3

A tutorial presented at AAAI-24 on AI evaluation that focuses on estimating capabilities and creating capability profiles of AI systems (e.g., reinforcement learning agents and large language models) using a Bayesian framework.

We use directed acyclic graphs (DAGs) to investigate the chances that two variables are causally connected, correlated, and that a covariate is inducing a correlation when controlled for.

A justification of the Inference to the Best Explanation (IBE) by finding conditions under which the best explanation of evidence can provide a confirmatory boost for the hypotheses under consideration.