Abstract:
Despite the remarkable capabilities of modern deep learning, purely statistical AI systems remain opaque and difficult to align with human values. Neurosymbolic AI, integrating neural learning with symbolic reasoning, formal ontologies, and knowledge graphs, offers a principled path toward systems that are interpretable, verifiable, and trustworthy. Yet technical trustworthiness alone is insufficient: AI systems must also be designed with and for the humans: who use them. This panel brings together NeXAI researchers to explore both dimensions of neurosymbolic methods that make AI auditable and formally grounded, and human-centred design approaches that ensure AI systems earn user trust, respect diverse values, and remain genuinely useful across social contexts. Through five focused talks and a moderated panel, the session addresses: How do symbolic constraints improve explainability for end users? How do we design neurosymbolic systems that people can meaningfully interact with and oversee? What does responsible Human-AI Interaction look like when powered by structured knowledge?
Goal:
The main goal of the session is to explore how Neurosymbolic AI can combine technical trustworthiness with human-centered design to create AI systems that are explainable, auditable, verifiable, and genuinely useful to people. More specifically, the session aims to: Showcase current Neurosymbolic AI research combining neural learning with symbolic reasoning, ontologies, knowledge graphs, and formal methods. Explore how symbolic knowledge and constraints can improve explainability, transparency, and auditability for AI users. Examine the human side of trustworthy AI, including how users can interact with, understand, and oversee Neurosymbolic systems. Discuss applications in real-world and high-stakes domains, such as healthcare and process-oriented settings. Bridge technical and societal perspectives, bringing together researchers from AI, knowledge representation, reasoning, Human-AI Interaction, and application domains. Identify open research challenges and future directions through an interactive panel involving the audience.
Submission:
Participants submit open research challenges at the session start, fed into the panel debate.
Intended Audience:
AI researchers (ML, KR, formal reasoning); PhD students; industry practitioners in high-stakes domains; policymakers interested in auditable AI; interdisciplinary researchers from health, law, and social sciences.
Organizers:
