Deontological decision theory

This thesis consists of four papers on ethics and decision theory. Although the papers are closely related, each is written to make a standalone contribution. So, the thesis should not be read as one continuous project.

Ignoring the Improbable. Many problems in decision theory appear to be solvable if agents simply ignore some possibilities. The utility of this approach has given rise to a substantial number of theories endorsing discounting: ignoring states whose probability is below a particular threshold. This paper argues that ignoring possibilities—even extremely remote ones—comes at a hefty cost for one’s ability to make rational decisions. First, the approach is inescapably partition-sensitive: agents will undertake different acts depending on how the world is described. Second, agents become insensitive to differences in the probability of excluded states; they will be indifferent between taking a small risk and a much smaller risk. Third, agents become insensitive to differences in the value of outcomes in excluded states; they will be indifferent between risking a bad outcome and the same risk of a much worse outcome. And fourth, excluding a state affects the expected value of all acts to which the state is relevant, generating implausible prescriptions for peripheral acts; for instance, agents will not take any bet on the excluded state since they have assigned it a probability of zero.

Superiority Discounting Implies the Preposterous Conclusion. Many population axiologies avoid the Repugnant Conclusion (RC) by endorsing Superiority: Some number of great lives is better than any number of mediocre lives. But as Nebel shows, RC follows (given plausible auxiliary assumptions) from the Intrapersonal Repugnant Conclusion (IRC): A guaranteed mediocre life is better than a sufficiently small probability of a great life. This result is concerning because IRC is plausible. Recently, Kosonen has argued that IRC can be true while RC is false if small probabilities are discounted to zero. This paper details the unique problems created by combining Superiority with discounting. The resultant view, Superiority Discounting, avoids the Repugnant Conclusion only at the cost of the Preposterous Conclusion: Near-certain hell for arbitrarily many people is better than near-certain heaven for arbitrarily many people.

Filtered Maximization. According to moral absolutism, consequentialist considerations may justify sufficiently small risks of violating a duty but not sufficiently large risks. But finding a value function that accommodates these preferences is notoriously difficult: Seemingly, if some amount of consequentialist value outweighs a small risk, then some larger amount should be able to outweigh a larger risk. Critics have taken this difficulty to warrant rejection of absolutism generally; others have attempted to solve the problem by offering decision-theoretic models of absolutist decision making. I outline five desiderata that such a model must satisfy and demonstrate that none of the leading four theories satisfy all five. Then, I present an alternative: Filtered Maximization. This theory models absolutists as assigning “duty value” to outcomes and filtering out acts whose expected duty value is sufficiently low. Then, of those remaining, agents perform the act that maximizes expected value. I illustrate how Filtered Maximization satisfies all five desiderata, then conclude by discussing some implications for existing absolutist theories.

Superiority and Separability. Superiority is the view that there exists some pair of valuable objects x and y such that some quantity of x is better than any quantity of y; it is very plausible when x is an important good and y is trivial, such as in the Repugnant Conclusion. This paper shows that (given modest auxiliary assumptions) Superiority is incompatible with Separability—the principle that in comparing the value of two outcomes, we may ignore people whose welfare and existence are unaffected.