Washington, Dec 11 (EFE).- A group of scientists has developed a software program that imitates the way humans learn new concepts, an “important step” in the field of artificial intelligence, a study published in the Science journal said.

What the scientists are trying to do, according to Joshua Tenenbaum, one of the researchers and a professor at the MIT Department of Brain and Cognitive Sciences, is to reduce the difference in learning capability between humans and machines, and to discover why “people learning new concepts can often generalize from just a single example.”

Computers, on the other hand, “require tens or hundreds of examples to perform similarly,” according to the study.

The chief virtue of humans is the speed and dimension with which they learn new concepts and apply them to new situations, while for computers to generalize from individual examples has been next to impossible, Brendan Lake of New York University and lead author of the study said in a conference call to present the groundbreaking study.

The researchers focused on having the new software, known as Bayesian Program Learning, or B.P.L., learn handwritten characters in a range of languages, and developed an algorithm that allows the program to make accurate generalizations from very few examples.

Up to now, computers haven’t had one program applicable to every situation, but rather a complex of different learning programs for each circumstance, Tenenbaum said during the same conference call.

Comparing the capability of this new software, endowed with a technique known as “deep learning,” to accomplish learning tasks, including concept generalization from examples seen only a few times, with that of other computers and humans, it far surpassed any previously developed machine intelligence and equaled humans.

The learning capabilities of this new cognitive model “in many cases are indistinguishable from human behavior,” the study says. EFE