneation”.

Isn’t that wonderful?

[This was given a fairly significant edit on 5 July 2016]