The final version of this draft was published in the book Lines of Thoughts: Rethinking Philosophical Assumptions (Cambridge Scholars Publishing, 2014). I find the view defended here relevant because the basic idea, replacing Moore's old paraphrase, not only seems to be correct, but solves the consequence argument and gives a new compatibilist answer to Frankfurt's cases without abandoning PAP.
How a Compatibilist can do Otherwise
Wishing to escape, the person tries the door, but it is locked; she tries the window, but it has bars; then she turns back only to see that the way has been always open.
When an agent freely chooses to act, he is aware that he can later say or think: ‘I could have chosen to do otherwise’. Accepting this seems to imply a libertarian position, for if I could have chosen to do otherwise, then with a mere decision I could have changed the causal chain of events that actually occurred, thereby overriding the determinist constraint that this actual causal chain is unique and unchangeable. In the discussion below, I propose a new analysis of the sentence ‘I could have chosen to do otherwise’ and show that it has relevant implications for the compatibilist approach to free will.
1. Moore’s Traditional Paraphrase
A traditional compatibilist response to the challenge presented by ‘I could have chosen to do otherwise’ consists in trying to find an acceptable paraphrase for it. One classic attempt that we owe to G. E. Moore is as follows:
(1) I could have chosen to do otherwise
really means the same thing as
(2) If I had chosen to do otherwise, then
I would have done otherwise. 
If (1) is understood as having the same sense as (2), then (1) does not mean that the agent could have changed the actual causal chain of events, but rather the trivial conclusion that if the causal chain had been different, the agent would have followed another chain and acted differently. Since (2) expresses a counterfactual conditional that assumes only that if the causal conditions were different, then the consequences would also be different, the feeling that we could have done otherwise does not represent any threat to compatibilism. This is because the determinist thesis usually assumed by compatibilism concerns only the non-changeability of the actual causal chain, and not a relation between the actual causal chain and any other different causal chain that could have existed, but that in fact is a merely possible causal chain.
Nevertheless, Moore’s paraphrase has always seemed trivial. Moreover, it is easy to show that it is incorrect. As Roderick Chisholm has noted, statement (2) cannot mean the same thing as statement (1), since it is possible that (2) is true while (1) is false. If an alcoholic drinks because he cannot do otherwise, it remains true that if he had chosen to do otherwise (in a possible world where he was not an alcoholic), he would have done otherwise. Because of its clear inadequacy, Moore’s paraphrase has done more harm than good for the compatibilist cause.
2. Proposal for a New Paraphrase
If Moore’s paraphrase is incorrect, the compatibilist will have to look for a more adequate one. This is my first purpose in this paper. In order to find a better paraphrase, I suggest we pay attention to the following point. The sentence ‘I know that I could have chosen to do otherwise’ makes sense for the libertarian only if we understand it to mean the same thing as:
(3) I could have chosen to do otherwise, all antecedent
conditions remaining the same.
Indeed, the ability to do otherwise can only break the causal chain if we are able to do otherwise even with the same antecedent causal conditions, for in a case where the antecedent conditions are different, we have nothing more than the mere possibility of a different deterministically oriented causal chain.
Now, my compatibilist contention is that (3) only seems to be true. To arrive at this conclusion, we simply need to examine some concrete examples where a person is entitled to say, ‘I could have chosen to do otherwise’. Thus, suppose that Carl voted for Richard Nixon in 1972. After the Watergate scandal, he decides that he could indeed have voted otherwise. But what does he mean when he says, ‘I could have done otherwise’? As the following dialogue suggests, by saying that he could have voted otherwise, Carl means that at the time he had no very strong reason to vote for Nixon:
Jane: ‘Could you have voted for McGovern instead of Nixon?’
Carl: ‘To tell the truth, I have never felt much sympathy for Nixon, but since I believed that he was a more experienced politician than McGovern, I decided to take a chance on him. However, I think that I could have chosen otherwise’.
Jane: ‘What entitles you to say that?’
Carl: ‘Well, if I had been only somewhat more attentive to current events, if I had only taken the time to find more information, I guess that I would not have voted for Nixon’.
This dialogue shows that what Carl means by the claim that he could have chosen to do otherwise is clearly not that he could have done so, all the antecedent conditions remaining the same, since he named various antecedent conditions that could have been different, such as his paying more attention to political developments or deciding to spend more time reading newspapers. If he were a staunch Republican, he would also not be entitled to say that he could have chosen to do otherwise, because for reasons of principle he would always vote for his party. In this case, the conditions required to justify the utterance would have been too hard to fulfill (he would have had to overcome his Republican bias, which greatly changes the antecedent conditions). The reason Carl says that he could have done otherwise seems to be that sufficiently small changes in the antecedent conditions underlying his decision (like paying attention to information in the media) would have been enough to change his decision.
