Strengths and weaknesses in the use of AI in the forensic field
In recent years, the introduction of artificial intelligence (AI) has dramatically transformed the working methods across various fields. For instance, within security contexts, law enforcement has long adopted Machine Learning and Deep Learning algorithms for tasks such as facial recognition and voice analysis. However, their applicability in judicial contexts remains significantly limited or, in some cases, impossible. For instance, in scenarios involving facial image comparison (face verification for judicial purposes), the current practice relies on a manual procedure conducted by an expert, employing an internationally accepted morphological approach.
In this context, the Italian Forensic Science Police Service of the Italian State Police has notably introduced a cutting-edge web portal known as AIM4SIE (Artificial Intelligence Methods for Smart Investigation of Evidence), which integrates advanced AI tools for multimedia analysis. The presentation will showcase specific examples of these analyses, demonstrating adaptive image denoising for faces or license plates, as well as the generation of aged faces or identikits. These examples will provide a comprehensive insight into the innovative capabilities of the AI tools within the portal.
Finally, some issues associated with the use of AI systems will be mentioned, including those related to bias problems and lack of transparency, encapsulated in the concept of the so called ‘black box’. We will explore how these issues can be partially mitigated and refer to the promising results of Explainable Artificial Intelligence (XAI) in addressing the ‘black box’ problem. In this context, the presentation will introduce the distinction between model-specific and model-agnostic eXplainable AI (XAI) approaches, along with the potential application of the latter in scenarios such as face recognition systems, typically reliant on closed commercial engines.