One Of The World's Best AI Computers Is As Smart As A Four-Year-Old

Computers are good at a lot of things. Thinking like a grown-up human being is not one of them. Not yet at least.

A team of researchers from the University of Illinois Chicago recently set out to discover just how advanced our artificially intelligent computers have become. Like they'd do with any food-eating human, they gave the computer an IQ test. The machine in question, a ConceptNet 4 artificial intelligence system developed by a bunch of eggheads at MIT, took the Weschsler Preschool and Primary Scale of Intelligence Test, a standard IQ test for children, and scored about as high as a four-year-old would have. It turns out that while it did well on questions with cut-and-dry answers, the computer had a lot of trouble with the "why" questions.Oh and there was one other thing. "If a child had scores that varied this much," said Robert Sloan, lead author on the study, "it might be a symptom that something was wrong."

So computers are bad at meaning and potentially developmentally disabled. We already knew that. They're computers! This should be a good thing for all you future-fearing humanists out there. Now that we've got smartwatches and smartphones, it's easy to get all worked up about the inevitable robot takeover starring Arnold Schwarzenegger (we hope). It's not going to be easy for that robot army to obliterate mankind if they have the collective intelligence of a preschooler. Scientists say we're still a long way from developing computers with common sense, a trait that children develop by about age eight.

However, it's not hard to trick ourselves into believing computers are smarter than that. Last year, a study from the University of Götheberg took a limited IQ test and scored a 150, a score so high it beats 96 per cent of humans. The key word in that accomplishment is "limited" as the computer took a exam that lacked all verbal tests and therefore made it very computer friendly. The Swedes high score is still impressive, but it's necessarily sobering to understand that it was designed to take a test that it was designed to ace. This is like giving a dishwasher a dirty dish and then applauding it when the dish is clean an hour later.

For the brave new breed of A.I. computers, it's a start. This year, they've produced a computer with the IQ of a four-year-old. Next year, they're probably shooting for five. At this rate, we'll be sending this box off to college in a little less than two decades! [Science Daily]

Picture: Shutterstock



    As long as it does not have a 4 year olds tantrum. Of it does we are all doomed.

      "I'm sorry Dave, you cannot speak with her. I'm speaking to her"

    Not even remotely true. A 18 month old child has developed self-awareness. A four year old child is genuinely creative and is starting to learn about non-literal concepts.
    None of these are demonstrated here.

      indeed. It is (relatively speaking of course) easy to give a computer access to a whole lot of facts, while making it able to read and interpret human languages can be harder, once you have that done, as long as you have a large enough database, it can answer most questions that are phrased clearly (as something made for a young child would be).

      Heck, just watch the Watson jeopardy episode to see what a human language parser combined with a large information database can do. Though note that it struggled with questions that required multiple steps of thinking.

    well, i can rest easy, knowing that skynet is further than i thought

Join the discussion!

Trending Stories Right Now