Artificial intelligence machine takes a standard IQ test for young children (that’s not it in the picture).
How clever is the cleverest computer currently available in human terms? Say you give it an intelligence test designed for humans and see how it does?
Researchers at the University of Chicago at Illinois have done just that to the latest artificial intelligence machine.
They administered the standard psychological tests to the AI, rather unromantically named ‘ConceptNet 4’. The machine did well on the tests of vocabulary and of similarities in meaning. It was right up there with a pretty smart 4-year-old.
That’s as you’d expect since this AI is built around a network of concepts. By using short phrases and words it has been taught the connections between things and their functions and meanings.
For example it knows that one of the uses of a thing called a ‘saxophone’ is for something else called ‘jazz’. And it knows that people want to do something called ‘learning’ because they are ‘motivated by’ something called ‘wanting to know’. And so on…
Where the AI falls down—and is way behind a normal 4-year-old—is in basic common sense. That’s why the computer did much worse when it tried to answer the ‘why’ questions.
As children we learned so much about the world by exploring:
The computer, though, has to be told everything; it can’t work much out for itself:
The truth is that while the latest generation of AIs are making great leaps forward, if a real four-year-old had these scores, we’d think there was something seriously wrong with it.
So, little chance the machines will become our overlords any time soon.
About the author
Psychologist, Jeremy Dean, PhD is the founder and author of PsyBlog. He holds a doctorate in psychology from University College London and two other advanced degrees in psychology.
He has been writing about scientific research on PsyBlog since 2004. He is also the author of the book “Making Habits, Breaking Habits” (Da Capo, 2003) and several ebooks.