One thing I dislike about certain thought-experiments (and empirical experiments!) is that they do not cleanly differentiate between actions that are best done in “player vs player” and “player vs environment” domains.
For example, a lot of the force of our intuitions behind Pascal’s mugging comes from wanting to avoid being “mugged” (ie, predictably lose resources to an adversarial and probably-lying entity). However, most people frame it as a question about small probabilities and large payoffs, without the adversarial component.
Similarly, empirical social psych experiments on hyperbolic discounting feel suspicious to me. Indifference between receiving $15 immediately vs $30 in a week (but less aggressive differences between 30 weeks and 31 weeks) might track a real difference in discount rates across time, or it could be people’s System 1 being naturally suspicious that the experimenters would actually pay up a week from now (as opposed to immediately).
So generally I think people should be careful in thinking about, and potentially cleanly differentiating, the “best policy for making decisions in normal contexts” vs “best policy for making decisions in contexts where someone is actively out to get you.”
One thing I dislike about certain thought-experiments (and empirical experiments!) is that they do not cleanly differentiate between actions that are best done in “player vs player” and “player vs environment” domains.
For example, a lot of the force of our intuitions behind Pascal’s mugging comes from wanting to avoid being “mugged” (ie, predictably lose resources to an adversarial and probably-lying entity). However, most people frame it as a question about small probabilities and large payoffs, without the adversarial component.
Similarly, empirical social psych experiments on hyperbolic discounting feel suspicious to me. Indifference between receiving $15 immediately vs $30 in a week (but less aggressive differences between 30 weeks and 31 weeks) might track a real difference in discount rates across time, or it could be people’s System 1 being naturally suspicious that the experimenters would actually pay up a week from now (as opposed to immediately).
So generally I think people should be careful in thinking about, and potentially cleanly differentiating, the “best policy for making decisions in normal contexts” vs “best policy for making decisions in contexts where someone is actively out to get you.”