Duncan MacIntosh

Hume said that the reasons that determine the rationality of one's actions are the desires one has when acting: one's actions are rational iff they advance these desires. Thomas Nagel says this entails calling rational, actions absurdly conflicting in aims over time. For one might have reason, in one's current desires, to begin trying to cause states one foresees having reason, in one's foreseen desires, to prevent. Instead, then, real reasons must be timeless, so that current and foreseen reasons cannot conflict. I say the desire theory does not have absurd consequences. A rational agent's desires would rationally evolve, never requiring actions conflicting in aims over time, except where it was instrumentally rational for her to change in her desires, whence such conflicts are rationally appropriate. Further, whatever sorts of things count as real reasons, since reasons can rationally require their own revision, they cannot be necessarily timeless.

While Autonomous Weapons Systems have obvious military advantages, there are prima facie moral objections to using them. By way of general reply to these objections, I point out similarities between the structure of law and morality on the one hand and of automata on the other. I argue that these, plus the fact that automata can be designed to lack the biases and other failings of humans, require us to automate the formulation, administration, and enforcement of law as much as possible, including the elements of law and morality that are operated by combatants in war. I suggest that, ethically speaking, deploying a legally competent robot in some legally regulated realm is not much different from deploying a more or less well-armed, vulnerable, obedient, or morally discerning soldier or general into battle, a police officer onto patrol, or a lawyer or judge into a trial. All feature automaticity in the sense of deputation to an agent we do not then directly control. Such relations are well understood and well-regulated in morality and law; so there is not much challenging philosophically in having robots be some of these agents — excepting the implications of the limits of robot technology at a given time for responsible deputation. I then consider this proposal in light of the differences between two conceptions of law. These are distinguished by whether each conception sees law as unambiguous rules inherently uncontroversial in each application; and I consider the prospects for robotizing law on each. Likewise for the prospects of robotizing moral theorizing and moral decision-making. Finally I identify certain elements of law and morality, noted by the philosopher Immanuel Kant, which robots can participate in only upon being able to set ends and emotionally invest in their attainment. One conclusion is that while affectless autonomous devices might be fit to rule us, they would not be fit to vote with us. For voting is a process for summing felt preferences, and affectless devices would have none to weigh into the sum. Since they don't care which outcomes obtain, they don't get to vote on which ones to bring about.

Scientific Realists argue that it would be a miracle if scientific theories were getting more predictive without getting closer to the truth; so they must be getting closer to the truth. Van Fraassen, Laudan et al. argue that owing to the underdetermination of theory by data (UDT) for all we know, it is a miracle, a fluke. So we should not believe in even the approximate truth of theories. I argue that there is a test for who is right: suppose we are at the limit of inquiry. Suppose that we then have all the logically possible theories that are adequate to all the actual data. If they all resembled in their theoretical claims, since one of them must be true, all of them would then resemble it, whichever it is. We would thus be justified in saying they all approximated the truth in the degree to which they co-resembled. If they don't all co-resemble, the SRs are wrong; more predictive theories are not necessarily closer to the theoretical truth. Prior to the limit, if, in spite of our best efforts to the contrary, all the theories we can make adequate to current data tend to co-resemble, we have inductive warrant for thinking more predictive theories are closer to the truth. If they don't resemble, we have inductive warrant for thinking that more predictive theories are not necessarily closer to the truth.

David Gauthier thinks agents facing a prisoner's dilemma ('pd') should find it rational to dispose themselves to co-operate with those inclined to reciprocate (i.e., to acquire a constrained maximizer--'cm'--disposition), and to co-operate with other 'cmers'. Richmond Campbell argues that since dominance reasoning shows it remains to the agent's advantage to defect, his co-operation is only rational if cm "determines" him to co-operate, forcing him not to cheat. I argue that if cm "forces" the agent to co-operate, he is not acting at all, never mind rationally. Thus, neither author has shown that co-operation is rational action in a pd.

