DEEP RISKS OF MACHINE LEARNING
Dmitriy Vladimirovich Vinnik
Financial University under the Government of the Russian Federation, 49, Leningradsky Prospekt, 125993, Moscow, Russia
Keywords: artificial intelligence, neural networks, machine learning, risk, logical transparency, optimal brain surgery, perceptron, network-centric warfare, bureaucracy, cognitive functions
Abstract
The article assesses specific machine learning technologies (as methods of computer realization of specific cognitive functions) in terms of the need for rational explanation of their work, reliability and efficiency, as well as related social properties (responsibility for the use of results, the threat of degeneration of human skills, etc.). The risks of introducing machine learning technologies in various types of activities and various fields are analyzed in view of whether logical transparency is a crucial criterion or not. The conclusion is made that the logical opacity of most machine learning results is a serious challenge to rational thinking and reasonable argumentation in public administration It is argued that negative consequence of the imprudent introduction of neurocomputers may be a partial or complete loss of personnel’s skills to perform routine intellectual procedures (filling in tables, arithmetic calculations, classifying information and documents into categories, self-service research for information).
|