Can Computers Learn the Way Humans Do?

Human and machine learning may soon become indistinguishable.

Artificial intelligence may get a whole lot smarter, thanks to a new algorithm developed by a team of researchers from New York University (NYU). Published in the latest issue of the journal Science, Eureka Alert reported that the algorithm captures human learning abilities which enables computers to see and draw simple visual concepts. The same statement claimed that the visual concepts created are almost indistinguishable from humans -- a significant breakthrough as it shortens the duration of "learning time" for computers and therefore allotting more time to learn more creative tasks and applications.

"Our results show that by reverse engineering how people think about a problem, we can develop better algorithms," explained lead author Brenden Lake, a Moore-Sloan Data Science Fellow at New York University. "Moreover, this work points to promising methods to narrow the gap for other machine learning tasks," Lake continued.

The team of NYU scientists have successfully developed a "Bayesian Program Learning (BPL)" algorithm that transforms "concepts into simple computer programs" which allows computers a huge class of data from a single example, according to a report by Mashable.

The model developed is said to glean from the same knowledge gained from previous concepts. Mashable exemplified: "For example, if the computer knows the Latin alphabet, that can help it learn the similar Greek alphabet"

Lake revealed that the discovery happened from their observations of people drawing a character. "If you ask a handful of people to draw a novel character, there is remarkable consistency in the way people draw.... They do not see characters as just static visual objects. Instead people see richer structure... that describes how to efficiently produce new examples of the concept," he explained.

Lake furthered that his team's objective was to develop an algorithm with a similar capacity and skill that is comparable with a person's.

The scientists then proceeded to testing by asking both humans and computers to reproduce or recreate as series of handwritten characters after being shown a single example of each. Both outputs were then compared through "visual Turing tests." Human judges determined whether the drawings were made by a computer or a person.

According to Eureka Alert, less than 25 percent of the judges guessed "performed significantly better than chance in assessing whether a machine or a human produced a given set of symbols." Given this result, the future of Robotics, particularly Artificial Intelligence, could either be an scary or exciting thought.

Allen Institute for Artificial Intelligence CEO Oren Etzioni said though, as per Mashable, that "they didn't beat the Turing test any more than a calculator does by out-multiplying a human," and the work is best classified as a "scientific contribution."

"While the authors pose a fascinating research question, many researchers have used related methods to achieve strong results. Still, the paper is an invaluable reminder that we need methods that can generalize from small numbers of examples both to model human abilities and to move AI forward," furthered Etzioni based on the same report.