When is an alternative possibility robust?

According to some, free will requires alternative possibilities. But not any old alternative possibility will do. Sometimes, being able to bring about an alternative does not bestow any control on an agent. In order to bestow control, and so be directly relevant qua alternative to grounding the agent's moral responsibility, alternatives need to be robust. Here, I investigate the nature of robust alternatives. I argue that Derk Pereboom's latest robustness criterion is too strong, and I suggest a different criterion based on the idea that what agents need to be able to do is keep open the possibility of securing their blamelessness, rather than needing to directly ensure their own blamelessness at the time of decision.

In the discussion which followed it was quickly pointed out that many of the early cases relied on what has come to be known as a prior sign: an indicator which the intervener-Black, in the case above-uses to work out whether to intervene. Fischer talks above of Jones "showing an inclination" of voting for Carter without specifying the nature of that inclination. Others talked about Jones blushing or exhibiting some neural pattern which reveals what he is going to do. Whatever the precise form, a worry arises for the proponent of the cases. If the prior sign is a perfectly reliable indicator of Jones's decision, then it seems that there must be a deterministic connection between the prior sign and the decision. While the compatibilist proponent of PAP will have no grounds on which to object to this, the incompatibilist proponent of PAP now has a reason to dismiss the case. That's because the incompatibilist holds that the making of free decisions is incompatible with those decisions being determined. On the other hand, if the prior sign isn't a reliable indicator of Jones's subsequent decision, then it seems like Jones might have had an alternative after all. If these latter FSCs demonstrate anything, then it is has not yet been shown that it is anything which needs to worry the incompatibilist. This reply is known as the Dilemma Defence and was advanced by Robert Kane (1985, p. 51 fn. 25), Carl Ginet (1996), and David Widerker (2003).

| PEREBOOM AND HIS ROBUSTNESS CRITERION
One line of reply to the Dilemma Defence has been to develop explicitly indeterministic FSCs without any deterministic connection between the prior sign and the agent's subsequent decision. The most powerful of these are the socalled buffer cases invented independently by David Hunt (2005) and Derk Pereboom (2001, pp. 18ff). The latest such case put forward by Pereboom runs as follows: Tax Cut: Jones can vote for or against a modest tax cut for those in his high-income group by pushing either the "yes" or the "no" button in the voting booth. Once he has entered the voting booth, he has exactly two minutes to vote, and a downward-to-zero ticking timer is prominently displayed. If he does not vote, he will have to pay a fine, substantial enough so that in his situation he is committed with certainty to voting (either for or against), and this is underlain by the fact that the prospect of the fine, together with background conditions, causally determines him to vote. Jones has concluded that voting for the tax cut is barely on balance morally wrong, since he believes it would not stimulate the economy appreciably, while adding wealth to the already wealthy without helping the less well off, despite how it has been advertised. He is receptive and reactive to these general sorts of moral reasons: he would vote against a substantially larger tax cut for his income group on account of reasons of this sort, and has actually done so in the past. He spends some time in the voting booth rehearsing the relevant moral and self-interested reasons. But what would be required for him to decide to vote against the tax cut is for him to vividly imagine that his boss would find out, whereupon due to her political leanings she 2 KITTLE would punish him by not promoting him to a better position. In this situation it is causally necessary for his not deciding to vote for the tax cut, and to vote against it instead, that he vividly imagine her finding out and not being promoted, which can occur to him involuntarily or else voluntarily by his libertarian free will.
Jones understands that imagining the punishment scenario will put him in a motivational position to vote against. But so imagining is not causally sufficient for his deciding to vote against the tax cut, for even then he could still, by his libertarian free will, either decide to vote for or against (without the intervener's device in place). However, a neuroscientist has, unbeknownst to him, implanted a device in his brain, which, were it to sense his vividly imagining the punishment scenario, would stimulate his brain so as to causally determine the decision to vote for the tax cut. Jones's imagination is not exercised in this way, and he decides to vote in favor while the device remains idle (Pereboom, 2012, pp. 308-309).
Ignoring the neuroscientist intervener for the moment, the structure of this case is as follows. Jones is consider-  (Moya, 2006, p. 58). Since then, Pereboom has adopted and endorsed the terminology. For Pereboom, an alternative is exempting only if, were the agent to avail himself of the alternative, and were there no intervener, the agent would succeed in bearing no blame for performing an action of the type in question (Pereboom, 2014, pp. 16-17). To be explicit: Pereboom assesses whether an alternative is exempting by asking what would happen in that alternative "without the device in place" (Pereboom, 2014, p. 16 [Pereboom, 2014, p. 17]). Pereboom formalises these ideas in the following robustness criterion (Pereboom, 2014, p. 13):

