PDTAI: Provability, Decision Theory and Artificial Intelligence

PDTAI: Provability, Decision Theory and Artificial Intelligence: Modal Combat: a sandbox for decision theory Seminar | October 6 | 5:10-6:30 p.m. | 732 Evans Hall Speaker: Patrick LaVictoire, Machine Intelligence Research Institute Sponsor: Department of Mathematics How might we study the dynamics of intelligent agents on various decision problems, when we can’t currently write a general artificial intelligence? It turns out that there’s a simple way to encode some virtual agents and environments in the modal logic of provability (GL); we will present a polynomial-time algorithm which decides the actions of a certain class of “modal agents” (which use formal verification and vast computing power) on a certain class of “modal decision problems”. We will then explore some of the self-referential phenomena we’ve found there, and show an optimality theorem for a…


Link to Full Article: PDTAI: Provability, Decision Theory and Artificial Intelligence