Artificial intelligence not as smart as a 4-year-old

Despite a deluge of media coverage on the heels of this year's Association for the Advancement of Artificial Intelligence conference, computers have not attained the same level of intelligence as 4-year-old children, writes Miles Brundage, a doctoral student in the Consortium for Science, Policy and Outcomes at Arizona State University.

Researchers from the University of Illinois at Chicago recently administered a standard verbal IQ test to ConceptNet4, a semantic network program that helps computers interpret various concepts and language. While the program’s overall score matched up with that of a 4-year old, the results were extremely uneven: it scored rather high on vocabulary but performed poorly on tasks related to language comprehension (the “why questions,” as the researchers put it). In other words, according to Brundage, the semantic network “did well on precisely the parts of the test one would expect computers to excel at.”

Brundage argues that the very idea of comparing human and machine intelligence in terms of age level is counterproductive. Instead of designing machines to replace or simulate human intelligence, we should develop machines that collaborate with humans and perform tasks “that humans aren’t naturally good at, don’t want to do, or are just too dangerous.” After all, “we already have plenty of humans who can do all sorts of cool things, many of whom are unemployed.”

Future Tense is a collaboration among ASU, the New America Foundation and Slate magazine that explores how emerging technologies affect policy and society.