AI Can Now Learn As Fast As Humans
This post is also available in: 
עברית (Hebrew)
According to a recent article published in the journal Science, researchers have developed a computational model that allows Artificial Intelligence (AI) to learn in a way similar to that of humans. It provides AI with the ability to recognise handwritten letters as fast as humans can.
AI has been able to recognise patterns in visual and audio information for quite some time now, and make deductions accordingly. However, it could only do these things after being presented with many thousands of examples. Humans, on the other hand, can make such inferences and start learning a thing after seeing only a handful of cases.
Now, a group of inter-institution researchers from New York University, MIT, and the University of Toronto have managed to create an AI algorithm that approaches learning and problem solving in a way similar to humans.
This breakthrough work propels AI research closer to giving machines learning capabilities on par with those of humans.
“Compared to our best machine learning algorithms, people can learn a new concept from less data and generalize in richer and more powerful ways,” says Brenden Lake, a Moore-Sloan Data Science Fellow at New York University. “We were able to capture these human learning abilities for a large class of simple visual concepts — handwritten characters from the world’s alphabets.”
AI possessing human-like deductive capabilities opens up a slew of new, exciting possibilities. Online education courses would benefit from automated processes, with AI personalising content according to students’ needs, says Susan Eustis, co-founder and senior researcher at WinterGreen Research.
“Medical markets stand to gain diagnostic flexibility, greatly speeding the time to achieve an accurate diagnosis and refer patients to an appropriate treatment regime,” she says. “Supply chain modernization will also benefit from the ability to manage thousands of varying inputs and quickly notify all the appropriate managerial endpoints as to the state of shipments and inventory in a timely manner.”
The research was supported by the (US) National Science Foundation, Army Research Office, the Office of Naval Research, and the Moore-Sloan Data Science Environment.
