Assessing the Admissibility and Reliability of Evidence Obtained Using Artificial Intelligence

The obtaining of evidence is increasingly mediated by artificial intelligence — from facial recognition and traffic analysis to the detection of synthetic media content — which compels a rethinking of how such evidence is assessed. The subject of the study comprises the criteria for assessing the admissibility and reliability of evidence obtained with the use of artificial intelligence, their relation to the general rules of proof, and the specific problems that artificial intelligence creates at each level of assessment. The aim of the work is to distinguish the assessment of admissibility from that of reliability and to propose a system of working criteria for each level. The methodological basis is the formal-legal, systemic and functional methods combined with elements of the comparative-legal approach. It is substantiated that admissibility and reliability are two independent criteria of assessment that must not be conflated: the output of an algorithm may be obtained by lawful means yet remain unreliable, and vice versa. Two systems of criteria are formulated — of admissibility (a proper subject, source, method and form) and of the reliability of the result of algorithmic analysis (validation of the method, known error rate, reproducibility, explainability, integrity of the data, documentation). The authors conclude that the output of artificial intelligence does not constitute a new type of evidence and enters the case through the existing means of proof, while a nonverifiable (closed) algorithm cannot serve as the basis of reliable evidence.