Transcript slides
QuickTime™ and a decompressor are needed to see this picture. What is AI All About? QuickTime™ and a decompressor are needed to see this picture. What is AI All About? QuickTime™ and a decompressor are needed to see this picture. What is AI All About? Artificial General Intelligence Narrow AI • Each program carries out specific tasks that humans consider intelligent • Limited capability for adaptation to novel contexts AGI • A single program capable of intelligent behaviors with a variety roughly equal to, or greater than, the variety a typical human demonstrates • Ability to generalize its knowledge so as to adapt to novel contexts The Time is Ripe for AGI QuickTime™ and a decompressor are needed to see this picture. QuickTime™ and a decompressor are needed to see this picture. QuickTime™ and a decompressor are needed to see this picture. Multiple Approaches to AGI QuickTime™ and a decompressor are needed to see this picture. QuickTime™ and a decompressor are needed to see this picture. What is AGI? • g-factor from human psychology? • Ability to achieve complex goals in complex environments? • Legg and Hutter’s formal definition? • Ability to adapt to the environment using limited resources? • Ability to achieve goals in environments, assessed by weighting (goal, environment) pairs according to some distribution that is neither too broad nor too narrow? Embodied, Social AGI • Ability to achieve goals in environments, assessed by weighting each (goal, environment) pair according to the ease with which an embodied social agent could communicate it to its peers, using a combination of methods such as – – – – Language Dramatic depiction Imitation Indication Will a Deep, Rigorous, General AGI Theory Ever Exist? • Can there be a general theory of AGI? • A general theory of computationally feasible AGI? • A general theory of computationally feasible, embodied, social AGI? Possible General Principles of AGI • Occam’s Razor • Cognitive Synergy • ?? QuickTime™ and a decompressor are needed to see this picture. What is the “end goal” of AGI R&D? • • • • Turing test? Artificial, superhuman scientist/engineer? Putting us all out of work? Maybe there isn’t one! Challenges in Moving the Field Ahead • Temptations of Narrow AI research • Absence of a general theory of feasible, relevant general intelligence • Difficulty of measuring incremental progress • Lack of easily usable, common environments/tasks for comparing different approaches • Lack of a collectively agreed-upon “AGI Roadmap” • Paucity of appropriate publication venues • Difficulty of comparing detailed approaches, sharing insights • Public fear of AGI, spurred in part by dystopian fiction • Lack of money! AGI-09 • Keynote: Juergen Schmidhuber • Presentations + themed discussions: – – – – – – – • • • • • • Evaluation/metrics Robotics/embodiment Cognitive architectures Learning Reasoning Language Self-awareness Expert elicitation Sat night: demo/poster session Sun: lunch biz meeting Sun: optional conf dinner at Jaleo Monday: Future of AI workshop Kurzweil Best Paper Prize