According to the "technology review" website reported at the Massachusetts institute of technology, an international research team gain inspiration from the learning method of the human brain, developed a more efficient, with complex learning ability of artificial intelligence software. The latest study published recently in the journal science.

This kind of new artificial intelligence software can only need to look at an example like people identify precisely to write words, and by far the most common machine learning algorithm using A technique called "deep learning", the need to tens of thousands of cases of handwritten characters to distinguish the letters A and Z.

Over the past few years, the computer has become more and more intelligent, they can learn to identify facial, understand the content of the speech, and even safe driving a car, but the ability to learn: there are serious defects even learn how to perform the simple task, they also need a large amount of data.

The research team used a technique called "bayesian learning system" algorithm, this algorithm can use virtual stroke for each text generated unique system. Then, they use a probabilistic programming technology will each system matching with the text, or to an unfamiliar word to generate new system. This system does not imitate children get the process of reading and writing ability, but to imitate adult's learning process, which is on the basis of the already know how to learn to recognize and write the new words.

In order to test the accuracy of the system, the researchers and software at the same time to see the new words after the case writing out the words, and then let another group of people to judge what words were written by people, which is written by robots. They found that less than 25% of the "the referee" can find the different.

The research team, said the technology can be extended to the more practical application. For example, it can let the computer quickly learn how to recognize and use the new vocabulary in spoken English, also can let a computer to quickly identify new object.

A cognitive scientist at the university of New York Gary marcus thinks, this study represents the new direction of artificial intelligence technology, because in many cases, and there is not enough data for the computer to learn. Marcus believed that language will be a "killer application" of this kind of system. "When artificial intelligence can really understand the language, the robot speaks there will be a real inflection." Marcus, stressed that "when they turn to ordinary language is not just a job, but really understand what you mean."