Mathematicians have come up with a problem that artificial intelligence is not able to solve in principle

An international team of machine learning and artificial intelligence researchers have come to the disappointing conclusion that they are more limited than we thought. This stems from the mathematical nature of their architecture and working methods. According to Gödel's Incompleteness Theorem and the provisions of his Second Theorem, in reality, far from all mathematical problems are solvable. And since machine learning is precisely mathematical in nature, it has its limit.

One of the most pressing problems in machine learning is "maximizing". It can be illustrated by the following example: there is a certain site that will be visited by an unknown number of users, whose interests are also unknown in advance, but in general the set of parameters is finite. It is necessary to create an algorithm that will ensure that all of them display targeted advertising with an accuracy close to absolute. When simulating such a situation, scientists came to an unambiguous similarity with the conditions of the "Continuum Hypothesis", which for a long time was on the list of unsolved problems in mathematics.

To be more precise, for both the Incompleteness Theorem and the Continuum Hypothesis, there is no answer in the form familiar to a machine. AI, even the most advanced one, when solving such a problem will come to such a step when it cannot give an assessment of "true" or "false". A person would simply wave his hand, introduce some additional condition or ignore the importance of choice, make a decision intuitively. Machine learning algorithms do not allow this kind of liberties, so the AI ​​will not be able to continue working.

The burden of unprovability, alas, is inherent in too many mathematical problems, and therefore the probability that AI will sooner or later face a similar situation approaches 100%. This means that we need today to figure out how to allow him to get around such paradoxes. But at the same time, preserve the accuracy of analyzing the situation and making decisions, for the sake of which we are trying to teach our artificial mind. And this is all the more difficult, the greater the deviation from the rules is allowed in his work.