17:30 GMT +316 December 2019
Listen Live
    Human Brain

    Artificial Intelligence Machines Match IQ Test Results of 4-Year-Olds

    © Flickr / Ars Electronica
    Tech
    Get short URL
    7105
    Subscribe

    Some may be wary and some may celebrate, but no one can stop machines from developing intelligence. Fresh studies show AI has reached intellectual levels of 4-year-old humans in IQ tests. What’s next?

    From the start, robots were outplaying humans in algebra. Recently they’ve begun to lead in chess and other games. They are good with face recognition, too. But what about comparing them to humans in more general terms?

    Psychometrics were designed by psychologists as a means to measure how intellectually developed a person is. Probably the most widely known psychometrics is the IQ test. A team of researchers with the University of Illinois at Chicago, led by Professor Stellan Ohlsson, tried the method on one of the most advanced computer systems.

    The results they came up with are just astonishing. It appears that modern AI systems have reached the level of an average 4-year-old kid.

    The verbal part of the Wechsler Preschool and Primary Scale of Intelligence (WPPSI-III) was tested on the ConceptNet 4 AI system. The results were recently published at arXiv.org.

    “We chose the WPPSI-III because we expected some of its subtests to highlight limitations of current AI systems, as opposed to some PAI [Psychometric Artificial Intelligence] work with other psychometric tests of verbal abilities that highlights the progress that AI systems have made over the decades,” the paper says.

    Researchers wrote a program in Python language to input into the ConceptNet.

    While the findings are impressive, improvements are needed before an AI could match a 5-year-old child.

    “ConceptNet does well on Vocabulary and Similarities, middling on Information, and poorly on Word Reasoning and Comprehension,” the paper highlights.

    This means that the AI system easily answers questions like “What is a cat?,” “Ball and apple are both___,” but struggles with abstract reasoning. “Why do people shake hands?” would be a more difficult question.

    “Future work on the query interface, the knowledge base, and the inference routines will inevitably raise the question of where the boundary is to be drawn between natural language understanding on the one hand, and common sense reasoning on the other,” the paper concludes.

    Related:

    Artificial Intelligence “Existential Threat” or Boon?
    Artificial Intelligence Taking Over? Robot Threatens to Cage Creators
    Artificial Intelligence: Myths, Facts and Future
    US Navy Accelerates Artificial Intelligence Development for Combat
    IBM Teams with Apple on Artificial Intelligence Health Program
    Elon Musk Donates $10 Mln to Keep Artificial Intelligence Human-Friendly
    Tags:
    IQ test, Artificial Intelligence (AI), research, ConceptNet 4, University of Illinois at Chicago
    Community standardsDiscussion
    Comment via FacebookComment via Sputnik