Keynote

Architecting AI Systems for Dialogue and Other Complex Tasks

Kristinn R. Thórisson, Reykjavik University / IIIM

Abstract:

If we wanted to build a system that could communicate with people, using the natural modes of communication that humans use with each other, such as speech, gesture, body language, etc., what would the underlying system look like? This is a big undertaking which obviously has not been achieved to date, although some examples exist of systems capable of parts of this task. The question revolves around many a multitude of issues, but ultimately one must address the fundamental issue of architecture. The architecture of an AI system enables – or limits – all else in such systems. After half a century of AI research, enormous gaps can still be seen when comparing artificial and natural intelligence. The differences in capabilities are readily apparent on virtually every scale we might want to compare them on, from adaptability to resilience, flexibility to robustness, to applicability. Today the only way to build a complex software system that behaves predictably is via manual labor: programming by hand is an unavoidable part of the task. This is likely to be a dead end – building by hand is not going to get us the kind of architectures capable of supporting intelligence at the level of communicative humans. Unfortunately all current methodologies derive from a constructionist approach, putting manual coding and custom algorithm design and implementation at their center. Taking a fundamentally different approach, and based on new constructivist development principles, the AERA system goes well beyond many of the limitations of present AI systems. AERA can automatically acquire complex skills through observation and imitation. We are testing the system on a challenging task: Learning a subset of socio-communicative skills by observing humans engaged in a simulated TV interview.

Bio:

Dr. Thórisson has been researching artificial intelligence for two decades, both in academia and industry. His research centers on realtime interactive intelligences, complex cognitive architectures and cognitive robotics. At MIT he pioneered new ideas in the area of communicative, multimodal intelligent agents. Recent projects include developing a cognitive architecture for the humanoid robot ASIMO by Honda Motor Corporation. He is the co-founder of CADIA, Iceland’s first AI lab, and is the Founding Director of the Icelandic Institute for Intelligent Machines in Reykjavik. He has taught AI courses at Columbia University, KTH and Reykjavik University, and consulted for NASA and British Telecom, among others. Kris has authored numerous scientific papers and sits on the editorial board of the Journal of Artificial General Intelligenceand the LNCS Transactions on Computational Collective Intelligence.

-- TimoBaumann -- 02 Oct 2012