Skip to yearly menu bar Skip to main content

Workshop: Workshop on Formal Verification of Machine Learning

Invited Talk 2 (Anton Dahbura): Undeterminism and the AI Uncertainty Principle

Anton Dahbura


As AI/ML-based systems are deployed in increasingly autonomous and safety-critical applications, there is increasing concern for many reasons about the behaviors of the ML in the wild, perhaps most of all because the problems for which ML is applied are so complex that it’s not possible to know in all cases what the ML will do. In this talk I will introduce the concept of undeterminism, extracted from my experience in developing techniques for communication protocol conformance testing, and argue that undeterminism yields inherent uncertainty. I show why systems cannot have certainty and ML simultaneously and that there is a clear set of tradeoffs that should guide research in formal methods for ML for the foreseeable future.

Chat is not available.