In order to have φed freely, is it necessary that – without changing the circumstances of the action in any way – it was possible for me to have not-φed?

Essay

No, it is not necessary. This categorical definition of free will is overly strong, and would require that agents be able to act against their interests. In this essay, I will first investigate our motivations for producing an analysis of free will at all, and then explore the categorical, conditional, and actual-sequence views, concluding that the actual-sequence account is the definition that we ought to use.

The question above invites a discussion about what we mean by the word “freely”. We cannot test empirically whether a definition is correct, as changing the concept that a particular string of characters refers to would not lead us to different, falsifiable predictions about reality. Definitions allow us to distinguish between concepts, but these distinctions are not mind-independent features of the world ( Putnam 1988, 114

From the SEP.

). So, to judge between competing definitions of “freely”, I will rely on two criteria: the definition should be internally consistent, and the conclusions it leads to should line up with my considered intuitions.

Why argue over the meaning of “freely” at all? Justifications are often given in terms of importance for morality (Watson 1982, 8-11):

P1: S’s freely doing φ is a necessary condition for S to be morally responsible for φ.

P2: S’s moral responsibility for φ is a necessary condition for S to be a legitimate object of praise or blame for φ.

C: Each of the necessary conditions for S doing φ freely is individually required for S to be legitimately praised or blamed for φ.

If this argument is sound (for it is valid via modus ponens), then an analysis of freedom would help with deliberations over whether or not to praise or blame others for their actions. However, P2 is not necessarily true. An act-utilitarian would point out that the actions of praising and blaming are instrumental towards maximising utility (Smart 1993, 49-50): if it happens that utility is maximised when S is reproached for an action they are not morally responsible for, then so be it. Therefore, at least insofar as applications to morality are concerned, the following analysis of freedom is useful

See also “Disguised Queries”

only for those who are not act-utilitarians.

There is an attractive simplicity to the proposition that, for an action φ to have been done freely, it must have been possible (without a change in circumstances) to have not-φed.¹ An immediate problem

Maybe the Finch & Warfield (1998) paper could be brought in somewhere here too?

with the categorical account of freedom, though, is that it unintuitively

On reflection, it might not be so ridiculous to claim this is possible. I’m not arguing that you actually would not-φ given that you wanted to φ, but rather that it would not bring about a contradiction for you to want to φ and subsequently not-φ. We see this in our lives all the time: there are many things which I superficially might want to do yet subsequently fail to do. This is related to weak will and also higher-order desires, neither of which get enough space in this essay.

seems to require that it be possible for me to act in direct opposition to my desires and interests. If my φing came about because I wanted to, then that desire to φ would have been part of the circumstances of the action. So, according to the definition, in order to have φed freely, it must have been possible for me to have not-φed, even given that I wanted to (Wolf 1980, 152-153). If the categorical definition is to consider me free, then there must be no deterministic link between the circumstances of an action and that action. But if that condition is met, where is the space for me as an agent to make decisions based on my interests and act upon them? The indeterminism demanded here prevents S from being the sort of individual that we can describe as “freely” doing anything at all, and this incoherence should lead us to reject the overly strong categorical condition for freedom.

Relaxing the requirement for S to have φed freely to “if S had wanted to not-φ, S would have not-φed” is one natural response to the argument from reasons (Wolf 1980, 154). By no longer requiring that the circumstances of the action remain the same, this conditional definition avoids the awkward implication that S must be able to act in opposition to their interests, and fits with our intuition that freedom is about being able to do what you want. However, there are several issues with the conditional approach. First, a practical one: the conditional presented is a counterfactual. This poses problems for evaluating its truth-value when trying to ascertain whether a particular action φ was done freely by S. One could appeal to the “closest possible world” in which S wanted to not-φ, and try and establish whether S would have φed in it, but that presupposes both that (a) one has some objective metric to judge the closeness of possible worlds, and (b) one is able to determine what happens in a hypothetical world we are not in. Since we want an analysis of freedom to answer other questions (such as around moral responsibility), it is unhelpful to have a definition that cannot be evaluated.

Moreover, the conditional account does not fit with our intuitions about freedom. Take the following example, based loosely on McKenna and Coates (2023):

Suppose Victoria is a smoking addict. A friend offers her a cigarette. Victoria wants it very much and starts to smoke.

If Victoria had wanted to not smoke, then she would not have done so. But it seems strange to say that Victoria freely chose to smoke: our intuition is that she had no choice but to. These inconsistencies between the conditional view and our intuitions

The conditional account says that the individual in Frankfurt-style cases isn’t acting freely, and so does the categorical account I was originally planning on arguing for (it captures most closely what we seem to mean by and care about from free will, even though I think free will on that account is impossible). So, what’s the difference between the two, and why does the categorical account have advantages? Maybe I’d be arguing that the conditional account is also necessary (since it’s a pure weakening of the categorical account), but not sufficient? The categorical account gives the strictest possible set of requirements (maybe?), and it’s therefore necessary only if no weaker requirement is sufficient.

are made clearer by Frankfurt-style cases (FSCs), where – contradicting the conditional account – an individual seems to act freely even though they could not have acted otherwise, should they have so desired. Details vary, but here is a typical FSC (Frankfurt 1969, 835):

Suppose Scarlett has implanted a chip into Jones’s brain which lets her manipulate him. She hopes Jones will steal some money and monitors him, ready to switch on the chip at the first sign that he doesn’t wish to steal the money. Jones does decide to steal, so the chip remains inactivated.

