P1070538 Doug Lenat with CFI'ers
Doug Lenat (right), Steve Bratteng (center) and Scott B chat after Doug Lenat's lecture "CYC-ology - Using AI to Organize Knowledge" at the CFI Austin. In this talk Doug Lenat addressed why it is so hard to create general purpose artificial intelligence, and how Cycorp (his company) approaches it.
Human thinking differs from computer logic in a variety of so-called translogical behaviors: incorrect but predictable decisions people make. Some of those phenomena Doug Lenat listed are quite interesting, such as conjunction fallacy, skewed perception of risk and reward, reflection framing effect (that illustrates, among other things, why organ donor rates vary greatly between neighboring countries), or an especially poignant one, failure to understand regression to the mean, which, in Lenat's opinion, explains much of human cruelty.
And yet, computers are no match for humans in understanding semantics of the language, not just syntax. Even the best today's search engines, such as Google, are syntactic. Lenat believes it's possible to achieve synergy of human and machine strengths and weaknesses. He told us about the artificial intelligence engine CYCorp is building, and showed examples of semantic queries it's able to answer.
The whole article can be found in my blog.