Lewis's Nihil Obstat
The latest issue of The Monist contains an outline of an unpublished paper by Lewis: "Nihil Obstat: An Analysis of Ability", along with a useful commentary by Helen Beebee, Maria Svedberg, and Ann Whittle.
Lewis's analysis of ability goes as follows:
You're able to φ iff, for some basic action
(1) doing it would be φing, and
(2) there is no obstacle to doing it.
It is clear from this analysis, and from the context in which it is presented, that Lewis is only interested in a rather specific sense of 'ability'. He wants to spell out the sense in which we are able to perform particular intentional actions that we don't actually perform, even if the world is deterministic. He is not interested in our ability to regrow injured skin, or in the ability of tardigrades and steel to withstand high temperatures. He also isn't interested in what are sometimes called "general abilities", like my ability to play the piano that I have even when I don't have access to a piano. (At any rate, this kind of ability is not covered by his analysis, and it isn't relevant to compatibilism.)
For basic actions, Lewis's analysis simplifies to "ability iff no obstacles". At some points in the paper, Lewis actually seems to assume this simpler analysis, even for non-basic actions. For example, he discusses lack of funds as an obstacle for making a purchase, and suggests that if you are able to overcome this obstacle (by earning or borrowing the money), we might say that (in some extended sense) you are able to make the purchase. On the above analysis, however, what matters are only obstacles to performing basic actions, and making a purchase is surely not a basic action.
Lewis does not explain why he officially restricts "ability iff no obstacles" to basic actions. Perhaps he thought that the simpler "no obstacles" analysis would make abilities too easy. Even if you're bad a darts, there is arguably no obstacle to you hitting the bulls eye. Nor is there an obstacle to drawing the Queen of Hears from a shuffled deck of cards. But we're reluctant to attribute these abilities.
Immediately after stating his analysis, Lewis remarks that we sometimes require that you know which basic action would constitute φing. This is the "effective"/"transparent" ambiguity that I discuss at length in my recent paper on abilities. Lewis uses the same example: if you don't know the combination to a safe, there's a sense in which you can open it (by dialing 417-154-832) and a sense in which you can't (you don't know that you have to dial 417-154-832).
How does this ambiguity arise? In their commentary, Beebee et al suggest that Lewis understands not knowing which basic action would constitute φing as a potential obstacle. I don't think that's what Lewis meant. As Beebee et al acknowledge, Lewis never lists ignorance as an obstacle, and it's unclear how this would generate the two readings. Also, recall that the relevant obstacles in the analysis are not obstacles to φing, but obstacles to performing the basic action. But why would not knowing the safe's combination be an obstacle to dialling 417-154-832 (or whatever else is the basic action here)? Moreover, if we count the ignorance as an obstacle (as Beebee et al suggest), we should be equally reluctant to say that you can dial 417-154-832 as we are to say that you can open the safe. But it's hard to hear 'you can dial 417-154-832' as false.
I think what Lewis had in mind is what I also suggest in my paper: that sometimes we impose a third condition on an agent's ability to φ. The third condition is that the agent knows which basic action would constitute φing.
On the present interpretation, Lewis's analysis is similar to "Analysis 1" in my paper, which runs as follows:
S can φ (effectively) iff there is an available variation of S's volitional state that would cause S to φ.
S can φ (transparently) iff there is an available variation V of S’s volitional state such that (i) V would cause S to φ, and (ii) S knows that V would cause her to φ provided that φing is under her volitional control.
There are three main differences.
First, I speak of variations of the agent's volitional state where Lewis speaks of basic actions. (Accordingly, I say that these would cause φing, rather than constitute φing). I don't think this makes a big difference. Plausibly, every relevant basic action is the result of some variation of the agent's volitional state, and every such variation leads to some basic action.
Second, my analysis of the "transparent" reading has a further clause: the agent must know which basic act/volition would do the job on the assumption that some basic act/volition does the job. Without this condition, the analysis predicts that if you falsely believe that a door is locked, then you're not able to open it, in the same ("transparent") sense in which you're not able to open a safe if you don't know the combination. But there is no reasonable sense in which you're not able to open a door merely because you falsely believe that it is locked. Lewis's analysis gets this wrong.
Lewis and I both impose a restriction on the relevant basic actions or volitions. Lewis's condition (2) says that there must be "no obstacle" to performing the basic act. I say that the volitional state must be "available". In either case, the restriction is meant to ensure that, for example, an arachnophobic does not count as being able to pick up a spider: I say that willing to pick up the spider is not available for the arachnophobic; Lewis says that the phobia is an "obstacle" to performing the act.
But my availability restriction is not equivalent to Lewis's no-obstacle restriction. (This is the third difference.) I suggest that a volitional state is available iff (roughly) the agent could be reasoned into in. Lewis suggests that an obstacle is a robust preventer: something that would cause you not to perform the relevant act, and that would not go away if things were a little different.
As examples of obstacles, Lewis lists phobias, paralysis, being at the wrong spot, lacking funds, and lacking tools. As I said above, some of these are not plausible obstacles to performing any basic actions, so it's not clear why they are mentioned. As Beebee et al point out, it is also hard to make sense of Lewis's claim in section 5, that an agent who can remove an obstacle to φing may count as φing. If I lack the funds to make a purchase, there is no basic action that would amount to making the purchase: I fail condition (1), before we even get to the existence or non-existence of obstacles in condition (2).
Beebee et al suggest that we could make sense of Lewis's claim by filling in different time parameters. If I can earn the money to make a purchase, then I am not able to make the purchase now, but I am able to make it later. Perhaps that's what Lewis had in mind. Or perhaps he was slipping into thinking about more "general" abilities. If there's no piano around, I can't play the piano: intuitively, the absence of a piano is an obstacle to my playing. But if I can remove the obstacle, then there's another sense in which I can play the piano.
Anyway, all of this doesn't seem relevant to the analysis as presented in section 3, which only mentions obstacles to basic actions.
Recall that the aim of Lewis's analysis is to spell out a notion of ability on which it is clear that determinism is compatible with an ability to act otherwise. Given Lewis's proposal, this means that determinism should provide no "obstacles" to performing basic acts other than the ones we actually perform. What could these obstacles be? Two candidates come to mind: first, the past in conjunction with the laws; second, our preferences.
The past and the laws are easy to deal with: they aren't obstacles in Lewis's sense because they aren't robust preventers. In a deterministic world, the past and the laws do prevent us from acting otherwise, but they would not prevent us if things were a little different.
The case of preferences is discussed in section 6. Lewis suggests that contrary preferences can be regarded as obstacles if they are sufficiently strong. If "the balance of pros and cons" strongly favours the chosen act, then our preferences are robust preventers. But if the balance is "delicate", our preferences are not robust preventers, as they would not be preventers if things were a little different.
Here my analysis disagrees. I say that you are able to choose a different act as long as you could be reasoned into it (roughly). Usually, even if we have a strong preference to, say, order bitter rather than lager, we could be persuaded to make a different choice. On Lewis's account, we are able to act otherwise only if there's no strong reason to act the way we do. As Beebee et al point out, this is enough to square abilities to act otherwise with determinism, but it is not enough if we want to connect the ability to act otherwise with blame. I also think we need more if we want to ensure that ought implies can: sometimes we ought to do something even if that would completely go against our preferences.