A university study has shown that an AI outperformed 75% of Americans on the test.
Northwestern University researchers have created a computational model for an artificial intelligence program which outperforms most humans on a standard visual IQ test. The test features a series of patterns and asks the test-taker to select which pattern is the next natural progression of the series. The researchers say that the result is an important stepping stone to making AIs understand the world like humans do. “The model performs in the 75th percentile for American adults, making it better than average,” said Northwestern Engineering’s Ken Forbus, “The problems that are hard for people are also hard for the model, providing additional evidence that its operation is capturing some important properties of human cognition.”
The new computational model is built on top of the CogSketch AI platform, which is designed to interpret sketches. CogSketch was developed at Forbus’ laboratory at Northwestern University. Sketching is a natural way for humans to communicate ideas, and are used quite naturally in many scientific and engineering fields. CogSketch is designed for spatial understanding and reasoning, making it suitable for interpreting research based on sketches.
Forbus designed the AI platform on the notion that analogical reasoning is at the heart of visual problem solving. If we want computers to solve complex problems, it is of benefit for them to be able to understand analogies and draw conclusions linking two different subjects. The sketched images go through something called ‘structure-mapping’, comparing and aligning structures found in the sketch and in reference images, to identify common elements. The structure-mapping theory was developed by psychology professor Dedre Gentner, who also works at Northwestern.
An example of the type of question the AI encountered on the test
The IQ test didn’t involve any sketches, but the structure-mapping technique helped it excel at the test, a so called Raven’s Progressive Matrices Test. “The Raven’s test is the best existing predictor of what psychologists call ‘fluid intelligence, or the general ability to think abstractly, reason, identify patterns, solve problems, and discern relationships’,” said Forbus’ former colleague Andrew Lovett, now a researcher at the US Naval Research Laboratory, “Our results suggest that the ability to flexibly use relational representations, comparing and reinterpreting them, is important for fluid intelligence.”
source: Tom’s Hardware