A first point about these slightly different anteceding conditions that would be able to change the decision is that they must be internal. Conditions can be external (environmental factors…) or internal (psychological factors). They must be internal, even when they have external origins and effects. There are two changes in internal antecedent conditions that might have affected how Carl voted: paying more attention to the political process and spending more time keeping informed on politics. These conditions depend on another antecedent decision being different, namely, the decision to vote more carefully or deliberately. A second point to consider is that Carl himself could have sufficiently easily changed these conditions. He could easily have paid more attention to politics. A third point, linked to the second, is that alternatives like the level of attention paid to politics must be within the range of what Carl could reasonably have willed. Indeed, when an agent makes a mistake due to factors beyond the control of his will, he does not say: ‘I could have done otherwise’. This is the reason why the called-for internal changes must at least be minor enough to be reasonably subject to the control of the agent’s will. Finally, we can be led astray by differences between demands made in the first person and demands made in the third person. When I say that I could have done otherwise, I typically mean that I could have done otherwise had there been sufficiently minor and easily influenced differences in the antecedent conditions. But when we say that someone else could have done otherwise, we often mean that he could have behaved differently under very different conditions – insofar as we think that these were under the control of the agent’s will. It seems intuitive that first person and third person evaluations of ‘I could have done otherwise’ should be similar, although they are often quite different.
What does it mean to say that a reasonably small internal change of conditions is within the scope of the agent’s will or control or rational capabilities? At the least, it means that the agent could easily change the conditions by drawing on newly acquired information (knowing that) or easily acquired skills (knowing how). In short: at least in the case of Carl’s decision, when he says, ‘I could have chosen otherwise’, what he means is, at the least, ‘I would have chosen to do otherwise if some internal conditions subject to my control had been slightly different’.
Is such a conclusion applicable to all the relevant cases in which one might say ‘I could have done otherwise’? I will consider two cases that point to an affirmative answer.
First, take the differences in the following cases:
(a) A physician makes a wrong diagnosis because he was somewhat negligent.
(b) A physician makes a wrong diagnosis due to external circumstances beyond his control: an incorrect laboratory result.
(c) A physician makes a wrong diagnosis due to internal circumstances that are far beyond his control, such as his not having a level of competence that few specialists ever achieve.
In case (a) we can say of the physician: ‘He could have done otherwise’. In case (b), the assertion ‘He could have chosen to do otherwise’ is out of place, since the necessary changes in circumstances were external and beyond the reach of his will. In case (c), the necessary changes were internal, but too far beyond his capabilities to allow a meaningful use of the locution. After all, there is no point in speaking of changes if there is no chance of making them.
Consider, as a second example, the differences among the three following cases:
(a) a student who failed to pass a math exam because she made a small but compromising error,
(b) a student who failed to pass a math exam only because she had a migraine headache while she was trying to solve the problems, and
(c) a student who failed to pass a math exam because she simply lacks mathematical aptitude.
Here again, in case (a) the student can say: ‘I could have done otherwise’, and by saying this she means that if the internal conditions leading to her decisions had been somewhat different, say, in the absence of certain distractions, she would have done better. In case (b), the student cannot say: ‘I could have done otherwise’. The reason here is that the internal condition preventing this – her migraine headache – was not under her control. She is not able to change it (at least not in my example). Finally, in case (c) the student couldn’t have done otherwise under any conceivable circumstances, since the changes would have been beyond her will or control: no information, no training could have changed the outcome. Indeed, the (c) cases in the two examples demand changes in the antecedent conditions that are unreasonable under the given circumstances. The called-for changes must be reasonably small. But what does this mean? It means that the persons in the examples are only entitled to say that they could have done otherwise if the antecedent conditions were changeable by exerting a reasonable amount of cognitive and/or emotional effort. I understand this to be the same as saying that the agent was capable of making these changes through an act of will.
In short: what justifies saying ‘I could have done otherwise’ is usually that the changes in the causal conditions necessary to reach a different decision were sufficiently small to remain within the scope of the agent’s will and thus under his or her control. They were therefore easily enough changeable through cognitive and/or emotional effort, which also means that they must be internal, at least in origin.
We conclude that,
(1) I could have chosen to do otherwise.
(3) I could have chosen to do otherwise, all antecedent conditions remaining the same.
Instead, it means at least:
(4) I would have done otherwise if certain internal conditions under the control of my will had not been the same.
