A huge computer project, which simulates the human brain, is so accurate that it makes the same mistakes.
Boffins working on the SPAUN (Semantic Pointer Architecture Unified Network) project were jolly pleased when the computer simulation COULD play games, make predictions and pass simple IQ tests. But some of the tasks it struggles with are the same sort of things that human brains find trickly.
SPAUN, which has 2.5 million virtual neurons, is the brainchild of Professor Chris Eliasmith, who published his paper in the journal Science in November. He said that the only input to the model is an eye, which gets images like you’d take with a camera and the only output is a simulated arm, which we physically model to have mass, length and inertia. The brain uses the arm to draw its responses.
SPAUN can use this arm to copy numbers and letters that it sees with its eye but it doesn’t just copy the shapes, it understands their value and performs almost as well as a human in handwritten character recognition. But the brain can remember short lists of numbers but, like a human, has trouble when the list gets longer it starts to forget some numbers in the middle as the simulated neuron ‘spikes’ get harder to decode.
Unfortunately they have not worked out how to give the simulated brain a good cup of coffee.