?
Метрологическая модель процесса оценивания функциональных характеристик систем искусственного интеллекта
When making decisions on the possibility of using artificial intelligence systems for solving critical data processing and control tasks, it is crucial that the consumer and other stakeholders understand the functional characteristics of these systems under the foreseen operating conditions. The article attempts to formulate and interpret the task of assessing the functional characteristics of artificial intelligence systems in terms of metrology. It is shown that in the metrological context the task of evaluating the functional characteristics of artificial intelligence systems can be considered by analogy with the conformity assessment of measuring equipment. In this case, the latter represent test data sets, the representativeness of which determines the measurement error of functional characteristics. The mechanism of measurement error formation is examined. The models with the use of reference data sets, assessment of the representativeness of test data sets and reference machine learning algorithms are proposed as measurement models.