Contemporary Approaches to Engineering Artificial General Intelligence

In recent years an increasing number of AI researchers have begun to turn back toward the original goals of the AI field: creating software programs with general intelligence at the human level and ultimately beyond. In this talk I will review the theoretical distinctions between Artificial General Intelligence (AGI) and the task-specific "narrow AI" R&D that has dominated the AI field recently. I will then review a number of current projects aimed specifically at AGI, including classic projects such as SOAR and Cyc, and more recent approaches such as Nick Cassimatis's PolyScheme, Stan Franklin's LIDA, and my own open-source OpenCog system. The questions of teaching and evaluating AGI systems will also be discussed, with a focus on the ability current technologies give us to embody AI systems in online virtual worlds. Examples will be drawn from my current work creating AI-powered virtual pets in the Multiverse virtual world, and creating language processing and reasoning based AI system for government and commercial customers. The audience will be asked to consider the radical possibility that, perhaps, with sufficient focus of attention and resources, human-level AGI could be achieved faster than most researchers think.