Speakers at the conference Neural Information Processing Systems, experts in the field of artificial intelligence has stated that they no longer understand the principle of decision-making is guided by AI, reports Quartz. According to experts, to accept the actions of the AI without understanding its logic — rather lightly, after all, to accept a machine learning model, they need to know what exactly is guided by the AI when deciding how it should behave in a particular situation.
Often the decisions taken by the AI, are biased, in addition, associative “thinking” is also often less ideal, resulting in AI carry out a wrong analogy. Such errors can be costly if the AI will be to lead complex projects, such for example, as a trip to Mars — in this case, the wrong thing artificial intelligence can not only destroy expensive equipment and lead to death. Therefore, before you allow the AI to make their own important decisions, you must first learn the principles, which guides, explains Master Raghu, a specialist in AI from Google.
At the conference she presented a report which described the process of tracking the actions of individual parts of neroli. By looking at whole of their parts, it is possible to understand the logic of the AI, and then, if anything, and to proofread. By analyzing millions of transactions, it was able to identify individual iskusstvennye “neurons”, which focused on incorrect views, and then disable them, making the AI more amenable and correct.
This is something similar to how the teacher delivers the material and then asks the student to retell in your own words what he understood from the lecture, explains Kiri Wagstaff AI expert from NASA.
Judging by the results of this study, to correctly understand the actions iskusstvennogo intellekta not so difficult. The main thing — to solve the problem before it is too late.
Exactly how the AI makes decisions? Scientists no longer understand it
Vyacheslav Larionov