Computer at MIT matches verbal reasoning ability of a 4-year-old

New Scientist : ConceptNet is a computer system that uses a crowd-sourced database of simple relational statements describing everyday objects, such as “a fawn is a deer.” It was developed by Catherine Havasi of the MIT Media Lab and her colleagues. Robert Sloan and Stellan Ohlsson of the University of Illinois at Chicago, who were unaffiliated with ConceptNet’s development, evaluated the system using the Wechsler Preschool and Primary Scale of Intelligence test. The test asks a range of verbal questions including basic vocabulary definitions—"What is a house?"—and simple riddles—"You can see through it. It is a square and can be opened. What is it?” ConceptNet then searches its database for relevant statements and uses a variety of algorithms to find the best answer. The results were approximately equivalent to the verbal ability and knowledge of an average 4-year-old human. However, the test was limited to only verbal skill and did not examine spatial or symbolic reasoning ability. Havasi says that the version of ConceptNet that was tested had a database of 1 million statements; she expects that the most recent version, which has 17 million, might score even higher.