It is interesting to note that we use the expression ‘could have been otherwise’ in an analogous way when talking of purely physical phenomena. We say, ‘The dam could have withstood the flood’, or ‘It could have been otherwise’, meaning that the dam would have withstood the flood if certain conditions had been different, usually only slightly different, (e.g., if its construction had been reinforced, as was originally planned), and (forcefully) if it had been within the reach of human capabilities to change them. Thus, if an unforeseen tsunami produced a flood surge that no normal dam could be expected to withstand (except perhaps a gigantic dam), we would see no sense in claiming ‘The dam could have resisted the flood’.
It must also be noted that our paraphrase of ‘I could have chosen to do otherwise’ as ‘I would have chosen to do otherwise if certain internal conditions depending on my will had been different’ isn’t open to the objection that Chisholm made to Moore’s paraphrase. If it is true that an alcoholic couldn’t have done otherwise, then it is also true that under slightly different internal conditions leading to the decision he wouldn’t have done otherwise, for different behavior would still have been beyond the control of his will.
But, couldn’t we imagine cases in which, all antecedent conditions remaining the same, the decision could have been otherwise? The only case where we could say something like this is an irrelevant one – the case of the ‘coin-flip decision’. Suppose that by some random choice a gambler decides to choose the number 16 in roulette. The winning number is 17. It may seem that the gambler could have acted differently, could have chosen the number 17, all things remaining the same. It should be noted that in this case, first, the difference between choosing the number 16 instead of choosing the number 17 or any other number isn’t completely independent of causes, since (as psychologists have long known) there is always some hidden reason for a choice, even if in such cases this reason might be completely trivial and unconscious. Aside from this, one can say that in this case the ‘choice’ is ‘free’ in the sense that it is made in a nearly random way. Does this mean that freedom is something like non-causality, as the incompatibilist believes? I do not think so. Indeed, in a similar sense we could just as easily say that the ball ‘chose’ to stop on the number 17 and was ‘free’ to do so. But then we see that in such cases the words ‘freedom’ and ‘choice’ must be put in quotes. And the reason for this is that when the incompatibilist equates freedom with non-causality, the best he can do is to use ‘freedom’ in an analogous sense, suggested by the increase in the number of alternatives resulting from a reduction of constraints, which is typical of free choice.
3. Evolutionary Reasons
As compatibilists, we need to relate our paraphrase of ‘I could have done otherwise’ to right and wrong actions and the ability to correct our mistakes and increase the success of our future actions. Suppose that after performing some action, we come to see that the action didn’t have the intended effect, or that the action’s intended effect had undesirable consequences. This may persuade us to mentally revise related decisions leading to actions and also our reasons (beliefs, wishes) leading to these decisions. Consequently, to say here that we could have chosen to do otherwise typically means respectively that we would have chosen to do otherwise if our procedures of decision had not been the same, or we would have done otherwise if our reasons (beliefs, wishes) had not been the same. Even if an agent is deterministically compelled by his beliefs and wishes, he can realize later that an internal event under the control of his will – like a mistaken conclusion due to incorrect or insufficient information – has produced an undesirable practical consequence. In this case, the person cannot afford not to hold himself responsible for the mistake. Hence, here there is also a clue to the compatibilist (determinist) explanation for a feeling like that of responsibility for our actions: If this feeling isn’t really linked with our ability to do otherwise, all conditions remaining the same, but rather with our ability to do otherwise under slightly different internal conditions under the control of reason, then it should be evolutionarily linked with an ability to learn how to change the internal conditions leading to our decisions. This will enable us to do better when confronted with similar decisional circumstances in the future.
J. R. Searle once wrote that for reasons he didn’t really understand, evolution has given us a form of experience of voluntary action where the experience of freedom, that is to say, the experience of a sense of alternative possibilities, is built into the very structure of conscious, voluntary, intentional human behavior. While opposed to the libertarian viewpoint adopted by Searle, my solution also makes the feeling of alternative possibilities understandable from an evolutionary perspective. We have an experience of freedom, namely, an awareness that alternative possibilities are open to us, that is, an awareness that we could (would) have done otherwise (under slightly different conditions within the reach of our will). Based on this, the feeling that we could have done otherwise (namely, that in this case we would have performed a different action) should play an important evolutionary role. Lessons learned from poor choices made in the past may enable us to change our behavior in ways that reduce the chances of making bad choices in the future, insofar as this can be brought about by our will. Likewise, a rewarding sense of achievement resulting from good choices in the past motivates us to develop patterns that will most likely lead us in the future to make better choices that are within the reach of our will. Thus, the sense of alternative possibilities is evolutionarily linked with a learning dispositive that is essential to improving our cognitive and behavioral patterns, enabling us to repeat successful kinds of actions and avoid repeating unsuccessful ones. This means that our sense of alternative possibilities, with the resulting idea that we could change our actions in order to obtain certain kinds of rewards and to avoid certain kinds of blame, is part of an evolutionarily shaped mechanism for attitudinal and behavioral adaptation. Ironically for libertarians, it is also a consequence of evolutionary determinism.
