Putting "a strong form of AI" on the knee ...

You can talk about the dangers of AI for a long time, but it is much more interesting to participate in the process. Let's start with the analysis of why, what is being developed today will not lead to a breakthrough in this area.



It's funny that the last couple of years, all rushed to the idea of ​​creating a strong artificial intelligence. Dug up algorithms 50 years ago, for which suddenly began to be enough power. And they tell, encouraging each other, how everything will be cool if the number of neurons in artificial neural networks can be brought to equalize with the number of neurons in the human brain.



At the same time, no one really tells, why is there still a strong AI for humanity? What tasks will be in front of him. what is the purpose of its existence. All usually some general phrases "WELL THE TYPE WILL BE THE COOR OF THE SAME."



It becomes especially fun when some kind of grief start-up developer starts broadcasting “oh, we don’t even understand how the neural networks we create work”. Seriously? So, how do neural networks work, and why you should not expect High AI based on neural networks.

Do you know what is arithmetic mean?

The arithmetic average (in mathematics and statistics) is one of the most common measures of the central tendency, representing the sum of all recorded values ​​divided by their number. It was proposed (along with the geometric mean and harmonic mean) also by the Pythagoreans.

Now we take 1 million photos, for example cats I will simplify, for a better understanding, but not much.

')

Take all these photos of cats and make them an arithmetic average cat in a vacuum.





At the same time, the more photos of cats we feed, the better the recognition will be when comparing a new photo with an average of photos of cats. A couple of dozen photos of cats will not be enough.



Is it possible to build a Strong AI on such an algorithm? How, when increasing the size of neural networks, the calculation of the arithmetic mean will allow to proceed to the independent construction of algorithms? Apparently there is a calculation for some kind of magic, but when we make enough neurons in the network, there is something that will generate itself. Well, they believed earlier that mice were self-generated in a pile of rags.



At the same time, that I personally do not understand, in attempts to create a Strong AI, for some reason they are trying to copy the human brain. Although it would be more logical to copy the way of thinking. We will discuss this next time if the topic is interesting to someone. I have some thoughts.



The topic of the practical application of Strong AI is still open. What nails will we hammer with this microscope? I propose to speak in the comments.