Causal Decision Theory (CDT) and Evidential Decision Theory (EDT) are two leading theories of rational choice. Briefly and informally, CDT says that a rational agent does whatever in her view most effectively brings about her ends. EDT says that a rational agent does whatever in her view is the best evidence of what she wants.1

CDT and EDT are both intuitive but not both true. They conflict over cases like:

Newcomb’s Problem. You must choose between (o1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o_{1}$$\end{document}) taking only an opaque box (‘one-boxing’) and (o2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o_{2}$$\end{document}) taking both it and a transparent one in which $1000 (k\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k$$\end{document}) is visible (‘two-boxing’). You get to keep what you take. The opaque box already contains $1 M (m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document}) if and only if (S1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}$$\end{document}) a highly reliable predictor foresaw your taking only it. If (S2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}$$\end{document}) she foresaw your taking both boxes then the opaque box is empty. So the payoff schedule (in dollars, for which we assume linear increasing utility) is as follows (Table 1).

Payoffs in Newcomb’s Problem

Probably the most popular argument for one-boxing, and against CDT, is that one-boxing maximizes expected return. Since the predictor reliably foresees both one- and two-boxing, ‘one-boxers’ make on average just under m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document} per trial, whereas ‘two-boxers’ make on average more than k\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k$$\end{document} but considerably less than m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document} per trial. So followers of EDT can ask followers of CDT: ‘If you’re so smart, why ain’cha rich?’

To this argument, known as ‘Why ain’cha rich?’ or WAR, defenders of CDT have a simple reply. They say that its lower expected return has no bearing on the rationality of two-boxing, because two-boxers typically lack money-making opportunities that one-boxers typically enjoy. Most two-boxers face an opaque box that never contained any money, and most one-boxers face an opaque box that always contained $1 M. Call this the opportunity defence of CDT.2

This paper defends the following material conditional: if agents lack opportunities in another case, a ‘Frankfurt case’, then the opportunity defence fails for a version of Newcomb’s Problem that Sections 1–4 construct. For those who know the Frankfurt cases, the argument is that Black’s role in the relevant Frankfurt cases involves prediction in a way that makes those cases interpretable as Newcomb problems. (For those who do not know the Frankfurt cases, Section 2 outlines the relevant points.)

The conditional will prompt different people to draw different further conclusions. Anyone who maintains its antecedent will agree that the opportunity defence cannot blunt WAR. Anyone who maintains the opportunity defence will have to say that certain opportunities are available in Frankfurt cases. But even someone who is or becomes agnostic on both questions might find interest in the connection that the conditional establishes between the debate over practical rationality on which Newcomb’s problem bears, and the debate over free will on which Frankfurt cases bear.

1. The construction starts with a base case, then modifies it in three steps. Before describing the base case I’ll sketch the idea behind Frankfurt cases.

There is a debate over whether determinism is compatible with any kind of freedom that entails moral responsibility. One argument that they are incompatible turns on what Frankfurt called the:

Principle of Alternate Possibilities (PAP). A person is responsible for doing something only if he could have done otherwise.3

Frankfurt’s case against PAP involves examples where a person is morally responsible for something that she could not have avoided doing. Such Frankfurt cases originally took the following form.

Suppose someone – Black, let us say – wants Jones to perform a certain action. Black is prepared to go to considerable lengths to get his way, but he prefers to avoid showing his hand unnecessarily. So he waits until Jones makes up his mind what to do, and he does nothing unless it is clear to him (Black is an excellent judge of such things) that Jones is going to decide to do something other than what [Black] wants him to do. If it becomes clear that Jones is going to decide something else, Black takes effective steps to ensure that Jones decides to do, and that he does do, what [Black] wants him to do. Whatever Jones’s initial preferences and inclinations, then, Black will have his way… Now suppose that Black never has to show his hand because Jones, for reasons of his own, decides to perform and does perform the very action Black wants him to perform… In this example there are sufficient conditions for Jones’s performing the action in question. What action he performs is not up to him… He has no alternative but to do what Black wants him to do. If he does it on his own, however, his moral responsibility for doing it is not affected by the fact that Black was lurking in the background with sinister intent, since this intent never comes into play.4

But arguably, the case only threatens both PAP and relevantly similar principles if Black’s procedure meets certain constraints. After all, even if Jones can’t get around doing what Black wants, he could still try or choose not to do it. Black would then force his hand – but still, the fact that Jones could have thus tried or chosen might suffice to make him responsible for his actual, wholehearted performance.5 That may not save PAP itself: Jones still performs the act, is responsible for doing so and has no alternative to performing it. But it does immunize nearby principles, such as this variant of PAP: an agent is responsible for a performance only if she could have acted blamelessly in the circumstances.6 That principle seems well-placed to do what the incompatibilist wanted PAP to do.

