The term “artificial intelligence” often refers to neural network based on the technology of deep machine learning. Moreover, the technology of neural network learning is well established and is bearing fruit. However, not all scientists share the view that artificial intelligence should develop on this way. Someone even believes that such systems “do not trust”, and no good development will not result.
Artificial intelligence in the modern sense is not what many think
Why machine learning is bad for human development
In a massive work, published in the publication Тechnologyreview, a Professor at new York University, a specialist in the field of cognitive science (science of knowledge) Gary Marcus told what are the consequences of the widespread use of neural networks on the basis of deep machine learning.
First, the scientist believes that the technology has clear limitations. In particular, have long talks about what it takes to create the so-called “real AI”, which is suitable for a wide range of tasks, not a specific one, as is happening now. Existing AI systems are already came to the peak of its development and it is almost “nowhere to grow”. Besides, you can’t just, say first one to teach AI to drive, and the other to get to fix it, and then merge the system, creating a universal assistant. Artificial intelligence just can’t communicate as “learned differently”.
You can teach the AI to play Atari better than a man, but to make a good robomobile — hardly. Although this task is also quite specialized. Deep learning performs well in the analysis of big data, but the algorithms do not see the causal relationships and do not take any change of conditions. Slide elements in a computer game two to three pixels, and trained the AI will become ineffective. Make the field for the game of go is not square, but rectangular, and artificial intelligence will lose even the novice player.
How to make the AI smarter
In order for algorithms to become more effective, they need to “teach differently”. It is necessary to make so that they began to see the relationship between objects and the consequences of interaction with them. In this case, the best example will be you and me.
Type of students-interns, and they in a few days you will begin to work on any issue — from law to medicine. Not because all of them are smart. And the fact that people have a General idea about the world, rather than private.
Professor Gary Marcus
And what Markus is not new. The example described above is how scientists imagined “classical AI”. Only here for this AI to work effectively, we need to pre-program all possible outcomes. And it’s almost unreal. But there is a solution. By the way, what path of development of AI is preferable in your opinion? Tell us about it in our chat in Telegram.
See also: artificial intelligence
A solution could be a sort of symbiosis of “classic AI”, which sees the relationship and gets the solution in meaningful ways, and deep learning, are able to find the solution by “trial and error”. This can be a kind of basic system rules and regulations regarding the surrounding world. On the basis of their AI systems are already and will be able to develop yourself in a certain area. True AI needs to understand how everything works around in order to understand causal relationships and to easily switch from one task to another. Modern systems created using technologies for deep learning, for such are simply not capable of.