The Singularity Institute Blog

SIAI Director of Research Dr. Ben Goertzel reports on his participation in a recent workshop.

“I just returned from the University of Michigan at Ann Arbor, where I attended a weekend workshop on Evaluation and Metrics for Human-Level AI, sponsored by the Office of Naval Research and co-chaired by John Laird from the University of Michigan (who is a leader of the SOAR project, a long-standing project aimed at human-level AI, and pursued both in academia and, via Soartech, in industry) and Pat Langley from Arizona State University (who has a long and diverse history of achievement in the AI field, and has recently introduced the Icarus integrative AI system).

“Rather than a typical academic workshop focused on the presenting of papers, this was a ‘working workshop’ devoted to discussions aimed at collectively generating and filtering ideas regarding appropriate environments and tasks for software systems aimed at human-level AI. There will likely be a followup workshop early next year, focused more tightly on the description of specific environments and tasks; this one was more of a careful exploration of the relevant conceptual landscape, and a review of various requirements to be fulfilled and pitfalls to be avoided.

“Joscha Bach (creater of the MicroPsi AGI system) and I were the most radical optimists present, both of us believing it likely that with sufficient dedication of resources, human-level AI could plausibly be created within a decade. Most of the other participants believed it would take longer — but all considered human-level AI a valid and important area on which to focus current research attention.

“Among many other topics, there was some discussion of my proposal of an ‘AGI Preschool,’ which is summarized in ‘Intelligence Assessment for Early-Stage Software Systems Aimed at Human-Level, Roughly Human-Like AGI.’ Time permitting, some of the workshop participants may write a report of the main conclusions reached and submit it for publication.”