An event has happened that could become a landmark in the history of the development of artificial intelligence. Google's DeepMind system, which purposefully studied the school course of mathematics, algebra and partly geometry, scored only 14 points out of 40 according to the results of the test. And thus failed the test aimed at testing the knowledge of the level of 16-year-old students.
The study was conducted in the UK, in accordance with the norms of the national education program. It is not at all as complicated as it might seem - the DeepMind developers have already figured out what the problem was. It turns out that although AI can easily solve many mathematical problems, it still cannot recognize the formulation of the problem itself - in the form in which students receive it.
Instead of a clear set of data, the AI receives "abstract" from its point of view, text, where letters are mixed with numbers, and verbal descriptions are interspersed with formulas. What looks logical and understandable for a trained student forced DeepMind's algorithms to spend almost all of the system's resources simply on recognizing and extracting information. And it came as a surprise to everyone, since the AI seems to have been trained on this topic?
But no, the AI turned out to be, in fact, unable to solve the problem presented in free form. This means that artificial intelligence developers face a new, large and complex task. And fans of fantastic dystopias can sleep peacefully - the machines will not soon understand what is in this world.