In response, defenders of Frankfurt’s idea offer ‘prior sign’ cases where Black can intervene before Jones chooses whether to do what Black wants, and so also before he can try not to. Black can do this because he can observe some precursor of Jones’s choice. This example concerns the 2008 US Presidential election.

Because he dares to hope that the Democrats finally have a good chance of winning the White House, the benevolent but elderly neurosurgeon, Black, has come out of retirement to participate in yet another philosophical example… He has secretly inserted a chip in Jones’s brain that enables Black to monitor and control Jones’s activities. Black can exercise this control through a sophisticated computer that he has programmed so that, among other things, it monitors Jones’s voting behaviour. If Jones were to show any inclination to vote for McCain (or, let us say, anyone other than Obama), then the computer, through the chip in Jones’s brain, would intervene to assure that he actually decides to vote for Obama and does so vote. But if Jones decides on his own to vote for Obama (as Black, the old progressive would prefer), the computer does nothing but continue to monitor—without affecting—the goings-on in Jones’s head. Now suppose that Jones decides to vote for Obama on his own, just as he would have if Black had not inserted the chip in his head. It seems, upon first thinking about this case, that Jones can be held morally responsible for his choice and act of voting for Obama, although he could not have chosen otherwise and he could not have done otherwise.7

The base case is a ‘prior sign’ Frankfurt case. But its bearing on PAP is irrelevant. It won’t matter to my main argument whether Jones is morally responsible for what he does (although I shall revisit that point in Section 10). What matters is that Jones could not have chosen or done otherwise.

2. Start with a version of Fischer’s example—call it F1. Jones’s brain-state’s instantiating some neurological pattern N* is a necessary but not sufficient condition of Jones’s being about to decide (say, in the next seven seconds) to press ‘M’. Black has secretly implanted into Jones’s brain a chip that detects N*. If it does, it warns Black, who immediately intervenes through the chip to ensure that Jones chooses to vote for Obama (and does vote for Obama). As it happens, the chip never alerts Black, who never interferes, and Jones votes for Obama as he intended all along.

Grant the following:

(1) In F1 Jones lacks any opportunity to vote for McCain.

The intuition is that it suffices for foreclosure of an opportunity, that one’s choice be subject to the wrong kind of external control. It is hard to say precisely what counts as the ‘wrong kind’. Intuitively, persuading you not to buy something doesn’t count as denying you the opportunity to buy it. Nor does modestly raising its price. But neurosurgical intervention counts, as does a steep enough increase in price—like pointing a gun at your head. Still, it seems plausibly sufficient for Black’s having denied Jones the opportunity to vote for McCain that Black is intentionally and systematically ‘linking conditions so that all the conditions that are necessary for the opportunity to be acted upon are never jointly satisfied’.9 That does look like the wrong kind of external control.

In F1 the following conditions are nomologically necessary for Jones’s voting for McCain: (i) Jones’s brain-state instantiates N*; (ii) Jones doesn’t choose to vote for Obama. In F1 Black is systematically linking (i) and (ii) so that they are never jointly satisfied. If the chip detects that (i) is true then Black makes (ii) false. Black has therefore denied Jones the opportunity to vote for McCain in F1. So (1) is true.

Next consider case F2: the chip only monitors Jones. It alerts Black if Jones’s brain-state instantiates N*. But Black can’t influence Jones via the chip or otherwise. I claim:

(2) In F2 Jones has an opportunity to vote for McCain.

3. To summarize where we are: so far, I have introduced a ‘prior sign’ Frankfurt case (F1), and a variant case (F2) where Black is observing but not controlling the subject. The next case modifies F1 by changing the locus of Black’s control: instead of being able to manipulate his choice, Black can only manipulate the effect of his choice. I’ll argue that a comparison with F1 reveals that Jones lacks certain opportunities, but a comparison with F2 reveals that he now has certain others.