If one can get the targets of one's current wants only by acquiring new wants (as in the Prisoner's Dilemma), is it rational to do so? Arguably not. For this could justify adopting unsatisfiable wants, violating the rational duty to maximize one's utility. Further, why cause a want's target if one will not then want it? And people "are" their wants. So if these change, people will not survive to enjoy their wants' targets. I reply that one rationally need not advance one's future wants, only current ones. Furthermore, rational choice seeks not utility (the co-obtaining of a want and its target), but satisfaction (the eventual obtaining of what is now wanted) -- otherwise, it would be irrational to care now about what happens after one dies. Finally, persons survive "rational" changes of values. Thus reflection on the rational revision of values illuminates the conditions on personal identity and the bases and aims of rational choice.

David Gauthier suggested that all genuine moral problems are Prisoners Dilemmas (PDs), and that the morally and rationally required solution to a PD is to co-operate. I say there are four other forms of moral problem, each a different way of agents failing to be in PDs because of the agents’ preferences. This occurs when agents have preferences that are malevolent, self-enslaving, stingy, or bullying. I then analyze preferences as reasons for action, claiming that this means they must not target the impossible, they must be able to be acted on in the circumstances, their targets must be attainable, and having the preferences must make their targets more likely. For groups of agents to have a distribution of preferences, their preferences must jointly have those four features, this imposing a kind of universalizability requirement on possible preferences. I then claim that, if all agents began with preferences satisfying these requirements, their preferences would not be of the morally problematic sort (on pain, variously, of circularity or contradiction in the specification of their targets). Instead, they would be either morally innocent preferences, or ones which put the agents in PDs. And it would then be instrumentally rational for the agents to prefer mutual co-operation. Thus if all agents initially had rationally permissible preferences and made rational choices of actions and preferences thereafter, they would never acquire immoral preferences, and so never be rationally moved to immoral actions. Further, the states of affairs such agents would be moved to bring about would be compatible with what Rawls’ agents would chose behind a veil of ignorance. Morality therefore reduces to rationality; necessarily, the actions categorically required by morality are also categorically required by rationality.

Theories of practical rationality say when it is rational to form and fulfill intentions to do actions. David Gauthier says the correct theory would be the one our obeying would best advance the aim of rationality, something Humeans take to be the satisfaction of one’s desires. I use this test to evaluate the received theory and Gauthier’s 1984 and 1994 theories. I find problems with the theories and then offer a theory superior by Gauthier’s test and immune to the problems. On this theory, it is rational to treat something different as the aim when doing so would advance the original aim. I argue that the idea that this would be irrational bad faith entails contradictions and so is false, as must be theories saying that rationally we must always treat as the aim the bringing about of objectively good states of affairs or obeying a universalizable moral code. (Note: the published version differs somewhat from the version on the website of the Center for Ethics and the Rule of Law; please quote from the published version.)

Gauthier claims: (1) a non-maximizing action is rational if it maximized to intend it. If one intended to retaliate in order to deter an attack, (2) retaliation is rational, for it maximized to intend it. I argue that even on sympathetic theories of intentions, actions and choices, (1) is incoherent. But I defend (2) by arguing that an action is rational if it maximizes on preferences it maximized to adopt given one's antecedent preferences. (2) is true because it maximized to adopt preferences on which it maximizes to retaliate. I thus save the theory that rational actions must maximize, and extend it into the rational criticism of preferences.

This paper reviews six attempts to give cooperative solutions to Prisoners Dilemmas: symmetry (agents are in identical situations, so should choose the same way, so should both choose cooperation because that’s better for each), mechanism (each agent should delegate the decision to a machine which will choose cooperation for them provided the other does likewise), inducement (the agents should make a side bet which pays off only upon both cooperating), resolution (each agent should resolve to cooperate, then act on the resolution), alternative principle (rationality requires not dominance reasoning and so defecting, but constrained maximization or resolution compliance or “sophisticated” choosing, and so cooperation), and preference revision (each agent should acquire conditional preferences to cooperate, then cooperate provided the other agent has acquired similar preferences). It distinguishes one-stage from two-stage games, as well as endogenously available solutions (ones available to agents in the circumstances of the game merely due to the nature of rationality) from solutions requiring exogenous devices. It contrasts Edward McClennen’s endogenous solution, which is some combination of resolution and alternative principle solution, with David Gauthier’s endogenous alternative principle solution. It argues that all have problems, ones best solved by the endogenous preference revision solution defended by Duncan MacIntosh.