Superintelligence: Paths, Dangers, Strategies

by Nick Bostrom

One advantage of relating learning problems from specific domains to the general problem of Bayesian inference is that new algorithms that make Bayesian inference more efficient will then yield immediate improvements across many different areas.