Since Jones could not have failed to steal the money if he wanted to, the conditional account rules that he did not freely do so in this case. Yet our intuitions lead us to conclude that, since Scarlett did not actually intervene to manipulate him, Jones stole the money freely. We should, therefore

It’s not really as cut-and-dry as this. The SEP says that “the success of Frankfurt-style cases is hotly contested”, and there are several defences that proponents of alternate possibilities might give.

, seek alternatives to the conditional definition.

Actual-sequence accounts – arguing that in order for S to have φed freely, S’s φing must have been caused by a desire to φ – provide an elegant solution to FSCs. Jones’s inability to not steal is irrelevant to the actual-sequence account’s intuitive judgement that he stole freely (Frankfurt 1969, 837).

In cases like Victoria’s, the actual-sequence might appear to give the wrong verdict: Victoria’s desire to smoke caused her to do so, but it was not free. In particular, we might think that Victoria wishes that she were not addicted to cigarettes, and therefore does not want to want to smoke (a second-order desire). However, trying to overcome this failure by explicitly specifying that “S’s φing is free only if it is caused by a desire S desires to have” would be vulnerable to another counterexample, this time referencing conflicting third-order desires (and so on with higher-order desires).

We can successfully refine the actual-sequence account to address this entire class of objections by defining S’s φing as free only if, after reflection, both (a) $\exists n\mathbb{\in N}$ such that S endorses an n^th^-order desire which causes S’s φing; and (b) $\neg\exists m > n\mathbb{\in \ N\ }$ such that S’s m^th^-order desire conflicts with that n^th^-order desire (Frankfurt 1987, 164). In Victoria’s case, condition (b) is not met, since her second-order desire straightforwardly conflicts with the first-order one. More generally, this analysis fits neatly with our intuitions

that a person does something freely only if they wholeheartedly wish to do it, but also that superficially forced actions we willingly pre-commit to² can still be said to be free.

In light of these arguments, I conclude that the ability to have not-φed is not necessary in order to have φed freely – possessing the ability to do other than φ in the same circumstances would in fact be at odds with having agency at all. Whilst the conditional view is an improvement, it fails to deliver the correct judgements in instances of psychological compulsion or FSCs. As I have argued, the actual-sequence account provides the best analysis

The essay I would’ve liked to write is about how free will is a modelling trick and a relative concept. Frankfurt’s account of free will is a neat one, but I don’t think it really works, because it makes no claim that the higher-order desires arise freely, and I have a strong intuition that that’s what matters. Maybe “you” don’t need to be the first cause of your actions, but you do need to exert some causal influence over them. But now we are walking dangerously near to territory claimed by personal identity.

of the conditions required for S to do φ freely, in terms of internal consistency and its alignment with our intuitions.

References

Frankfurt, H., 1969, “Alternate Possibilities and Moral Responsibility”, The Journal of Philosophy, 66(23), 829–839.

———, 1987, “Identification and Wholeheartedness,” in Frankfurt, H., 1988, Philosophical Essays, Cambridge: Cambridge University Press.

McKenna, Michael and Coates, D. Justin, “Compatibilism”, The Stanford Encyclopedia of Philosophy (Winter 2023 Edition), Edward N. Zalta & Uri Nodelman (eds.)

Putnam, H., 1988, Representation and Reality, Cambridge, MA & London: Harvard University Press.

Smart, J.J.C., 1993, “An Outline of a System of Utilitarian Ethics”, in Utilitarianism: For and Against, Cambridge: Cambridge University Press.

Watson, G., 1982, “Introduction”, in Watson, G., (ed.), 1982, Free Will, Oxford: Oxford University Press.

Wolf, S., 1980, Asymmetrical Freedom, The Journal of Philosophy, 77(3), 151–166.

I will interpret “circumstances” broadly
: the conjunction $L \land P$ , where $L$ is the conjunction of all physical laws, and $P$ is the state of the universe at the point that φ is performed (Inwagen 1975, 191). We can say that an event $\varphi$ is “possible” just in case it is not impossible, i.e. $\neg\ \square\neg\ \varphi$ , where $\square\$ is the “necessary” operator. To put it another way, the conjunction $L \land P \land \neg\varphi$ must not be a contradiction. ↩︎
For example, I ask a friend to make me exercise tomorrow, because I know I will lack motivation, but will still want to want to exercise. ↩︎