In F3 then, Black can intervene. But if the chip detects N* then Black interferes not with Jones but with the voting machine. He remotely alters its program so that pressing either button records a vote for Obama. As before, Jones votes for Obama without interference, because his brain-state never instantiates N*, and Black never intervenes.

If (1) and (2) are true then so are:

(3) In F3 Jones lacks any opportunity to vote (i.e. to record a vote) for McCain.

(4) In F3 Jones has an opportunity to press ‘M’.

The case for (4) is that F3 shares with F2 those features that make (2) true. What makes it true in F2 that Jones can vote for McCain is that no external agent is stopping him. In F2, the chip only records N*. Black does nothing to influence Jones’s subsequent vote. Equally in F3, if the chip records N*, Black does nothing to prevent Jones from pressing ‘M’. He intervenes if at all only on the causal connection between the pressing of ‘M’ and the subsequent recording of a vote. If anyone ever has an opportunity to do anything that she doesn’t actually do, then Jones has an opportunity in F3 to press ‘M’. So (4) is true.

4. So far, I have argued that a comparison with F2 and the base case shows that Jones has certain opportunities in F3 and lacks certain others. I’ll now modify F3, but not in a way that makes any structural difference to Jones’s opportunities. All that changes are the possible outcomes. The result is a Newcomb Problem.

This fourth case, F4, resembles F3 except that the machine doesn’t record votes but transfers money. Jones chooses whether to press ‘M’ or ‘O’. As things are initially set up, pressing ‘M’ has no effect. But pressing ‘O’ transfers $1000 to Jones’s account. As before, a necessary condition of Jones’s being about to choose to press ‘M’ is that his brain-state instantiates N*.

Again, a chip in Jones’s brain can reliably detect N*. But this time Black’s overriding intention is to reward restraint. More particularly, he wants to award $1 M to anyone who presses ‘M’. So if the chip detects N*, Black remotely reprograms the machine so that Jones’s pressing either button causes $1 M to be wired to Jones’s account, on top of the possible $1000 that Jones stands to gain by pressing ‘O’. Finally, we suppose that in F4 Jones knows about Black’s powers and intentions.

As it happens, Black doesn’t act. Jones presses ‘O’. His brain-state never instantiates N*. He ends up $1000 richer.

It is as true in F4 as in F3 that Black is systematically controlling Jones’s environment to prevent the joint realization of the necessary conditions for something: not for Jones’s voting for McCain, but for his realizing any outcome in which he presses ‘M’ and does not get $1 M. Jones’s knowing that and how Black is controlling his situation has no tendency to diminish this control. Nor does it give Black control over Jones’s choice between pressing ‘M’ and pressing ‘O’.

Claim: if (3) and (4) are true then so are:

(5) In F4 Jones lacks any opportunity to: press ‘M’ and not make $1 M.

(6) In F4 Jones has an opportunity to press ‘M’.

The case for (6) is that F4 shares with F3 those features that make (4) true in F3. As in F3, if the chip detects N*, Black does nothing to stop Jones from pressing ‘M’. Jones has an opportunity to press ‘M’ if anyone ever has an opportunity to do anything that she doesn’t do. So (6) is true.

5. F4 is a Newcomb Problem where pressing ‘M’ corresponds to one-boxing and pressing ‘O’ corresponds to two-boxing. The payoffs in F4 are as follows (Table 2):

Payoffs in F4

In F4 Jones chooses the two-boxing option (i.e. he presses ‘O’).

In F4, pressing ‘O’ dominates. If Black has detected N* then Jones gets at least $1 M. But pressing ‘O’ returns an additional $1000. If Black has not detected N*, then pressing ‘M’ leaves Jones empty-handed, but pressing ‘O’ returns $1000. Moreover, whether Black has already detected N* when Jones chooses is causally independent of Jones’s choice, because it precedes Jones’s choice by about seven seconds. CDT therefore prefers pressing ‘O’.

But what Jones chooses to do is evidentially relevant to whether Black has detected N*, given Jones’s evidence when he chooses. Certainly, if Jones chooses to press ‘M’, Black has detected N*. But if Jones chooses to press ‘O’, it is less than certain that Black has detected N*, and may be most unlikely according to Jones’s credences. If Jones thinks that Black detects N* on fewer than 99.9% of occasions on which the agent presses ‘O’, EDT recommends pressing ‘M’.11

This completes the construction of the advertised Newcomb Problem i.e. F4. Here as in all Newcomb Problems, it is foreseeable that the average return to ‘one-boxing’ (pressing ‘M’) exceeds the average return to ‘two-boxing’ (pressing ‘O’). At any rate this is so if Black detects N* on all occasions on which the agent presses ‘M’ and on fewer than 99.9% of occasions on which the agent presses ‘O’. ‘One-boxers’, who all make $1 M per trial, can therefore press the WAR objection against Jones and other ‘two-boxers’, who on average make less, possibly much less, than $1 M per trial.

