Identification of manipulation in the dictum of neural networks

Abstract

This research aims to develop methods the manipulative acts identification by a new type of agent – neural networks. The authors analyse manipulative strategies and techniques generated by ChatGPT and Gemini (Bard) language models, taking into account their influences on the collective and individual paciens and the formation of opinions about events. Special attention is paid to the identification of functional, semantic and cognitive aspects of manipulative acts in the responses of neural networks. The study proposes a methodology to effectively assess and identify manipulative messages in the dictum of neural networks and hypothesises discourse markers that may indicate the presence of manipulative influence in the use of neural networks. It is concluded that the complex and ambivalent nature of manipulative markers in electronic dictum makes their typologization much more difficult. The results obtained may be used for the development of algorithms aimed at improving the reliability and efficiency of neural networks in various application areas, including natural language processing, data analysis, and others.