Robustness (B):
For an agent to have a robust alternative to her immoral action A, that is, an alternative relevant per se to explaining why she is blameworthy for performing A, it must be that i. she instead could have voluntarily acted or refrained from acting as a result of which she would be blameless, and ii. for at least one such exempting acting or refraining, she was cognitively sensitive to the fact that she could so voluntarily act or refrain, and to the fact that if she voluntarily so acted or refrained she would then be, or would likely be, blameless.

KITTLE
Is Robustness (B) a satisfactory criterion of robustness? To answer that we need to know what it applies to.
Taken at face value, "A" is a name for a particular action that some agent performs. This reading, I submit, is unsatisfactory. It states baldly that the agent is blameworthy for some particular action A, but a particular action is going to have lots of properties of which the agent is entirely unaware and which could not be controlled even if the agent was aware of them. Consider Jones's voting. Let's assume this involves making a mark in a box on a voting slip.
Jones's vote, like any particular instance of voting, will involve making a mark of some determinate length in one of the boxes, using ink of a particular shade, and so on. Let's assume the mark Jones makes is 5.667 mm long and in a shade half way between Columbia blue and Cornflower blue. On the current reading, Robustness (B) has it that Jones is responsible for the particular act he performed-all of it. That includes the properties being a vote made with a mark 5.677 mm long and being made in a shade of blue half way between Columbia and Cornflower blue. But is Jones responsible for such aspects of his action? That seems doubtful. To begin with, if Jones isn't aware of these properties of his action, they are not properties for which he can be responsible. But even if we gerrymandered a context where Jones was aware of such properties and where they had moral importance-e.g., Jones lives under a strange fundamentalist government which has decreed that all votes must be made with marks having an even length in microns, which would result in it being morally good to vote by making an even-length mark (because obeying the law, avoiding punishment for oneself, one's family, etc., are moral concerns)-that would not show that Jones can be responsible for them. That's because they lie outside the scope of his control: humans cannot control marks made on pieces of paper with that kind of precision. Obviously, this point generalises to all of our ordinary actions.
It is far more plausible to hold that agents only control some aspects of the particular actions they perform, and that therefore they are only morally responsible for some aspects of their particular performances. The following reframing of Robustness (B) aims to capture this idea while also retaining the spirit of Pereboom's Robustness (B):