4. Implications for the consequence argument
Now I will show how our analysis of the freedom to do otherwise can be applied to one incompatibilist argument based on the assumption that we could have done otherwise: the so-called consequence argument.
The compatibilist understanding of ‘I could have chosen to do otherwise’ makes possible consequence arguments like Peter van Inwagen’s, which can be read as a reductio ad absurdum of soft determinism. Exposed in a simpler form, this argument is as follows: Assuming that determinism is true and that our will is nevertheless free, suppose that at time t0 the conjunction of the totality of conditions in the world is C. Now, suppose that L is the totality of the laws of nature governing the world, and that because of the conjunction of C and L at t0 (following the principle of determinism) the agent is causally determined to perform the action A (e.g., raise his right hand) at a later time t1. We can formally summarize this causally necessary determination of A by the conjunction of C and L as: ‘(C & L) → A’. Now, it seems clear that, being free, the agent could have chosen to do otherwise. This means that he could have done ~A, that is, he could have chosen not to make A his choice (e.g., could have chosen not to raise his hand). However, if the agent could have done ~A, this means that – as the application of a modus tollens shows – he could have rendered ~(C & L) true. This means that he could have done ~C and/or ~L, namely, he could have changed the past and/or even have rendered our accepted physical laws false! However, for many of us, this is an unbearable result, since neither possibility seems conceivable. The consequence of this reasoning can be seen as the following reductio: Since the assumption that the agent could have done otherwise is incontestably true, and the acceptance of this principle and of determinism force us to reach absurd conclusions, determinism must be false.
Now, if we consider the consequence argument in the light of our analysis of (1) as (4), it is obvious that the reductio does not follow: What the defender of soft determinism assumes is only that if the action were different, e.g., A1, it would be because antecedent conditions had rendered the agent’s will somewhat different, since they were produced by a somewhat different configuration of conditions in the world at t0, e.g., C1. However, to say that at t0 we could find ‘(C1 & L) → A1’ instead of ‘(C & L) → A’ is consistent with determinism, for all we are assuming here is that if the agent had a different present, he would also have a different past. However, this trivial conclusion concerns only the relationship between the actual causal chain and a different, merely possible, causal chain – a contrast unable to pose a threat to determinism.
5. Implications for Frankfurtian cases
Traditionally it has been argued that the capacity for doing otherwise is necessary not only for freedom, but also for moral responsibility. This idea has been summarized in the principle of alternative possibilities, or PAP, which can be formulated as follows:
PAP: An agent is responsible for his action only
if he is able to do otherwise.
PAP seems intuitively obvious. A person suffering a seizure of temporal lobe epilepsy, for example, cannot be held responsible for her actions, since she cannot behave otherwise. The same is true for people being threatened or influenced by drug addiction or acting under post-hypnotic suggestion. Because of its dependence on the capacity for deciding to act otherwise, PAP has been used to support libertarianism.
Nevertheless, I think that PAP can be harmonized with compatibilism, if we show that it can be generated by our own interpretation of ‘I could have decided to do otherwise’. An argument that speaks for this view is the following. For the compatibilist, to be responsible for an action implies being sensitive to the rational and moral constraints involved in the decision that leads to the action. When we impute moral responsibility to someone, we are assuming that she is able to respond to the moral constraints involved, whether or not she actually does this. But what do we mean – as compatibilists – by this ability to respond to moral constraints? Well, it seems that we are talking about the fact that one would do otherwise under different internal conditions under the control of one’s will. Consequently, moral responsibility implies the truth of PAP, based on the capability to decide otherwise.
A further point concerns Harry Frankfurt’s challenge to PAP using a well-known thought experiment. In a situation imagined by this author, there is a man named Jones, whose deliberative process leading to decisions and actions would always be blocked before he acts by an evil scientist named Black, if Jones’ decisions do not conform with Black’s wishes. However, since Jones always chooses to do what Black wishes, Black never needs to intervene to change his decisions and actions. According to Frankfurt, Jones does not have recourse to PAP, for he couldn’t choose to do other than what Black wants him to do. Nevertheless, we are still inclined to say that Jones remains responsible for his actions. Consequently, personal responsibility does not seem to require PAP.
