Wednesday, February 08, 2006

Scientific notes

Some quick thoughts about science, inspired by this, via this.

A principle of reinforcement learning (for computers) is that an agent acts in an environment, and remembers in order to predict the future state of the environment, then (or at the same time) maximizes its total reward over its lifespan in the environment. Thus, learning systems solve problems embodied by the structure of their rewards. It seems to me that what science claims to do is almost the same thing: predict the future state of the (physical) environment given a certain action in a present state (some would prefer to omit the idea that actions exist). But science omits the other part of reinforcement learning: the maximization of reward. In science, there are no fixed rewards. Thus, there are no fixed problems.

How do we know that induction (essentially, science) leads somehow to justified true belief (knowledge)? Why should it be privileged over intuition or mysticism or revelation as a knowledge-seeking tool? As I understand Wittgenstein to say, there is nowhere to stand to judge such questions.

No comments: