Neural Networks in Forensic Expertology and Expert Practice: Problems and Prospects

Abstract

The article, from the perspective of the theory the forensic activity digitalization as a particular theory of forensic expertise, examines the prospects for the introduction of neural networks in forensic examination and the current problems arising in this case. The author notes changes in the methodology and technologies for developing expert techniques in connection with the introduction of artificial intelligence algorithms — neural networks. The areas neural networks’ application for solving various problems of science and practical activity are outlined. Specific examples demonstrate the possibilities of using supervised learning algorithms for neural networks in forensic practice. A detailed analysis of the reasons why the use of neural networks in forensic science can lead to erroneous conclusions is given. Particular attention is paid to hallucinations of deep learning neural networks on large language models. There is a danger that an expert, relying entirely on a neural network, may give the wrong conclusion, since self-learning generative neural networks do not provide an explanation for why they made a particular decision. To develop expert methods for solving typical expert problems based on neural networks, it is proposed to create databases (Dataset) for various forensic objects for analysis and machine learning. To store the Dataset, it is necessary to organize repositories that can contain data sets on types (kinds) of forensic examinations. Dataset and repositories will provide data quality control and model verification. The article substantiates the need for new competencies of a Data Scientist — a specialist who develops tools for solving forensic problems when introducing neural networks and other artificial intelligence algorithms into forensic science, as well as a machine learning engineer working in contact with him.