Warning for humans : the existing Artificial Intelligence is a lot smarter than the previous ones as they beat the Turing Test
Computer researchers have alleged that artificial-intelligence have advanced a lot and have beaten human capabilities for a thin series of vision-related tasks.
The developments are notable as the so-called machine-vision systems are turning out to be commonplace in various aspects of life. These involves systems for car-safety that senses pedestrians and bicyclists, as well as in video game controls, Internet search and factory robots.
Experts at the Massachusetts Institute of Technology, New York University and the University of Toronto recounted a novel type of “one shot” machine learning on Thursday in the journal Science, where a computer vision program outclassed a team of humans in classifying handwritten characters based on a single example.
The program was able to learn the characters in a short period of time amid a range of languages and generalizing from what it has acknowledged. The authors advocated the aptitude as identical to the way humans learn and understand concepts.
The new tactic, best known as Bayesian Program Learning, or B.P.L., is different from the existing machine learning technologies recognized as deep neural networks.
Neural networks can be drilled to distinguish human speech, spot objects in images or recognize various kinds of behavior by being exposed to massive sets of examples.
Even though such networks are sculpted after the behavior of biological neurons, they do not yet understand the way humans do — obtaining new concepts swiftly. As compared, the new software program elaborated in the Science article has the ability to learn and recognize handwritten characters after “seeing” only a few or even a single example.
Researchers have stated that the model uses information from previous concepts for learning. For instance, if the AI knows the Latin alphabet, it is able to help itself learn the similar Greek alphabet.
The lead author of the study, Brenden Lake at the Moore-Sloan Data Science Fellow at New York University said that the breakthrough was noticed when researchers saw that, “If you ask a handful of people to sketch a new character, there are extraordinary consistency in the way people draw…. They do not perceive characters as just static visual objects. In its place, humans view a richer structure… that explains how to efficiently harvest new examples of the concept.
Researchers in the study aimed to create an algorithm with the same ability and then compare it with people.
During a demonstration on their work, the researchers stated that they’ve not only erected a machine-learning program, “but what the program absorbs — its concepts — are also programs. They believe that it is true for humans as well. Our concepts are also programs, or sections of programs, stated Joshua Tenenbaum, of the Department of Brain and Cognitive Sciences, and Center for Brains, Minds, and Machines, at MIT.
Fascinatingly, when the AI was asked to produce fresh examples based on the actual concept, and those images were matched to examples created by humans, many were unable to distinguish if a person or a computer had