Abstract
Forensic science has played a central role in criminal investigations since the late 19th century but continues to face challenges. Scrutiny arising from the expert testimony of FBI hair analysts raised concerns about the validity and reliability of forensic hair analysis. However, the underreported root cause was not based on science but on management practices. This research addresses four key aims: (1) examining current forensic hair analysis methodologies, (2) understanding the workplaces and attitudes of the forensic science community, (3) developing objective methods, and (4) evaluating these methods.This thesis by publication comprises six studies. The survey research filled a substantial gap in understanding forensic workplaces and the perspectives of forensic scientists and forensic hair analysts on worldwide scale. HairNet, a computer vision, deep learning, supervised machine learning model, was successfully developed and deployed as an objective method to determine the human or non-human origin of hair and assess the suitability of human hair for DNA analysis. Supervised machine learning was found to be the most effective application of narrow AI to forensic science compared to other emerging technologies and objective methods. While narrow AI has significantly advanced in the 21st century, this research considers its limitations in forensic science, particularly the lack of critical thinking inherent in human expertise. This research emphasises the need for continuous development of forensic methodologies, especially objective methods, amidst robust forensic governance. Success of forensic science depends on balancing any emerging technological advancements with the perseverance of expertise.
The recent Sydney Declaration, introduced at the 22nd meeting of the International Association of Forensic Science (IAFS) in 2023, is a substantial attempt to retain this balance and highlighted the importance of worldwide collaboration to address future challenges. The notion of narrow AI possessing true intelligence remains within the realm of science fiction; however, understanding its limitations and the distinct differences between narrow AI and human expertise is critical, as is a renewed perspective on expertise for forensic hair analysis in light of this research, HairNet and other similar emerging technology capabilities.
| Date of Award | 2025 |
|---|---|
| Original language | English |
| Supervisor | James ROBERTSON (Supervisor) & Wanli MA (Supervisor) |