Research Paradigms in Artificial Intelligence

Artificial Intelligence (AI) has been much in the news of late, with reports of impressive technical achievements alongside commentary on social benefits, risks, and policy. Receiving less attention has been the development of AI as an intellectual undertaking, with roots stretching back to the 1940s. Many observers today conflate AI with Machine Learning, but historically AI has always pursued several competing “threads” each with its own intuitions, research methods, and ideas about how intelligence should be modeled and analyzed. Each survives to the present day, and for good reason. This talk reviews several of these paradigms--heuristic search, symbolic reasoning, machine learning, and neural networks--and argues that, despite its recent progress, AI still needs better theoretical foundations: less technique-driven, less derivative of anthropomorphic concepts, and more careful in defining its domain of inquiry.

 

Speaker: Stan Rosenschein, Stanford

Room 126