Here is a philosophical puzzle with respect to actions:
(1) If A is an action, then there is an event, E, such that A is identical to E.
(2) S is the agent of action A only if S causes E (A = E).
(3) An event can only be caused by another event.
(4) Persons are not events.

Therefore:
(5) A person can never be the agent of an action.

Also:
(6) S is the agent of an action only if S is a person.

Therefore:
(7) Actions do not have agents.
But:
(8) Actions do have agents.

[Also, one could write down “living being” instead of “person” in (4) and (6) and still infer (7)]

Now, I do not think that I want to reject (1) – when asked what kind of thing an action is, I would say that it is an event. Neither do I want to reject (3), because it fits my intuitions about causation. Nor (4), because this also fits my intuitions about what a person is (or at least it fits my intuitions about what a person is not). I could, however, reject (2) and propose a slightly different view about agency:

S is the agent of an action A only if:
(i) there is an event E, such that when S performs A, E occurs (so it could be said that A is identical to E),
and (ii) a change C occurring in S’s body is either a part of E or identical to E (it could be said that C is also an event)

(i) and (ii) are necessary conditions for agency, but they are not sufficient; for S to be the agent of A other conditions must also be met:
(iii) S assumes a basic (or zero-level) responsibility for C – such that, to the question ‘Who did/produced C?’, S would reply with ‘I did it’ (or something similar). / or at least S would say that C belongs to her/him (basic ownership);
(iv) S assumes a reason-giving responsibility for E – such that S would accept the question ‘Why did you do A*?’ (where A* is a description of A which S accepts) as meaningful (even if S would reply ‘I did it for no particular reason’). [see Anscombe]

*

One could, of course, try to find counterexamples to this view about agency. For instance: If I set my alarm to sound at 6 in the morning before going to sleep (in the evening), my setting the alarm is not a part of the alarm sounding in the morning (since these are two different events, separated in time), so it could not be said that I made the alarm sound at 6 A.M. But this is counter-intuitive. To this I would reply: „Imagine that I was using a sand-glass for my alarm. The setting of the alarm would be a part of a larger event which would also include the flow of sand and the sounding of the alarm at the end. The ‘two events’ are not completely disconnected.”

Another example: S does A but does not want to admit that she did it, so there is no description of A, A*, such that S would accept the question ‘Why did you do A*?’ as meaningful (as for C, S might admit basic ownership of it, but not basic responsibility for it). Reply: „If S did A, she would accept the question as meaningful. Acceptance of a question is acceptance to provide reasons (or confess that you cannot do it, although it could, in principle be done). One might not want to provide reasons to a particular person (or to any other person), but still be able to provide reasons to herself.”

One last example: S rolls a 6 with one dice and is willing to provide reasons for rolling a 6. In short, conditions (i)-(iv) are satisfied, but we would not say that ‘rolling a 6’ was actually an action performed by S. Reply: „S did roll the dice (the throw is here our C, perhaps). We do not regard S as the agent producing the result of that roll because we know that it was a random roll. This means that we view the throw of the dice and the final roll of the dice which produced the result as two disconnected events. The exact same throw could have produced a different result if the dice was held in a different starting position by S, but S did not chose the starting position of the dice and cannot replicate throws. We tend to view the two events as causally disconnected (as if ‘luck’ was the main cause of the second event – the final roll of the dice which produced the result), although this is not the case. What seems to matter here is the way in which we regard different practices. If S scored a goal from great distance in the last seconds of a basketball match, we would grant agency to S for scoring the goal, although it might be impossible for S to replicate that throw. Putting a ball through a hoop is something which we (humans) train for, while ‘rolling sixes’ is not. Even if the basketball player did score „due to luck”, she could still attempt to answer the question ‘Why did you throw the ball in that way?’, while the Yahtzee player is not going to answer a similar question – ‘Why did you throw the dice in that way’. Some preexistent intuitions about agency are already in place when we consider whether an event is part of a larger event or not.”

Other examples could be produced, of course. My proposal for a particular way in which we could conceive agency and solve the initial puzzle as a result could be improved. The main point of this proposal might be that we do not look around us and see natural events at first and then try to figure out whether they have some agents or not by attempting to work out some causal chains and see whether they go back to some person or not. In a special case, a detective might recognize the result of an action and try to figure out who did it by following back some causal chains which might lead to the body of a particular person. Even so, there would usually be no doubt that someone is the agent of that action. In regular cases, anyway, we directly observe the actions of others.

The naturalist view, according to which these actions must also be included among other natural events comes only afterwards. One needs scientific education in order to accept our premise (1). The problem is that this view seems to make it difficult for us to conceive ourselves as responsible for our actions. The philosophical problem of free will is the locus communis of that tension, but my puzzle is meant to show that the source of that problem is deeper. One could talk of persons as causing events (by challenging either (3), or (4), or even both) and still leave it untouched, since causation seems to exclude responsibility. Strangely enough, lack of causation seems to exclude responsibility as well. I cannot be responsible for an event E if I was not E’s cause, so to speak. But if something going on in my body was the cause of E, I could not be responsible for E unless it was my choice for E to happen. Of course, my choice must have caused E, but then, if my choice was also caused by something else, then it was not really a choice. And if it was not caused by anything, then I did not make it. Nothing makes it mine.

At this point one might wonder how could my proposal of a different concept of agency escape such problems, if the vocabulary used to state conditions (i)-(iv) was translated into naturalist talk. If to assume some sort of responsibility with respect to an event was also an event and thus would figure in some causal chains, the trouble would surely resurface. It would appear that whether or not I assume a responsibility for an event E could not be something which I can chose to do. However, I do not think that the vocabulary used to state (i)-(iv) can be translated into naturalist talk. To assume some sort of responsibility with respect to an event is neither an event, nor an action. This does not mean that a naturalist story of how we started to view ourselves as agents of our actions could not be told. A link between responsibility and accountability could be used to find out some ancestral forms of responsibility in other species, perhaps. Maybe one could talk about responsibility (in its different versions) as being socially constructed. This, however, should indicate precisely that the vocabulary of agency cannot be completely translated into a “pure” naturalist vocabulary.