This is a past event
Abstract: Recently, AI has produced impressive advances mostly by following a "standard model" based on data-driven exploration of large solution spaces and optimisation toward a well-defined objective function. Arguably, this process model has little in common with human intelligence, where objectives are often vaguely defined and conflicting, satisficing behaviour is often more important than optimal behaviour, and improvements occur in a non-linear fashion, often based only on very little additional experience.
In this talk, I propose an alternative view of intelligence that emphasises diversity among the components and agents in an intelligent system. I argue that enabling more human-like intelligence requires understanding how different entities that use different local representations and have different objectives can interact in a meaningful way by exploring how they can incrementally benefit from each other's "intelligence". I will give two examples from our recent work where we are trying to enable such diversity, and which highlight the challenges and opportunities of making AI methods more diversity-aware. One is taken from the area of recommender systems, where we are looking at how we can enable algorithms to cater for diversity. The other looks at information sharing among agents with heterogeneous ontologies, and focuses more on how agents can "translate" others' knowledge to their own representations in ways that help them achieve their objectives.
Bio: Michael is a Senior Lecturer at the School of Informatics at the University of Edinburgh and Director of the Centre for Intelligent Systems and their Applications. He works in agents and multiagent systems and has broader research interests in artificial intelligence and social computation. He is currently involved in two major research projects, the SmartSociety project and the ESSENCE Initial Training Network, which he also coordinates.
- Speaker
- Michael Rovatsos
- Hosted by
- Adam Wyner
- Venue
- Meston 311