Robustness (C):
For an agent to have a robust alternative to her having A-ed, which it was immoral to do, that is, an alternative relevant per se to explaining why she is blameworthy for having A-ed, it must be that i. she instead could have voluntarily acted or refrained from acting as a result of which she would be blameless with respect to having A-ed, and ii. for at least one such exempting acting or refraining, she was cognitively sensitive to the fact that she could so voluntarily act or refrain, and to the fact that if she voluntarily so acted or refrained she would then be, or would likely be, blameless with respect to having A-ed.
The idea with Robustness (C) is that "A" should be filled in with a description of what the agent has done; "A" is not a proper name of a particular action, and neither should it be replaced with a phrase that contains an indexical that refers to a particular action (such as "performed that action"). It should be replaced by phrases like "walked to work" or "punched Tom." This allows for a more fine-grained treatment because phrases like "having punched Tom" and "having punched Tom hard in the face" name facts or states of affairs. These contrast with phrases like "the punch Jones gave Tom" and "Jones's punching of Tom" which name particulars (See Jonathan Bennett (1988, ch 1) for elaboration). And while, say, Jones's particular action might be a punching of Tom by Jones, and a very powerful punch, and a punch in the face, the fact that Jones punched Tom is different to the fact that Jones punched Tom in the face and different again to the fact that Jones punched Tom in the face at full force. Robustness (B) is unsatisfactory because (taken at face value) it applies to particular actions and when satisfied yields the result that an agent is responsible for the entirety of that particular action. This, I have suggested, is implausible. And Robustness (C) addresses that problem because it concerns states of affairs. Not only does it allow us to avoid the conclusion that Jones is responsible for, say, punching Tom in the face with a force of precisely 99.67543 pounds (the idea here being that Jones will lack the ability to control the level of force he exerts to that level of specificity, meaning that he can't be responsible for such a determinate property of one of his actions), it also allows us to distinguish between, say, punching Tom in the face with all his strength and punching Tom. This is a welcome conclusion. If, for example, Jones's character (through no fault of his own) made it inevitable that he punch Tom, but didn't dictate how he did it (where 4 KITTLE he punched him, and the rough level of force used), we might be inclined to say that Jones is not responsible for punching Tom simpliciter but he is responsible for punching him in the face with all his strength.

| ROBUSTNESS AS KEEPING ONE'S OPTIONS OPEN
To recap so far. Robustness (B) is Pereboom's latest robustness criterion, but it applies to particular actions (action tokens). I have suggested that it is implausible that agents are ever in control of (and so responsible for) particular actions (as a whole); rather, agents are in control of performing actions of a certain type-that is, they are in control certain aspects or properties of the actions they perform. This can be expressed by saying that the agent is responsible for certain states of affairs. And Robustness (C) is a re-rendering of the idea behind Pereboom's Robustness (B) which is framed in terms of states of affairs.
Both Robustness (B) and Robustness (C) codify the idea that to have a robust AP an agent needs to be able to ensure that she will be exempt from blame (Pereboom, 2014, pp. 11, 16, 21). This is the key idea behind Pereboom's understanding of robustness: robustness with respect to performing an action of type A requires being able to secure one's blamelessness with respect to performing an action of that type. It is this equating of robustness with exempting oneself from blame that I want to challenge. I will pursue two strategies. First, I will present some examples where Pereboom's robustness criterion isn't satisfied but where I think it is plausible to consider the agent responsible in virtue of the alternatives available. Now, given that I believe it's plausible to think Jones's alternative in Tax Cut is robust, but Pereboom and others who agree with him don't, I must present examples where it is more plausible that the agent has a robust alternative than it is in Tax Cut if I am to stand any chance of persuading my opponents. That is one of the things I will aim to do. Second, I will propose a robustness criterion which I think explains why Jones in Tax Cut and the agents in my additional scenarios have robust alternatives.
Here is the first case, which will help us to see that Robustness (C) (i.e., Pereboom's suggestion about robustness, but re-formulated with respect to states of affairs) is too strong: Let me expand on the above scenario in a way which I hope makes the above point clearer. Suppose that Pete is reminded on Tuesday at 9 a.m. that he must choose what to eat, and suppose if he does not choose, then a meal will be selected for him at random, such that he's reasonably motivated to choose himself. If Pete were to decide at 9 a.m.
to find his details, login to the site and look at the options, he'd discover there were more options than he thought and so he would put them to one side and resolve to look them over properly at 12 p.m. during lunch. However, Pete decides at 9 a.m. to fire off a quick email to confirm his selection of the butternut squash dish. to install a brain-manipulation device into Pete's brain and to then use that device to make Pete choose the butternut squash if needed-all of which he can do in a matter of seconds without Pete noticing a thing, of course. None of those potential unfoldings of things seems to undermine the judgement, not just that Pete is responsible, but that he's responsible because of the choice he had at 9 a.m.
At 9 a.m., Pete faced a choice between choosing the butternut squash or choosing to look up his login details, visit the website and look at the menu. That choice is not symmetrical in the following sense: the second option does not resolve the same issue or uncertainty that the first option does. But does that matter? To be sure, it might be a natural enough thought that to be responsible for deciding to A an agent needs to be able to decide to B, where B is an alternative to A which settles the same issue that A would settle, in the way that choosing red is an alternative to choosing blue. Perhaps that is a natural thought. But is it correct? My suggestion is that the above case shows it to be too strong. Pete chooses dinner can be amended with a Frankfurtian intervener so that Pete is unable to ensure that he chooses other than the butternut squash. That doesn't show he doesn't have a robust alternative, nor that robust alternatives aren't needed for control-it shows that Robustness (C) is false.
In it's place, I want to propose a robustness criterion based on the following idea: what's important for robustness is not that the agent secures her blamelessness with respect to the performance of an action of some type, but that she do something which she believes keeps her options open. In this vein, I suggest the following:

