Medical diagnostics can be mentioned as an important application area of AI, but it does not mean that AI should be used for scientific reasoning, it can be a secondary or tertiary tool which can offer some hints to reduce human scientific efforts and tasks. As it is known, AlphaFold cannot explain why and how it finds promising protein’s 3D structure, it is a black box in this meaning (a Nobel Prize awarded box). And do not forget the noise effect, ANNs tend to learn a lot of noise (maybe the structured parts). For using deep learning algorithms in industry, the attention should be called to some security problems already discovered. Is "Deep learning" as a term just a marketing trick? Because almost everything was developed several years ago, e.g. Kolmogorov-Arnold representation theorem and so on. Of course, a new AI (algorithmic, systemic, methodological etc.) development can have scientific results, but a pure AI application?