The principle of alternate possibilities has a long history of seeming obvious. If I could not have done otherwise, then what I did was not up to me, and it is hard to blame someone for what was not up to them. Harry Frankfurt built a case in which the first condition is false and the verdict about blame does not seem to follow.
Jones is deciding whether to shoot someone. Black wants him to, and has arranged things so that if Jones shows any sign of deciding not to, Black will bring about the decision and the act in Jones by a device. Jones does not show such a sign. He decides for his own reasons, and shoots. Black does nothing. Had Jones wavered, Black would have acted, so Jones could not have done otherwise.
Build the model with three variables. D says Jones decides to shoot, and is a given, true. B says Black acts, and Black acts exactly when Jones does not decide. A says the shooting occurs, and it occurs if either Jones decides or Black acts.
D = 1 Jones decides
B = ¬D Black acts only if Jones does not decide
A = D ∨ B the shooting occurs either way
At the actual values, D is 1, B is 0 and A is 1. Now run the but-for test of “Counterfactual Causation and the But-For Test” in Science, Induction and Causation. Flip D to 0. Then B becomes 1, because Black acts, and A is still 1. Flipping Jones's decision alone leaves the act in place, so the decision is not a but-for cause of the act. That is the formal content of “Jones could not have done otherwise”.
Definition · Holding a variable fixedTo test whether D is a cause once Black's inaction is held fixed, set B at its actual value, 0, and flip D with B kept there. If the act then changes, D is a cause holding B fixed, and B is the witness.
Hold B at 0 and flip D to 0. Now A = D ∨ B is 0 ∨ 0, which is 0. The act does not occur. With Black's inaction fixed, the act depends on Jones's decision, and the lab reports D as a cause holding B fixed. This is the structure of preemption: Black is a backup that was never used, and a backup never used does not take the credit from the decision that did the work.
Frankfurt's conclusion is that Jones is responsible for the shooting, though he could not have done otherwise, because the act issued from his own decision in the actual sequence of events, and Black played no part in it. If so, PAP is false. The ability to do otherwise was never what grounded responsibility; the actual source of the act was.
Has Black been described too kindly?
The case has been pressed hard. One reply says that Black must read some sign to know whether to act, and that the agent could have shown the sign or not, a small alternative that is enough for responsibility. Another says that if the sign reliably predicts a decision that Jones has not yet made, then determinism has been assumed about Jones's choice, and the case begs the question against the libertarian who denies it. The model leaves both open. Its equation B = ¬D gives Black a perfect reader, which is exactly what the objections dispute.
Two further points. The case does not make anyone free in the sense of the previous lesson, since it grants that Jones cannot do otherwise. A philosopher can hold that determinism removes the ability to do otherwise, as the consequence argument says, and still hold that this does not remove responsibility. The position is called semi-compatibilism. Second, the lab computes a verdict about causes in the model, and whether the model fairly describes Jones is the question the case turns on.