O papel da perícia sobre a prova em vídeo na era das deepfakes: um estudo de casos

Abstract

In the digital society, the spread of videos and the advance of deepfakes strain the reliability of video evidence and bring forensic examination to the center of debates on authenticity and integrity. This study investigates the role of forensic examination over video evidence by reference to two high-profile cases flagged as possible deepfakes, examining the qualification of experts, the methods and techniques employed, the use or nonuse of artificial intelligence for detection, and the structure of forensic conclusions. Guided by the rationalist conception of evidence, according to which the function of evidence is the knowledge of facts, the study adopts the premise that only falsifiable and potentially reliable knowledge should enter the process, so quality standards must be set for the admissibility of a video, especially in view of the cognitive and interpretive biases of this digital means of proof. The need to state and apply probative standards for the admissibility and evaluation of video evidence is sustained, including minimum quality requirements for the material. Deepfakes, as byproducts of artificial intelligence and constantly improved by machine learning, can be detected through three avenues, namely through the chain of custody, through tools specifically developed for that purpose, and through human-based analysis, and the epistemic limitation of each method is addressed. The case study, which analyzed eight forensic reports, identified that the third avenue, grounded in the human eye as the predominant tool, was the one used by experts through frame comparison and comparative observation of elements such as lighting and synchrony, with software serving only as support and with indirect mentions to the chain of custody. The case study confirmed the importance of establishing quality standards, since limitations such as low resolution, sharpness, and lighting, which affect inferential robustness, were indicated by experts as epistemic limitations. It is concluded that hybrid detection methods that combine a rigorous chain of custody, tools for detection based on artificial intelligence, and a scientifically and methodologically guided human comparative examination, together with literacy and visual literacy as transversal competencies for reading videos, are the best instruments to prevent deepfakes from compromising the truth pursued as an ideal and attainable aspiration.