I believe that our compatibilist understanding of our capacity to decide and act otherwise allows us to keep the essence of PAP without rejecting compatibilism. What is important for the capability to decide to act otherwise is that if certain internal antecedent conditions that led to a decision under the agent’s control had not been the same, he would have decided and acted differently. However, the existence of unknown external conditions that would hinder a change in the internal antecedent conditions is irrelevant to the connection between the capability to do otherwise and responsibility. This is intuitive: I do not cease to be responsible to give my lecture tomorrow just because, without any advance warning, the classroom building will be destroyed by an earthquake before I can hold my class… In the Frankfurtian case, there is an external condition unknown to the agent that would block a change in a decision, which is the kind of control that the nefarious Black is able to exercise. Hence, all we need is to understand PAP as including an implicit ceteris paribus – a condition excluding consideration of all constraining events that could externally affect the deliberative process, leading the agent to do otherwise, but that are unknown to him. Our paraphrase of ‘I could have chosen to do otherwise’ will then be:
(5) I would have chosen to do otherwise under certain different internal conditions subject to the control of my will, assuming that an event unknown to me and beyond the influence of my will did not prevent me from doing so.
This formulation of the sentence ‘I could have done otherwise’ generates the following reformulation of the principle of alternative responsibilities:
PAP*: An agent is responsible for what he does only if:
(i) he would decide to act otherwise under certain different internal conditions subject to his will,
(ii) on the tacit assumption that in this decision to act otherwise he would not be hindered by any event beyond the reach of his will that he does not know about.
Because of condition (ii), the interpretation of PAP given in PAP* is satisfied by the agent in Frankfurt’s example. Jones satisfies PAP* because – under his tacit assumption that he would not be hindered by any event unknown by him (that is thus beyond the reach of his will to do otherwise) – he would decide to do otherwise under certain non-identical internal conditions. This presupposes, of course, that we are not taking into account the existence of an evil scientist furtively monitoring Jones’ every step when we morally evaluate his decisions and actions. Consequently, Jones satisfies PAP*. And he is also responsible for his actions, because he is able to do otherwise in our proposed compatibilist sense of the word.
 G. E. Moore, Ethics (Oxford: Oxford University Press, 1966 (1912)), chap. 6.
 Roderick Chisholm, ‘Human Freedom and the Self’, in Gary Watson (ed.) Free Will (Oxford: Oxford University Press, 1982), p. 27.
 A similar formulation can be found in D. J. O’Connor, Free Will (London: Macmillan, 1971), p. 82. See also J. R. Searle, Minds, Brains and Science (Cambridge: Cambridge University Press, 1984), p. 86.
 A similar distinction related to the principle of alternative possibilities is considered by D. J. O’Connor in his Free Will, ibid., pp. 82-83.
 J. R. Searle, Minds, Brains and Science (Cambridge: Cambridge University Press, 1984), p. 89.
 See P. v. Inwagen, An Essay on Free Will (Oxford: Oxford University Press, 1983).
 Although it does not change the results of my argument, I see the causal relation as necessary if it is true. Some theorists believe that the causal relation isn’t a matter of necessity, but of probability. However, this does not correspond to our ordinary notion of causality. For, suppose that I say, ‘Turning the key of my old car probabilistically causes the motor to start running’. This can only be understood as an odd way to say the same thing as, ‘Turning the key of my old car will probably cause the motor to start’, understanding by this that if turning the key does not start the motor, it is not the cause of its starting. Further, when it starts the motor it is (along with all the other necessary conditions) the necessary cause of the motor starting. Suppose, to show this more clearly, that I turn the key, and for some other, completely unexpected reason the motor starts. In this case, we would not say that turning the key caused the motor to start. However, the defender of the idea that causes are or can be probabilistic should be committed to the idea that even in this case turning the key caused the motor to start, since it confirms the probabilistic expectation. What we mean by a cause is a conjunction of factors conceived of as necessary for the effect, even if we are never able to know all of these factors, and indeed even if we are never able to know for sure when and if something is necessary, that is, when and if it is the cause of something else.
 Harry Frankfurt, ‘Alternate Possibilities and Moral Responsibility’, Journal of Philosophy, 66 (1969), 828-39. See also D. Widerkehr & M. McKenna (eds.): Moral Responsibilities and Alternative Possibilities (Aldershot: Ashgate, 2003). Most studies, however, understand PAP as committing us to libertarianism, which leads down the wrong path.