It is a little difficult to respond to the internal and external concerns raised by Geoffrey P. R. Wallace in his generous review of my book, because I mostly agree with them. The first internal problem Wallace raises is the spare way in which I model the costs to torture. At the risk of retreating to the last refuge of the formal modeler, one can add Wallace’s costs so that the cost parameter is high enough to make torture a dominated choice, even when it compels information. In fact, I argued so recently myself (“Review Essay: Interrogating Torture,” Journal of Politics, 79(4) 2017) and thus agree that in the real world, these costs were much higher than proponents anticipated, high enough to make torture counterproductive. The problem is that, for torturers and their supporters, these costs were perceived to be low.

Moreover, the assumption that interrogators perceive the costs as always outweighing the informational benefits would mean that torture would never happen. That, unfortunately, is not the case empirically, and the goal of the model is to examine the logic that generates those empirical results. This does not mean, however, that the costs are “minimal”; I actually assume that the costs of torturing an innocent are greater than the costs of failing to torture a detainee with information who refuses to divulge it. Nor are the costs to torture always less than not torturing. If the interrogator is confident enough that the detainee is innocent, then the interrogator prefers not to torture just because of those costs.

The second concern internal to the model is the single-shot structure of the game when torture is an “inherently repetitive process’’ going beyond the detainee being tortured. Once again, I can agree that the torture of an individual detainee is enmeshed in a much larger web of motivations and that there are unanticipated downstream problems. But my goal was to model the proponent argument that torture generates valuable information at little short-term cost. I did so for two reasons. First, neither the pro-torture philosophical and legal arguments nor the actual decision makers considered what the next administration might have to do in response. Second, if it can be shown that their argument fails on even those narrowly argued grounds, then it necessarily fails on the wider terrain when other costs and considerations are added.

Finally, there is the legitimate question about whether game theory is the appropriate approach for “obtaining a fuller and more convincing picture” of torture. A fuller picture of what aspect of torture, though? Once more I can agree that there are many aspects of torture that would be “better . . . explained by a closer and contextualized treatment of the underlying actors and dynamics at play” than by a formal model. As Wallace kindly acknowledges, however, my goal was to expose and draw inferences about “the most salient dynamics” in the proponent pain-information logic of interrogational torture. Game theory is well suited for this, admittedly limited, purpose. Still, reminders about limits are good, and I am grateful to Wallace for pushing the point that the problems associated with normative aversions, complexity, and other aspects of torture do indeed need to be questioned carefully.