Researchers at the University of Illinois at Chicago wanted to know how smart a top artificial intelligence system actually is. So they gave it an IQ test.
This light-weight exploratory study actually describes the administration of a normed, verbal IQ test to an AI by a trained psychologist!
I’m pretty pleased with the methodology/analysis by the researchers. Does anyone know of other examples of AIs being administered IQ tests?
An interesting detail from the study that didn’t seem to make it into news stories: When researchers allowed the system to offer its top 3 answers instead of just it’s top answer to questions, the system already performed at 5 year old levels instead of 4 year old levels.
So slight improvement in its confidence weighting algorithm or basic things like adding synonym disambiguation by even a single grad student researcher contributing to this MIT open source project could probably propel this to 1st grader verbal performance levels in a year or two.
See on www.techhive.com
