Top AI system ‘only at preschool level’
Fears that super-intelligent robots could render humans obsolete have been eased by one of the world’s most advanced artificial intelligence (AI) systems scoring at a young child’s level in an IQ test. ConceptNet 4, developed by the Massachusetts Institute of Technology, performed at an “average 4-year-old” level but much worse in comprehension.
“With common sense and reasoning, it’s clear there is a long way to go,” professor Robert Sloan, head of computer science at the University of Illinois at Chicago and lead author of the study, told Metro. “It struggled with concepts, like why people wear clothes, that young children understand.”
Although Sloan added that ConceptNet 4 performed better than expected with specialized and linguistic tasks, the results may dampen confidence in AI capability. A growing number of scientists have predicted that machines could match human intelligence within decades, boosted by successes such as IBM’s Watson computer that won the quiz show “Jeopardy!” in 2011.
“An embodied approach is needed to fully solve the common sense problem,” said leading AI scientist Ben Goertzel, chairman of the OpenCog Foundation. “AI needs to learn by controlling a robot that does common-sense tasks. The ability to draw analogies and understand words will then emerge from its ability to understand the world.”
Goertzel is developing a robot called Adam Z1, aiming to be “the first AI system with real common sense understanding.” The system will be ready in three years, he told Metro, and he expects to achieve full human-level AI by 2025.
This is unrealistic, according to other specialists. “The goals are not clearly defined enough yet,” said researcher Stuart Armstrong of the Oxford University’s Future of Humanity Institute. “I’m 80% confident it could happen in the next 100 years.”
As for those worried about losing their jobs to robots? “Most conceivable things will have no impact on employment,” says Armstrong, but added that any impact would increase inequality, as machines take over positions requiring lower skills and educationQ t