A team at the University of Illinois, Chicago, gave one of the world’s most advanced AI systems an IQ test.* They reported its IQ as the level of an average four year-old.
This research has to be taken with a pinch of salt. For a start, many psychologists are sceptical of the value of IQ tests. The fact that you can improve your test results with practice suggests that they don’t test anything more inherent than the ability to pass IQ tests.
Also, the system scored very differently on different parts of the test, to the extent that a child scoring the same would be diagnosed with serious problems. And the researchers noted that the system was lacking in that hard-to-define but vital asset known as commonsense.
That said, the report is intriguing. Broadly speaking, there are two main approaches to creating an artificial mind. One is to build an exact replica, or model, of a human brain and see if it generates thought and consciousness. The other is to take the most advanced AI systems available and have them learn.
If it can be said, however loosely, that today’s most advanced AI systems have the IQ of a four year-old, how long before they have the IQ of an adult human?
* Verbal IQ of a Four-Year Old Achieved by an AI System, presented at the AAAI Conference on Artificial Intelligence (AAAI-13), Bellevue, Washington, 2013. The authors (Stellan Ohlsson, Robert H. Sloan, György Turán, Aaron Urasky) gave ConceptNet 4, AI software developed at MIT, the verbal portions of the Weschsler Preschool and Primary Scale of Intelligence Test, a standard IQ assessment for young children.