This project sits at the intersection of AI/deep learning and psychological research into people’s ability for making sense of the behaviour of other people. When people observe the actions of others, they perceive them not only in terms of visually apparent body movements, but in terms of the meaning behind these movements: a friend’s excitement when opening a present, their disgust when brushing away a spider, or simply that they “want” the thing they’re currently reaching for. Yet, while such attributions of meaning to behaviour underpin all social interactions, the underlying mechanisms are surprisingly unclear.
This project builds upon recent advances in AI/deep learning to test novel theoretical frameworks that may explain how these attributions occur and implement them in AI/deep learning networks. These frameworks propose that people’s understanding of others’ actions emerges not from a simple reading of their behaviour, but from active attempts to project meaning onto it, to test whether what we think about others reflects what they actually do. The project harnesses the fact that strikingly similar projection mechanisms underpin recent advances in deep learning /generative models for artificial 2D/3D image generation (e.g., diffusion models) that achieve performance good-enough to fool human observers in mistaking artificial for real images. This alignment of mechanisms makes it possible for the first time to implement the proposed mechanisms in deep-learning architectures, and to test: (1) whether these mechanisms capture human-like performance and biases, (2) whether such models can, like human observers, combine information about the seen behaviour with outside contextual information, and, ultimately, (3) whether their output is good enough to fool human observers into mistaking artificially generated action videos with real ones.
The student will be part of an interdisciplinary team, with Prof Patric Bach from the School of Psychology as primary supervisor and Dr Georgios Leontidis from the School of Natural and Computing Sciences as secondary supervisor. They will be part of Bach’s Action Prediction Lab and be able to draw on its long-term expertise in human social interaction and perception, as well as of the University’s Interdisciplinary Centre for Data & Artificial Intelligence, headed by Georgios Leontidis.