6. The next step argues that in this Newcomb Problem the ‘opportunity defence’ is not available. Here is the informal argument. For formalization in a weak modal logic, see the Appendix.

(5) and (6) imply that Jones has an opportunity to press ‘M’ but not to press ‘M’ without making $1 M. But if Jones can in this sense press ‘M’ but can’t press ‘M’ without making $1 M, then he can get $1 M. Therefore, Jones has an opportunity to make $1 M. He cannot invoke the opportunity defence.

Or if he does, then he must explain which of (5) or (6) he rejects, and why. (6) looks unassailable; and I argued that (5) is true if (1) is. Hence my conditional conclusion: if in Frankfurt cases like F1 Black has deprived Jones of any opportunity to realize what Black wants to avert, then F4 is a Newcomb case where the opportunity defence is unavailable.

That concludes the main argument. I turn to three objections.

7. Nothing in these examples presupposes determinism about Jones’s choices, in the sense that the state of the world at any earlier time nomologically determines Jones’s choice. It is not true, for instance, that N* nomologically necessitates that Jones will choose to press ‘M’ in F4, nor that the absence of N* nomologically necessitates that Jones will choose to press ‘O’. The absence of N* at any time only necessitates that Jones will not choose to press ‘M’ in the next few seconds; this is nomologically consistent with his doing so later.

But even that limited determinism, not about Jones’s ultimate choice but about any choice he makes in a given time window, might seem to threaten (6). The objection is that when he presses ‘O’ Jones lacks any opportunity to press ‘M’, because his brain-state was not recently instantiating N*. Nor did he have that opportunity earlier if his brain-state never instantiated N*. So Jones can complain (contrary to (6)) that he never had an opportunity to press ‘M’.12

There are three responses. The most defensive response is that the objection would only appeal to someone who thinks that any pre-t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document} state of the world, that nomologically determines that one will not choose an option at t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document}, eliminates any opportunity to choose that option at t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document}. One might instead think that the only determining factors that are genuinely opportunity-depriving are those involving external impediments or external interference by another. On this view, the fact that your present choice is somehow determined by your past (internal) brain-state cannot by itself deprive you of any present opportunity to do whatever that brain-state rules out. Thus (apparently) Dennett, who grants in support of (5) that Frankfurt-style interference deprives Jones of an opportunity,13 whilst maintaining—in support of (6)—that mere determination-by-the-past does not.

[A] real opportunity is an occasion where a self-controller ‘faces’ (is informed about) a situation in which the outcome of its subsequent ‘deliberation’ will be a decisive (as we say) factor. In such a situation more than one outcome is ‘possible’ so far as the agent or self-controller is concerned; that is, the critical nexus passes through its deliberation. That is what we mean when we say that the outcome is up to the agent.14

Waiving that, the second point is that even if Jones lacked any opportunity to press ‘M’ when he chose to press ‘O’, it doesn’t follow that he lacked any opportunity to press ‘M’. When he pressed ‘O’ he could instead have waited. And there need have been nothing about the state of his brain or of the world at that time that was nomologically inconsistent with his pressing ‘M’ later. So even if Jones lacked any opportunity to press ‘M’ when he chose to press ‘O’, he had an opportunity to press ‘M’, and he had an opportunity to make $1 M.15

The third, more tentative point is that it may be unnecessary to suppose that the prior appearance of N* is a necessary condition for Jones to press ‘M’. Suppose instead that N* is ‘nearly’ necessary: at the outset there is a very high chancep\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p$$\end{document} that his brain-state instantiates N* at some later time, given that he does at some later time choose to press ‘M’; and that there is a lower chance q<p-km\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q < p - \frac{k}{m}$$\end{document} that his brain-state ever instantiates N*, given that he chooses to press ‘O’. If these conditional chances control the corresponding frequencies then this indeterministic case is a Newcomb Problem where pressing ‘M’ generates a foreseeably higher average return than pressing ‘O’. If Jones makes $1 K, can he now plead the opportunity defence?

