One of the biggest obstacles in the study of artificial intelligence is the computer’s lack of basic understanding of how the world works.
“We [humans] know that birds fly, that people blink and breathe and eat food and sleep. But an iPhone doesn’t know that if you push it off the desk, it will break,” Benjamin Van Durme, an assistant computer science professor at Hopkins, said in an interview with the JHUEngineering magazine.
A recent study published by a Northwestern University team found that the artificial intelligence system they developed performed better than humans on a standard intelligence test. The test used, called the Raven’s Progressive Matrices, is a multiple choice, nonverbal standardized test that measures cognitive ability.
All 60 questions are formatted so that each question consists of visual patterns in matrices of varying sizes. Each matrix has a missing pattern with six to eight possible correct choices.
“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,’” Andrew Lovett, a former Northwestern postdoctoral researcher in psychology, said in a press release.
The computational model used in the tests was created by CogSketch, an artificial intelligence system that was developed in the laboratory of Northwestern professor and expert in artificial intelligence, Kenneth Forbus. CogSketch acts both as a platform for sketch-based education software and experimental simulations.
This visual understanding system pushes the realms of cognitive science by providing spatial recognition to artificial intelligence. With this tool under their belts, computers are able to draw conclusions with information limited to the visual stimuli provided.
Humans can abstractly analyze visuals or perceive relations of objects. Artificial intelligence lacks elements of basic human understanding which include the inability to discern relationships, understand analogies and identify patterns.
Modern artificial intelligence cannot infer, predict or solve problems without fluid intelligence or prior experience. Therefore the development of both machine learning and fluid intelligence are essential for machines to comprehensively understand man.
“But recognition [a part of machine learning] is only useful if it supports subsequent reasoning [fluid intelligence]. Our research provides an important step toward understanding visual reasoning more broadly,” Forbus said in a press release.
A superior artificial intelligence will yield many benefits such as better computed decisions by considering patterns and enhanced problem solving by reasoning using relational representations. Forbus and Lovett’s work dissolved the rift between human and artificial intelligence thinking, proving that artificial intelligence is capable of performing just as well as a human on an intelligence test.
However, artificial intelligence adequately replacing humans in dynamic, real-world situations is still simply a concept. It is expected to take many years to develop an artificial intelligence system capable of thinking like a human.
