580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can philosophical and social science perspectives improve explanations in AI beyond simplified approximations?
Research on explainable AI (xAI) primarily produces simplified models approximating complex decision functions but often lacks alignment with philosophical and social science theories of explanation. This theme investigates how integrating insights from philosophy, cognitive science, and sociology about contrastive, selective, and interactive explanations can create more meaningful, trustworthy, and contestable AI explanations suited for diverse stakeholders.
Key finding: This paper identifies that current xAI methods mostly yield simplified approximations akin to scientific models, which provide locally reliable but potentially globally misleading explanations. Drawing on the social sciences,... Read more
Key finding: Building on the need for user-centered explanations highlighted in xAI, this work expands an Explanation Ontology (EO) to semantically represent various explanation types, connecting explanation outputs, data, and system... Read more
Key finding: This paper presents a Defeasible Logic Programming server enabling agents to engage in contextual, defeasible argumentative reasoning using both shared and private knowledge. By facilitating reasoning that mirrors human... Read more
2. What formal logical frameworks can unify knowledge and belief reasoning for agents, and how do they address knowledge representation in multi-agent contexts?
This research explores epistemic logic and argumentation frameworks to formally characterize knowledge and belief among agents, enabling reasoning about information states, higher-order knowledge, and dialogues in multi-agent systems. It investigates formal languages, semantics, and proof systems that describe how agents acquire, share, and contest knowledge, as well as how argumentation can represent defeasible and non-monotonic reasoning within resource-constrained and interactive settings.
Key finding: This chapter introduces formal epistemic logic frameworks including syntax, semantics, and proof systems for reasoning about knowledge and beliefs of multiple agents. It explains how public announcements and common knowledge... Read more
Key finding: The paper reformulates classical logic-based argumentation (ClAr) to enable resource-bounded agents to reason dialectically in real-world dialogues without prohibitive computational costs. It distinguishes premises assumed... Read more
Key finding: Here, Argumentation Logic (AL) is developed by connecting classical propositional logic with argumentation theory, redefining reductio ad absurdum to handle defeasible reasoning. AL avoids trivialization by inconsistent... Read more
Key finding: This work revisits David Lewis's foundational but informal theory of common knowledge based on 'reasons to believe' instead of mental states. It analyzes existing formalizations, identifies gaps, and proposes a novel... Read more
3. How can hybrid and multi-level knowledge representation frameworks integrate symbolic, sub-symbolic, and defeasible reasoning for AI systems?
This area studies frameworks that combine multiple representation approaches (symbolic, sub-symbolic, defeasible, ontological) to more effectively model real-world knowledge and reasoning, including uncertainty, exceptions, multi-level classifications, and human-like cognitive processes. It analyses formal languages, ontologies, and cognitive architectures that address scalability, expressivity, and normative constraints while supporting complex reasoning tasks in AI systems.
Key finding: This paper provides a comprehensive overview of knowledge representation techniques used in AI, including implicit and explicit knowledge, and categories such as rules, ontologies, frames, semantic networks, and logic. It... Read more
Key finding: DLN extends description logics with prioritized defeasible inclusions to represent prototypical knowledge and exceptions. It supports overriding of general axioms by exceptions without increasing computational complexity... Read more
Key finding: The paper argues that Conceptual Spaces serve as an intermediate representational level bridging symbolic, sub-symbolic, and diagrammatic representations within cognitive architectures. This framework alleviates... Read more
Key finding: This work proposes an OWL-based vocabulary and integrity constraints grounded in the axiomatic MLT theory to support multi-level ontologies with classes of classes (metaclasses), facilitating representation of domains like... Read more
Key finding: KRRT is a web-based tutoring system designed to teach first-order logic as a knowledge representation and reasoning language, focusing on formalization from natural language and constructing formal proofs. It integrates... Read more