MIRIx: Introduction to decision theory

This meetup's open to everyone, but note that it'll be math-heavy and more structured than our usual Less Wrong format.

How would an AI make its decisions? We don't want it to be in a position where there's an "obvious" right course of action which it refuses to take, or that it could somehow be corralled into acting unwisely.

The study of how agents should, and do, make decisions is called decision theory and it extends far beyond the realm of artificial intelligence research. But the Machine Intelligence Research Institute (MIRI) believes the correct decision theory is a crucial part of the architecture, or specification, of an advanced machine intelligence.

We'll look at:

• Causal decision theory, the academically dominant normative decision theory

• Simple games where it seems to get the answer right, and others where it seems to get the answer wrong

• Alternative decision theories, in particular of the "updateless decision theory" flavour

• Why this matters in real life

MIRIxToronto is (or will be) one of several independent research projects loosely affiliated with the Machine Intelligence Research Institute ( https://intelligence.org/ ). We're in the process of spinning it off from the Toronto Less Wrong meetup. Don't worry, the regular Less Wrong meetups aren't going away!