Northwestern University’s Professors Created An AI System Capable Of Scoring Better Than 75 Percent Of The US Population In Raven’s Progressive Matrices Test

One of the most important aspects of general intelligence model created by Charles Spearman. Spearman is one of the most recognized names in psychology because he was first to propose a model of general intelligence based on one common factor (a general ability factor, or simply the g factor) influencing every aspect of intelligence.

Today, Spearman’s theory is considered as a base for all intelligence tests, with IQ scores being estimates of a person’s standing on the g factor, compared to the general population. One of the most famous non-verbal intelligence tests used today is Raven’s Progressive Matrices test (RPM), a visual test designed to measure fluid intelligence, individual’s reasoning ability, a meaning-making component of Spearman’s g factor. In other words, it measures meaning making component of general intelligence.

One of the problems taken from RVP test. The correct answer is 8, since a figure from one column is added to or subtracted from another column to produce the third.

Now, while humans are quite good at noticing different patterns and discovering meaning between them – our brain is wired in a way that drives it to find patterns and meaning even between to phenomena that aren’t connected – the current state of artificial intelligence doesn’t really allow it to rock on this kind of tests.

Today’s AI systems are based on analyzing the data and finding patterns and connections based on data analysis, machine learning, and “deep learning.” But them solving complex visual problems isn’t really an area of expertise for the AI systems of today, or so we thought so.

A team of researchers at Northwestern University managed to develop a new AI system capable of achieving better scores than 75 percent of the US adult population. Ken Forbus, the leading researcher stated that “The model performs in the 75th percentile for American adults, making it better than average,” and that “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.” Their research has been published in the journal Psychological Review.

Forbus developed CogSketch, an artificial intelligence platform with the ability to solve visual problems and give interactive feedback when presented with various sketches. The new computational model that’s better than most of the adult population of the US is built on CogSketch.

The existence of an AI system capable of solving complex visual problems is an important breakthrough in AI research. Solving complex visual problems is one of the trademarks of human intelligence, and making a machine to think like a human (although just in one, quite narrow, form) can drive forward the entire AI research.

“Most artificial intelligence research today concerning vision focuses on recognition, or labeling what is in a scene rather than reasoning about it,” Forbus said. “But recognition is only useful if it supports subsequent reasoning. Our research provides an important step toward understanding visual reasoning more broadly.”

It would be interesting to see will the researchers working on huge AI projects (like Google’s Deep Mind or Facebook’s AI project) implement the model in their research.

Leave a Reply

Your email address will not be published. Required fields are marked *