Clearly (6) still holds, because now Jones’s choice is undetermined by anything in its past. Everything turns on whether (5) holds i.e. on whether, in this indeterministic case, Black has deprived Jones of an opportunity to: press ‘M’ and not make $1 M. Has he? To see why he has, compare these cases.

(a) Black shuts Jones in a room from which the only exit is a door that Black has secured with a steel padlock, and Jones doesn’t have a key.

(b) Black shuts Jones in a room from which the only exit is a door that Black has secured with a combination lock with 100 million settings, and Jones doesn’t know the combination.

(6) can therefore tolerate some injection of chanciness into F4. Black can ensure that Jones does not achieve this or that outcome without making anything physically impossible. For Black to do that, it suffices that (i) it doesn’t happen because (ii) Black has made it sufficiently unlikely. But then even in this indeterministic version of F4, Black has eliminated Jones’s opportunity to get anything other than $1 M by pressing ‘M’. Since Jones has an opportunity to press ‘M’, it follows that he has an opportunity to make $1 M, so cannot plead the opportunity defence.

Somebody might object that there is, in this indeterministic version, something that Jones knows how to do—pressing ‘M’—such that, if he were to do it, then he would be falsifying the material conditional o1→m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o_{1} \to m$$\end{document}, at least on a non-back-tracking reading of ‘If he were to do it…’.17 But then in (b), there is something that he knows how to do—entering the sequence 01271756—such that, if he were to do it, then he would be opening the combination lock. If it is consistent with the latter that Black has locked Jones into the room in (b) then it is consistent with the former that Black has locked Jones out of falsifying o1→m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o_{1} \to m$$\end{document} in F4.

For that matter there is, on the non-back-tracking reading, something that Jones knows how to do in the original Frankfurt case—pressing ‘M’—such that if he were to do it, he would be voting for McCain. So anyone who takes this line against (5) must also reject (1). The objection is therefore impotent against the conditional: if Jones lacks the opportunities that he is supposed to lack in ‘prior sign’ Frankfurt cases, then the opportunity defence fails for both F4 and this indeterministic version of it.

A chancy case like this may be feasible. In a recent study, fMRI subjects would ‘fixat[e] on the centre of [a] screen where a stream of letters was presented. At some point, when they felt the urge to do so, they were to freely decide between one of two buttons, operated by the left and right index fingers, and press it immediately’.18 Experimenters found that:

two brain regions encoded with high accuracy whether the subject was about to choose the left or right response prior to the conscious decision… The first region was in frontopolar cortex, BA10. The predictive information in the fMRI signals from this brain region was already present 7 s. before the subject’s motor decision.19

8. A second objection is that F4 is not a Newcomb case at all. A genuine Newcomb case involves exactly two options: ‘one-boxing’ and ‘two-boxing’. But F4 offers Jones three options at any time. He can press ‘M’ now; he can press ‘O’ now; and he can wait before choosing.

This makes a difference because CDT seems to recommend waiting. After all, the effect of waiting can only be to make N* more likely to occur at some time before Jones chooses what to press. But F4 is a Newcomb case only if CDT recommends pressing ‘O’.

But even if CDT sometimes recommends waiting, this doesn’t help CDT. Whether or not F4 is strictly a Newcomb case, followers of CDT will wait either forever (and never win anything) or until the marginal costs of waiting exceed the marginal benefits of increasing the probability that N* is at some point realized in their brains, and at that point press ‘O’ and end up with $1000. In either case those agents end up worse than followers of EDT, and in either case they can’t complain that they lacked the opportunity to win $1 M.

Besides, it isn’t clear that CDT does recommend waiting. If (as we might suppose) N* only ever occurs within a fixed time interval preceding the subject’s choice, there is no reason to expect waiting to increase the probability that N* precedes whatever choice the subject ultimately makes. Compare: my space capsule only deploys a parachute when it is a minute away from landing. Waiting before starting my descent (rather than descending immediately) doesn’t make it more likely that the parachute is deployed at some time before landing.

But if CDT doesn’t recommend waiting (and of course neither does EDT) then the only plausible options at any time are pressing ‘M’ and pressing ‘O’, so F4 is after all a Newcomb case.

