Human performance and next-generation facial identification technologies.

点击翻译按钮获取中文摘要

Our perception and recognition of human faces is central to life. Our brains continuously make perceptual judgements, e.g. whether a face is familiar or a facial expression indicates happiness. We can generally introspect to assess our uncertainty in these judgements, attributing confidence. Confidence attribution is studied as a canonical example of metacognition, with the capacity to reflect on our thoughts considered an archetypal characteristic of consciousness.Recently, there have been huge technological advances in Artificial Intelligence (AI)-powered synthesised material, including faces. The Generative Adversarial Networks (GANs) deep-learning method can synthesise images of realistic-looking fictional people (Fig1). This raises questions about how humans perceive and introspect about real versus AI-generated faces, and facilitates large stimuli sets that can vary on several dimensions (e.g., similarity, ethnicity) for testing theories of face perception, recognition, and metacognition, and identifying their neural correlates.Jointly studying human experiences and GANs-technologies vastly advances theoretical frameworks: New perceptual reality monitoring accounts propose that human consciousness operates computationally like GANs (Lau, 2019). Joint study also opens new avenues for practical application in multiple domains, including forensics such as the development of face-composites from witnesses' memory, identity-parades,and mugshot books. Visionmetric (https://visionmetric.com/), is a successful micro-SME in the Security and Policing sector, best known for its innovative facial-composite system EFIT6. EFIT6 is licensed to ~75% of UK police constabularies, and law enforcement agencies in 30 countries. Visionmetric wants to expand their product-base to include identity-parades and mug-shot books using GANs-generated faces, and further develop systems to extract memories from a witness (e.g.,via EEG) without relying on their behavioural response.This project combines Visionmetric's industry-grade databases and software tools, with Colloff and Bowman's theoretical understanding of face recognition and neural-correlates of familiarity to answer important theoretical and applied questions about face perception, recognition, and introspection:1) Can humans detect a real face from a group of simultaneously presented GANs faces?2) What factors (e.g., face similarity; ethnicity, gender or age of the observer and the faces) influence human:(a) detection of real faces from GANs faces?(b) judgements of real and GANs faces (e.g., trustworthiness)?(c) recognition memory for a previously seen real face when surrounded by a group of GANs faces?3) Can we identify electrophysiological markers of face familiarity and confidence to directly measure face recognition memory and introspection from the brain?