Artificial intelligence program ConceptNet matches 4-year-old child in IQ test

Artificial intelligence researchers have developed a system capable of matching the IQ of an average four-year-old child in psychometric tests. ConceptNet 4, the open-source AI project used in the tests, has been under development at the Massachusetts Institute of Technology (MIT) since the 1990's.

The study, Measuring an Artificial Intelligence System's Performance on a Verbal IQ Test for Young Children, reveals the results of verbal test questions, such as "Why do we shake hands?" and "Where can you find a penguin". While ConceptNet 4 was able to score test results that are average for a four-year-old, its lack of word reasoning and comprehension skills meant that the wrong answers given were entirely unlike the answers given by children of that age.

For example, to the question "why do we shake hands?" ConceptNet 4 produced the answer "epileptic fit". Delivering questions to a modified form of the AI system that forced it to consider only single concepts, it was able to come up with more suitable answers, such as "flirt" and "meet a friend".

The results are significant in that they show artificial intelligence can match human intelligence, albeit only up to a certain level. Previously, advances in AI have focussed on creating systems that excel at specific tasks, such as chess or pattern recognition tasks.

"In general, recent successes in AI have been mostly learning driven, resting upon statistics, large quantities of data, and machine learning," the study's authors claimed. "The era of knowledge driven AI, resting upon logic, reasoning and knowledge, appears to have passed.

"The areas of common sense and natural language within AI would benefit by having some benchmarks to drive research over the intermediate term... Our use of IQ tests may be [one way of achieving this]."

The ConceptNet 4 system used was first released in March 2012 and since then a new generation, ConceptNet 5, has been developed. The authors of the paper concluded: "In general, more powerful natural language processing tools would likely improve system performance."

Read the full study: Measuring an Artificial Intelligence System's Performance on a Verbal IQ Test for Young Children