9. The third objection involves a notion of opportunity that falsifies (5).

C-opportunity: At any time:

You have a counterfactual opportunity (a C-opportunity) to realize the truth of a propositionp\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p$$\end{document}by choosingo\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o$$\end{document} iff o\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o$$\end{document} is an option at that time such that if you had at that time realized o\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$o$$\end{document}, then p\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p$$\end{document}would have been true.21

On this definition of opportunity, (5) is false when Jones chooses to press ‘O’. At that time, he has the option to press ‘M’. And if—in the non-back-tracking sense—Jones had then pressed ‘M’, it would have been true that: he presses ‘M’ and fails to make $1 M. Jones therefore has a C-opportunity to press ‘M’ and fail to make $1 M.

Similarly, if opportunities are just C-opportunities then Jones lacks any opportunity to make $1 M. This is because when he chooses, Black has not detected N* and has not transferred $1 M to Jones’s account. These facts are causally independent of what Jones chooses. So if—in the non-back-tracking sense—Jones had chosen either option, it would not have been the case that he made $1 M. Jones therefore lacks a C-opportunity to make $1 M.23

I concede that if opportunities are C-opportunities then (5) is false. But I deny that this undermines the argument. I defended the material conditional: if Jones lacks an opportunity to defy Black in a Frankfurt case like F1, then he has an opportunity to make $1 M in a Newcomb case like F4. If opportunities are C-opportunities then the consequent of this conditional is false, but so is its antecedent. The conditional is still true.

The antecedent is false in F1 because the implant has not detected N* when Jones chooses to vote for Obama. So Black is not intervening. This fact is causally independent of Jones’s choice. So at the closest worlds at which Jones chooses to vote for McCain, Black is not intervening. At these worlds Jones presses ‘M’ and votes for McCain. So Jones has a C-opportunity to vote for McCain. Anyone who thinks that opportunities are C-opportunities should deny that Black eliminates them in ‘prior sign’ Frankfurt cases.24 But this is no objection to the conditional defended here: if Jones lacks an opportunity to defy Black in a Frankfurt case like F1, then he has an opportunity to make m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document} in a Newcomb case like F4.25

10. The next step in this dialectic – which I cannot pursue at length – turns on a question that this conditional raises. We saw two kinds of opportunity in play: C-opportunity, and the kind of opportunity that Black’s presence, intentions and capacities are denying to Jones in the Frankfurt cases—call it ‘F-opportunity’.26 The question is whether lack of C-opportunity to do better is by itself sufficient for exculpation. If so, then Jones cannot be blamed for making only $1000, because he lacks any C-opportunity to make more. On the other hand, if lack of C-opportunity is insufficient, then the opportunity defence is at best incomplete, because as we saw, it appeals only to Jones’s lack of C-opportunity, not to any lack of F-opportunity.

One prima facie basis for thinking that lack of C-opportunity cannot by itself be exculpatory arises from reflection on something like the original Frankfurt case (not the ‘prior sign’ cases). Suppose that Black plans to wait until Jones has decided what to do, and then if necessary—if he has decided to vote for McCain—to coerce or force him into voting for Obama. Jones then lacks any C-opportunity to vote for McCain, but he is still responsible for voting for Obama if he does it wholeheartedly. Similarly, we might suspect that someone who autonomously chooses to exercise, and does exercise, the ‘two-boxing’ option in the Newcomb Problem is responsible for his (relative) impoverishment. (Of course this is at best a sketch of an argument).27

It might also help to examine moral Newcomb Problems: cases like F4 where the payoffs carry moral significance. The literature hasn’t covered these extensively, but one notable recent treatment is due to MacAskill, who offers the following:

Box A is opaque; Box B, transparent. If the Predictor predicts that you choose Box A only, then he puts one wish into Box A. With that wish, you would save the lives of one million terminally ill children. If he predicts that you choose both Box A and Box B, then he will put nothing into Box A. Box B—transparent to you—contains a stick of gum. You have two options only: Choose Box A, or choose both Box A and Box B.28

Obviously there is more to be said about the interplay of ethical, decision-theoretic and forensic reasoning that this case involves, and also about whether the conditional that this essay defends serves better as the major premise of a modus ponens or as the major premise of a modus tollens (if either). Still, the conditional itself is true: whatever kind of opportunity is being foreclosed in the Frankfurt cases is not being denied to two-boxers like Jones in the Newcomb case that we studied. And I hope that it gets us closer to resolving at least one of the two long-standing philosophical problems that it connects.