Wherein Musk is wrong, speaking about AI

Musk, undoubtedly, is the embodiment of progress. But, from my point of view, one of the theses that he transmits is erroneous.

I believe that there is no threat from a strong AI. I will try to explain this originated fear and show that there are no prerequisites for it.

In the 60–70s of the last century, such authorities as Marvin Minsky, John McCarthy and Seymour Papert have masterminded the development and popularization of the theme of AI. It is they, who described two possible, but fundamentally different approaches in the creating of AI: top-down approach (or Scruffy School) and bottom-up approach (Neat). The top-down approach presupposes the creation of a strong AI by copying the human psyche, primarily its cognitive processes. The bottom-up approach is the gradual evolution of the mathematical apparatus from the reproduction of the neuron’s work with the subsequent complication of the system to a level where it can understand and solve complex problems. During the last few years, the attention has been focused on this way of creating AI. To a large extent, this is caused by the progress in the development of neural networks and the opportunity to obtain practical results immediately.

As a result, in the minds of most people, such concepts as “AI” and “neural network” have become identical. But what is a neural network from a human’s point of view? As for me, it is an image of a black box in which from one side the information arrives, and from the other side it gives out some conclusions and solutions. If a device with a similar architecture does not generate stress in the process of solving some mathematical and engineering problems, I will not dare to commit my health and even my life to this black box’s trust. If we talk about a strong AI, built according to such architecture, then, for my part, it looks more like an alien. I cannot even guess its motives, its train of thoughts and the principles of decision-making. Owing to the uncertainty and the inability to predict its actions appears the fear. This fear is quite relevant.

My colleagues and I are working on the creation of AI, using a top-down approach. I have already written about our work — The development of strong AI, by copying the structures and processes of the human psyche and 5 advantages of the top-down approach in the creation of AI. We copy the structures and processes of a person, that’s why such architecture is completely transparent for the observer. We see the logic in the conclusions, made by the AI, and the whole apparatus that forms these conclusions is available to our understanding. The AI’s reactions are predictable and understandable. It “says” such things which the person would say in its place. It does what we would do in its place. We consistently increase the capabilities of the AI, reproducing the process of a person’s growing up, and now, when the age of our sample corresponds to 7–8 years, everything that happens inside it is understandable and predictable for us, people. We have no reason to assume that its further maturation will change something in the possibilities for understanding the AI’s processes and structures. This is no more an alien, but rather, we can see our likeness. Due to the understandability and predictability of the processes and conclusions, there is no reason for fears and suspicions. The acutest task in terms of the risks, facing us now, is to minimize the risk of AI’s suicide when it grows up.

I think that a full-fledged and strong AI will be created at the turn of two approaches — Scruffy School and Neat. The neural networks are good at recognizing objects. Perfect. We do not need to understand the whole kitchen of this process, we can just use this function. But the AI’s blocks, in which decisions are made, motives and goals are formed, generalized conclusions are drawn, should be realized as an understandable structure that is similar to the human psyche (i.e., built according to the top-down approach).

Therefore, we are not afraid of a strong AI. On the contrary, we are sure that it will help mankind to solve a large number of tasks, including those problems that a person used to solve only with the help of another person, for example, the problem of loneliness.