About this talk
This talk, presented by Haishi Bai from Microsoft, delves into how autonomous systems arrive at decisions and the importance of explainable AI in fostering transparency and trust in machine reasoning. It examines varied approaches to artificial intelligence, including rule-based systems, expert systems, and machine learning, and discusses the challenges of model opacity that arise when decisions are derived from complex data patterns.
The session outlines the risks associated with opaque models, including reliance on unintended training data features and vulnerabilities introduced by prompt-based interactions. It also introduces explainable AI techniques, such as post-hoc analysis and white-box models, alongside the innovative concept of graded logic for modeling decision-making. The BACON project is highlighted as a key development in creating data-driven decision models that align with human reasoning, ensuring that interpretability and policy constraints are embedded within the decision process.
