Evaluation of Human-level Intelligence

In the last five years, there has been a renewed interest in creating systems that had general intelligence as indicated by the first conference
on Artificial General Intelligence, held last spring. Although there is much
enthusiasm and activity, there are no accepted standards for evaluation in
this field, which significantly inhibits comparison and evaluation of
alternative approaches, and makes it difficult to determine when progress
has been made. Moreover, shared evaluation procedures, testbeds, and
infrastructure are missing. In this talk I will present the results of a
workshop on evaluation of human-level AI systems held at Michigan last month
and open up the presentation to discussion of the issue with the audience.