Artificial intelligence for translators

Today everyone knows how neural networks work, a groundbreaking technology in many fields of activity. Such as in the field of translation. Previously, machine translation was an assistant – today, at first glance, it is an aid to human work.
Neural networks are based not on algorithms, but on the knowledge obtained from people. The analysis of information and the application of the obtained new structures in practice is the principle of machine translation of our days. With each newly analyzed array of information, neural networks build more and more connections, and the translation becomes more and more qualitative. Why can’t automated translation completely replace human translation in this case?
According to linguists, today you can rely on machine translation for scientific texts and documents. In texts with a large number of so-called “scripts” and unambiguous terms. Where logical connections are easy to establish, the phrase has meaning: “There is no other way to say it.” But experts are convinced that a human editor is indispensable even for machine translation of similar texts created according to templates.