For the first time since the computers "learned" the meaning of various words and sentences with the help of programmers, machine beat people in the intelligence test. The technological advance achieved by programmers thanks to a new approach to language helped computers to outmatch the people in a part of IQ test that applies to verbal reasoning.

The questions in the test were divided into three categories: logic (recognizing patterns in parts of the photo), mathematics (recognizing patterns in numbers) and verbal reasoning.

Computers have previously had difficulties in the final category, which refers to the analogy, classification, synonyms and antonyms.

When a new computer and 200 people answered the same questions in the field of verbal reasoning on Amazon's Mechanical Turk site, the computer performed better than the average man. This happened for the first time since there are tests of this kind.

Scientists from the Microsoft in collaboration with the Chinese University of Science and Technology have constructed a machine with the ability of deep learning which approaches problems in a completely new way.

Earlier computers were using technique of data analysis, where they took a large amount of texts in order to find a connection to other words. This method is efficient for tasks such as translation, but is based on the assumption that a word has one meaning. That is why computers have difficulties with synonyms and antonyms - tasks that people resolve relatively easy by context. To overcome this problem programmers had to develop an entirely new method that would later be used by their computer.

First, the computer recognizes whether the question from the test relates to the analogy, classification, synonym or antonym. In the second phase, the computer considers several meanings, analyzing other related words, mostly using reasoning to determine the context. Finally, using the appropriate meaning, the computer solves the problem.

Scientists say that the results are very encouraging and suggest that "with adequate advancement of technology of deep learning we can get closer to the true human intelligence in computers."