Robustness (D):
For an agent to have a robust alternative to her having A-ed, which it was immoral to do, that is, an alternative relevant per se to explaining why she is blameworthy for A-ing, it must be that i. she instead could have voluntarily acted or refrained from acting as a result of which she would have maintained the same level of blameworthiness with respect to having A-ed, and ii. for at least one such acting or refraining, she was cognitively sensitive to the fact that she could so voluntarily act or refrain, and to the fact that if she voluntarily so acted or refrained she would have maintained that same level of blameworthiness with respect to having A-ed, and iii. for at least one of the actings or refrainings which satisfies both i and ii, the agent believed either: (a) that realising that alternative would have precluded her from blame, or (b) that realising that alternative would have kept her options open with respect to having A-ed.

KITTLE
Clauses (i) and (ii) here parallel clauses (i) and (ii) in Robustness (C), with the amendment that instead of requiring the agent to be able to do something which would ensure her blamelessness, the agent needs only to be able to do something that would maintain her current level of blame with respect to performances of the action type in question. This change, which is discussed more below, is due to not requiring that the agent secure her blamelessness but only that she be able to keep her options open. The construction "blameworthiness with respect to having A-ed" is due to the move to a robustness principle which covers states of affairs. Given that the reason this move was made is because agents are only ever in control of (and so responsible for) some but not all aspects of what they do, the states of affairs in question will necessarily be general in character (the whole point was to abstract away from those properties which are not controllable). Unless we then time index these states of affairs, which I have not required in Robustness (D), they might obtain more than once. So Stefan might steal £5 from his brother in the morning, and then again in the evening. Stefan is blameworthy for having stolen £5 after the first theft. After the second theft Stefan is blameworthy for having stolen £5 twice, or, equivalently, has increased his level of blameworthiness with respect to having stolen £5. Here is a more detailed example illustrating the need for this construction:

etc). Now let's say that Larry successfully employs one of these techniques several times that morning, but in the afternoon he actively embraces a lustful thought when it occurs to him.
This example highlights the fact that some types of action are ones we can perform many times each day or even each hour. We might be repeatedly responsible for such actions. Robustness (B) did not need any special machinery to handle such cases because, by employing a proper name for the action in question, it applied to token actions. But as already argued agents are not responsible for action particulars but rather for performing actions of certain types with certain properties. So once we make this move we need to handle the case where agents perform actions of the same type multiple times. Hence, Robustness (D)'s talk of maintaining the same level of blameworthiness with respect to the type of action performed.
The third clause of Robustness (D) is intended to capture the kind of connection needed between the alternative and the decision which is in fact made (or the action which is in fact performed other reason than to scratch the itch is not a robust alternative. But an alternative where Pete scratches the itch in order to keep his options open-perhaps he sees this as buying himself a bit more time-is a robust alternative. Whether an alternative is robust, then, depends on the reasons for which the decision in the "alternative sequence" was made.
If this thought is on the right lines, the easiest way to accommodate the insight is to add a fourth clause as follows: iv. for at least one of the actings or refrainings which satisfies (i)-(iii), the agent is able to bring about the alternative either (a) in order to preclude herself being blameworthy, or ( In most contexts, eating a biscuit is not connected to deciding to go and meet someone from the airport. But what I hope the Walter scenario makes clear is that such an alternative could be so connected. If Walter decides to sit down and eat a biscuit in order to keep his options open-that is, in order to be better able to make a decision that is not based mainly on his tiredness levels-then it is a robust alternative to his deciding to go straight home. Walter has options: he can decide to go home, or decide to sit down and take stock. The latter would make it possible for him to make a particular decision which at the start of the scenario he cannot make. For that reason, if Walter were to decide right away to go home, he has exercised a robust form of control and so could be morally responsible for so deciding, if it is appropriate to assess that decision morally. This is so even if, say, were Walter to have sat down to eat a biscuit he would have been beaten-up and robbed and thus rendered unable to go anywhere. Walter, Larry, Pete and Jones all have robust alternatives for this same reason.

