Reconciling Agent's Control with Determinism: A Re-evaluation of Moral Responsibility — Epoche C2
The case that made the problem, and what it was designed to show In "Alternate Possibilities and Moral Responsibility" (1969) Harry Frankfurt described a man who decides on his own to do a certain thing while a second man stands by with the power to make him do it and never has occasion to use that power. Frankfurt's target was a principle he stated and named in that paper — the principle of alternate possibilities, PAP — according to which a person is morally responsible for what he has done only if he could have done otherwise. In Frankfurt's own telling, Black wants Jones to perform a certain action; Black waits, does nothing if Jones is going to decide on his own to do it, and is ready to take effective steps if it becomes clear that Jones will decide otherwise. Frankfurt deliberately declined to specify the intervention, offering threats, a potion and hypnosis as equally serviceable. The familiar version in which Black has implanted a chip in Jones's brain and is set to trigger it at the first sign that Jones will vote the wrong way is a later sharpening rather than Frankfurt's text, useful because the chip removes any suggestion of mere persuasion. The structure is what matters. Jones deliberates, decides and votes exactly as he would have done had Black never existed: the actual causal sequence running to the vote contains no trace of Black. Yet in no alternative available to Jones does he do anything else, because Black has closed off every such alternative in advance. If we judge Jones responsible — and Frankfurt's claim is that we plainly do — then responsibility does not require the ability to have done otherwise, and PAP is false. Frankfurt offered a replacement: a person is not morally responsible for what he did if he did it only because he could not have done otherwise. The qualifier carries the whole load. What excuses is not the bare absence of alternatives but the fact that their absence is what explains the act. One thing has to be said plainly, because the contrary is often assumed and was asserted in an earlier version of this argument. Frankfurt-style cases do not sharpen the threat determinism poses to responsibility; they are the compatibilist's chief instrument against it. The classical incompatibilist argument runs through the ability to do otherwise, and Frankfurt cases are meant to remove that middle term. Why the ability to do otherwise does look incompatible with determinism That the ability to do otherwise is incompatible with determinism is a conclusion, not a datum. The argument for it is Peter van Inwagen's consequence argument, in An Essay on Free Will (1983). Write $\mathsf{N}p$ for "$p$, and no one has, or ever had, any choice about whether $p$". Two rules are used: if $p$ is a necessary truth then $\mathsf{N}p$; and a transfer rule for powerlessness, $$\mathsf{N}p, \quad \mathsf{N}(p \supset q) \quad \therefore \quad \mathsf{N}q.$$ Let $P_{0}$ be a complete description of the world at some instant before any present agent existed, and let $L$ be the conjunction of the laws of nature. Determinism is the thesis that $P_{0}$ together with $L$ entails every truth about the present, so for any present truth $q$ the conditional $(P_{0} \wedge L) \supset q$ is necessary, and hence so is $P_{0} \supset (L \supset q)$. The first rule gives $\mathsf{N}(P_{0} \supset (L \supset q))$. Nobody has any choice about $P_{0}$, which obtained before anyone was born, so the transfer rule yields $\mathsf{N}(L \supset q)$; nobody has any choice about $L$, so a second application yields $\mathsf{N}q$. Whatever I am now doing, I never had any choice about it. The argument leaks at the transfer rule. Thomas McKay and David Johnson showed in 1996 that it entails a principle it should not: agglomeration, the inference from $\mathsf{N}p$ and $\mathsf{N}q$ to $\mathsf{N}(p \wedge q)$. Since $p \supset (q \supset (p \wedge q))$ is a tautology, the first rule gives it an $\mathsf{N}$; transfer with $\mathsf{N}p$ gives $\mathsf{N}(q \supset (p \wedge q))$; transfer again with $\mathsf{N}q$ gives $\mathsf{N}(p \wedge q)$. And agglomeration has counterexamples. Suppose I can toss a coin and never do. "The coin does not land heads" is true and I had no choice about it, since tossing would not have ensured heads; the same holds of "the coin does not land tails". Their conjunction is equivalent to "the coin is not tossed", about which I plainly did have a choice. The transfer rule is therefore invalid as stated, and later versions of the argument replace it with a stronger notion of unavoidability that is closed under conjunction by construction. Most participants think some version survives the repair. The point here is narrower: the incompatibility of control-with-alternatives and determinism rests on a contested formal argument, not on an intuition. Has the Frankfurt case actually been made? That was the easy half. The hard half is that Frankfurt's case has a structural weakness the reconstruction below must survive. Black has to know, before Jones decides, which way Jones is going, so he needs a prior sign — a flicker of intention, a neural precursor, a blush. David Widerker, in "Libertarianism and Frankfurt's Attack on the Principle of Alternative Possibilities" (1995), posed the resulting dilemma. If the sign is deterministically connected to the decision, the case contains a deterministic chain running to the decision, and an incompatibilist who denies that Jones is responsible has a reason to hand that has nothing to do with alternatives. The case begs the question it was built to settle. If the connection is not deterministic — which is what the libertarian believes — the sign leaves it open that Jones will decide otherwise. Black cannot rely on it, and Jones retains an alternative: he could have shown the other sign. PAP is untouched. John Martin Fischer's reply, given at length in Responsibility and Control: A Theory of Moral Responsibility (1998) with Mark Ravizza, is that t