Reflections on Artificial Intelligence for Humanity

Raja Chatila, Virginia Dignum, Michael Fisher, Fosca Giannotti, Katharina Morik, Stuart Russell, Karen Yeung. 2021.ĪI Development for the Public Interest: From Abstraction Traps to Sociotechnical Risks. McKane Andrus, Sarah Dean, Thomas Krendl Gilbert, Nathan Lambert, Tom Zick. Overviews of societal-scale risks from AI Human Compatible: Artificial Intelligence and The Problem of Control.

Human-Like Machine IntelligenceĪrtificial Intelligence: A Modern Approach (Textbook, 4th Edition). Human-Compatible Artificial Intelligence. Models of bounded or imperfect rationality Multi-agent perspectives and applications Value alignment and inverse reinforcement learning Foundations of rational agency & causality Many of our publications are collaborations with other AI research groups we view collaborations as key to integrating our perspectives into mainstream AI research. We also advise governments and international organizations on policies relevant to ensuring AI technologies will benefit society, and offer insight on a variety of individual-scale and societal-scale risks from AI, such as pertaining to autonomous weapons, the future of employment, and public health and safety.īelow is a list of CHAI's publications since we began operating in 2016. In addition to purely academic work, CHAI strives to produce intellectual outputs for general audiences as well. Transparency and interpretability methods. Other areas of interest to our mission includeĪdversarial training and testing for ML systems, Models of bounded or imperfect rationality. Multi-agent perspectives and applications, and Value alignment and inverse reinforcement learning, The foundations of rational agency and causality, Our areas of greatest focus so far have been However, our interests extend to a variety of other problems in the development of provably beneficial AI systems. Therefore, much of CHAI's research efforts to date have focussed on developing and communicating a new model of AI development, in which AI systems should be uncertain of their objectives, and should be deferent to humans in light of that uncertainty. This way of formulating objectives stands in contrast to the standard model for AI, in which the AI system's objective is assumed to be known completely and correctly. This means we need to somehow represent uncertainty in the objectives of AI systems. In short, any initial formal specification of human values is bound to be wrong in important ways. Currently, it is not possible to specify a formula for human values in any form that we know would provably benefit humanity, if that formula were instated as the objective of a powerful AI system. CHAI aims to reorient the foundations of AI research toward the development of provably beneficial systems.

Human compatible stuart russell