ARTIFICIAL INTELLIGENCE AND HUMAN EXPERTISE IN CRIMINAL INVESTIGATIONS: INTEGRATING NLP TECHNIQUES TO ENHANCE JUSTICE

This study explores AI's role in analyzing language interactions within investigative chambers and courts, using natural language processing (NLP) technologies such as Bert and GPT-4. The focus was on assessing questions posed and responses provided by witnesses or defendants, where AI was used to analyze emotional biases, inconsistencies in testimonies, and deliberate ambiguity in responses. The study draws on AI tools to detect violations of conversational principles, such as ambiguity and asymmetry, that may affect the integrity of investigations. The study results showed that AI can identify violations of conversational principles with an accuracy that exceeds human analysis by up to 30%, with GPT -4 and Roberta detecting violations such as intentional ambiguity (45%) and asymmetry (40%) in responses. AI has also shown a high ability to generate complex investigative questions that reflect legal and social context, such as questions intended to link testimonies with available evidence, enhancing analysis accuracy and reducing emotional biases that may affect human investigators. However, several challenges associated with the use of AI have been identified in this context, most notably algorithmic bias, where data biases used to train models may affect fairness and transparency in outcomes. The black box problem, which relates to the difficulty of interpreting how models make their decisions, raises questions about transparency and accountability in the judicial system. The study found that integrating AI with human expertise could improve criminal justice through hybrid judicial platforms, where AI is used to analyze raw data. In contrast, the human investigator remains responsible for interpreting and making final decisions. Based on these findings, recommendations were made to develop ethical frameworks to ensure the transparent and reliable use of AI in judicial investigations.