Explainability in systems: from AI to FM and back
In talk, I will discuss the topic of explainability in systems, with a specific emphasis on the intertwined concepts of causality and harm. Causal inference, as formulated by Halpern and Pearl, involves using structural equation models to understand causal relationships between the variables defining a system. This approach employs mathematical representations to capture the causal mechanisms underlying observed behaviours, allowing one to infer cause-and-effect relationships and make predictions about the consequences of interventions or changes to variables in a system. While causal inference and related concepts have emerged as a cornerstone in AI disciplines, I will showcase their expanded role in formal verification contexts. Beyond their applications in AI, I will examine how causal models have been embraced and integrated into Formal Methods (FM), for enhancing system transparency and reliability. Furthermore, I will provide some pointers to the reciprocal relationship between AI and FM, with an emphasis on how FM can contribute to AI’s quest for trustworthiness.
Tue 16 AprDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
15:00 - 15:50 | |||
15:00 25mTalk | A Logical Basis for the Verification of Message-Passing Programs Dutch Formal Methods Day 2024 Jorge A. Pérez University of Groningen Pre-print | ||
15:25 25mTalk | Explainability in systems: from AI to FM and back Dutch Formal Methods Day 2024 Georgiana Caltais University of Twente |