The term "artificial intelligence" often refers to neural networks built on deep machine learning technologies. Moreover, the technology of training neural networks is well developed and is bearing fruit. However, not all scientists share the opinion that artificial intelligence should develop along this path. Someone even believes that such systems “should not be trusted” and their development will not lead to anything good.
Artificial intelligence in the modern sense - this is not at all what many think.
Why machine learning is bad for human development
In a large-scale work published on the pages of Technologyreview, a professor at New York University, an expert in the field of cognitive science (the science of cognition) Gary Marcus spoke about the risks of widespread use of neural networks based on deep machine learning.
First, the scientist believes that the technology has clear limitations. In particular, there has been talk for a long time about the need to create a so-called "real AI", which is suitable for solving a wide range of problems, and not just one specific one, as is happening now. The existing AI systems have already reached the peak of their development and they have practically "nowhere to grow". In addition, you cannot just take and, say, first teach one AI to drive a car, and force another to repair it and then combine the systems, creating a universal assistant. Artificial intelligences simply will not be able to interact, as they "learned in different ways."
How to make AI smarter
Promotional video:
For algorithms to become more efficient, they need to be "trained differently." It is necessary to make sure that they begin to see the relationship of objects and the consequences of interacting with them. In this case, we will serve as the best example.
Professor Gary Marcus.
Moreover, what Marcus offers is not new at all. The example described above is how scientists envisioned "classical AI." But in order for such an AI to work effectively, we need to program all possible outcomes in advance. And this is almost unrealistic. But there is a way out.
The solution can be a kind of symbiosis of "classical AI", which sees relationships and obtains solutions in an understandable way, and deep learning, which is able to find a solution through "trial and error". It can be some kind of basic system of rules and regulations concerning the surrounding world. On their basis, AI systems will already be able to develop themselves in a certain area. Real artificial intelligence must understand how everything works around in order to understand cause-effect relationships and easily switch from one task to another. Modern systems built using deep learning technology are simply not capable of this.
Vladimir Kuznetsov