| THE HUNT AND SHABO TEST FOR ROBUSTNESS
In the previous section I argued that Pereboom's robustness criterion was too strong and I suggested an alternative criterion according to which Jones in Tax Cut does have a robust alternative. In putting forward this criterion I must contend with an argument for a certain class of robustness criteria that can be found in a test for robustness that has been put forward (independently) by David Hunt (2005) and Seth Shabo (2007) and then further developed in Hunt and Shabo (2013). If this test is acceptable, it will serve as an argument against my robustness criterion because it excludes as robust alternatives which my criterion counts as robust.
Hunt states the key thought behind the test as follows: "if … an agent is morally responsible in virtue of being blameworthy, a PAP-relevant alternative should be such that, by accessing it, the agent would avoid (or mitigate) blame" (Hunt, 2005, p. 138). The affinity to Pereboom's understanding of exempting alternatives is obvious. And like Pereboom, Hunt and Shabo would agree that the alternative Jones has-deciding to imagine being punished-is plausibly a robust alternative for Jones's not imagining being punished. They would identify the crucial question as whether Jones's being able to decide to imagine being punished is a robust alternative for Jones's decision to vote for the tax cut. They address this question in two ways. First, drawing from Hunt's earlier paper, they present the following case (Hunt, 2005, p. 138): Jack is a smoker who wants to quit; he correctly believes that he can quit only by joining and completing a 12-step programme. Mack is Jack's smoking buddy who doesn't want Jack to quit and who would, if Jack joined the programme, slip Jack a pill to make his desire to smoke irresistible. In this scenario, Jack cannot decide to quit smoking, but he can join the 12-step programme, and that seems like it would be a robust alternative for Jack's continued smoking. Hunt and Shabo agree and suggest that this is because were Jack to join the programme he would be doing so in order to quit: in other words, he'd be taking the first step towards deciding to quit (and quitting) and so already trying to quit. So Jack's alternative is robust because it's a means to an end-namely, deciding to quit-which is incompatible with ("symmetrical to") his decision to continue smoking. One (relatively) develop my reply somewhat, I conceive of the decision that Jones in Tax Cut would make if he decided to imagine being punished as a means to making a better informed decision. It's not part of the means of deciding to vote against the tax cut. But in making the decision to imagine being punished, Jones is making a decision to keep an option open, he's making a decision to alter his motivational state slightly, with the knowledge that might then make it easier for him to vote against the tax cut, and he's making a decision to avoid committing himself there and then to voting for the tax cut. And that is enough to make it a robust alternative. Now Hunt and Shabo agree that the means-end constraint as they state it is in need of more defence, and so they go on to develop their key thought in a second direction which doesn't appeal to the role of reasons. This is the idea that Jones's alternative would not "by itself eliminate or mitigate" Jones's blameworthiness for deciding to vote for the tax cut (Hunt & Shabo, 2013, p. 619 It's important to be clear about how Hunt and Shabo's argument is supposed to work. They are not claiming that in order to be robust, we must be able to answer the following question positively: if there had been no intervener, would exploiting the leeway available to the agent in an FSC prior to then deciding in the same manner have mitigated or eliminated the agent's blameworthiness? That is, they are not saying that a necessary condition on an alternative being robust is that, if it were realised in the absence of the intervener, prior to the agent then deciding as she (in fact) does, the agent's blame would be lessened. But they are saying that if a positive answer cannot be given to that question, then the burden of explanation is on the defender of PAP to explain why.
However, it is telling that Hunt and Shabo don't offer this test as a necessary condition on robustness. They do not do so, because they cannot do so: to make this a necessary condition on robustness would prove far too much because it would mean that deciding not to A would not-at least in most cases-count as a robust alternative for deciding to A. As Hunt and Shabo themselves explain: suppose we consider a standard (non-buffer) Frankfurt-style case where the agent, who actually decides to do A, has the alternative of deciding to refrain from A-ing (suppose that the intervener would then intervene, causing the agent to decide to A). Hunt and Shabo agree that it will be "plausible" to think that deciding to refrain from A-ing is a robust alternative to deciding to A (this is something of an understatement: if anything counts as a robust alternative for deciding to A, surely it is deciding to refrain from A-ing). But now suppose the agent is not subject to an intervener, decides to refrain from A-ing, then changes her mind and decides to A. Is her blame lessened or even removed altogether? No, at least not in most ordinary cases.
Making a decision to not-A rarely ensures (by itself) that an agent incurs no blame for a decision to A because it is (usually) open to an agent to change her mind and decide to A after all. So Hunt and Shabo cannot possibly hold that their robustness test is a necessary condition on robustness.
10 KITTLE Yet they suggest that it is incumbent on the defender of PAP to explain just why an alternative is robust if it fails this test. And they maintain this even though the paradigm case of a robust alternative-a decision to refrain from Aing-fails this test. Given that the paradigm robust alternative fails the test, I think it fair for the defender of PAP to ask for further details on why the test should be thought in any way plausible. This point is strengthened, I think, once we recognise that the PAP defender can and should see the alternatives that remain in cases like Tax Cut as decisions (as I have suggested above). It is, to be sure, not a decision to vote against the tax cut, and neither is it a decision to take the first step towards an end-voting against the tax cut-that Jones has settled on (indeed, if Jones had that end, he would have already made a decision to vote against the tax cut). Nevertheless, Jones's decision to imagine being punished is a decision to keep his options open, to potentially adjust his motivational state in the direction of that which he knows is right, as well as a decision to not commit there and then to voting for the tax cut. For those reasons Jones's decision is a robust alternative. This decision fails Hunt and Shabo's test for robustness, but so does the paradigm case of a robust alternative, so their test as currently framed should do nothing to trouble the defender of PAP.

