The emergence of generative artificial intelligence (GAI) technology, which has made it possible to introduce resources such as ChatGPT, Deepseek, etc. into private and professional life, has begun to influence the professional practices of psychologists. The purpose of our research is to determine whether it is possible to use the "explanations" provided by so-called AI when solving classification problems in psychodiagnostics to address the issue of expert trust in machine decisions. The corpus of essays written by depressed patients and healthy people was divided into two classes by a large language model (LLM), and each text was accompanied by an explanation of the machine's decision. Various models and prompts were used for classification. The results of the model that achieved the highest classification accuracy (based on the F1-measure) were subjected to expert analysis. A psychological typology of errors found in the "explanations" generated by GII was developed, and the issue of trust in "AI explanations" was addressed.
Download the proceedings (PDF) at the website of Institute of Psychology (in Russian): https://pirao.ru/science/publications/publication/117483/
Download the proceedings (PDF) from eLibrary (in Russian, registration required): https://elibrary.ru/item.asp?id=90488307
Article page at eLibrary: https://www.elibrary.ru/item.asp?id=91288132
Proceedings at the HSE website: https://publications.hse.ru/en/books/1203086819
Чуганская А. А., Чудова Н. В., Кузьмин Г. Ю. Страхи перед робототехникой в зеркале сетевых дискуссий // Экопсихологические исследования — 7: экологичность взаимодействий человека с окружающей средой: сборник научных трудов / Федеральный научный центр психологических и междисциплинарных исследований ; под общ. ред. чл.-корр. РАО В. И. Панова ; ред.-сост. Е. С. Александрова, Э. В. Лидская. — М.: ФНЦ ПМИ, 2026. — С. 425–429.