| CONCLUSION
In this essay, I have investigated the notion of robust alternatives, that is, the question of when alternative possibilities per se explain an agent's responsibility. I presented several cases which provide good reason for thinking that Pereboom's robustness criterion is too strong, and I have developed a replacement principle which does justice to the cases presented. I have also offered a reply to Hunt and Shabo who have presented a test for robustness which is at odds with my criterion; I have suggested that further reason needs to be given why we should take their test seriously given that the paradigm case of a robust alternative fails the test, and demonstrated how the alternatives available to the agents in the so-called buffer-style Frankfurt cases, being the result of decisions, are closer the paradigm case than is sometimes realised.

ENDNOTES
1 This literature is best thought of as an investigation into the nature of the ability to do otherwise rather than a body of work on the kind of cases Frankfurt highlighted (especially since much of the discussion concerns the metaphysics of control which Frankfurt hoped to avoid). 2 Frankfurt deliberately called this the Principle of Alternate-and not Alternative-Possibilities. This was because he thinks that the alternative options are merely second best: actions the agent would perform only had he not been able to do "what he clearly preferred to do" (Frankfurt, 1999, p. 372). Proponents of the Principle will reject that idea out of hand, so the apparently innocent name change is in fact non-incidental.
3 Some call those who endorse the need for APs leeway theorists. I find this name tendentious as it gives the impression that such theorists are engaged in a futile attempt to squeeze control out of alternatives. 4 I would like to thank an anonymous referee for raising this point.