Modelling general knowledge and common sense

Human augmentation systems combined with artificial intelligence have great potential for facilitating and improving cognitive performance, particularly under conditions of stress or high cognitive load. Our goal is to apply user-centred design principles to create augmentations that are intuitive and effortless to use, and that provide seamless integration between human and machine intelligence. We welcome applications in this space related to the four specific topics details below.

Modelling General Knowledge and Common Sense 

Modern AI systems are statistical models. They can generate the most likely sentence in response to a prompt or classify every pixel in an image but there is no intrinsic meaning in either of these outputs. Humans understand the world in terms of objects and their relationships with the world.

Object rules such as permanence and size invariance help us keep track of things without needing to re-analyse every pixel constantly. Physical rules help us predict where things are when they become occluded and limit possible outcomes. Modelling these rules is essential for an independent AI system working in the real world. Previous attempts at modelling rules with knowledge bases, logic, knowledge graphs etc. have failed to capture the relationship between sensory input and models of reality.

New representations such as word and image embeddings have started to provide a way to connect sensory data to knowledge, but only in a very limited way. This PhD would aim to build far richer models that provide a grounding for sensory inputs in relation to a world model. Methods to be explored include fact embedding, knowledge graphs and multi-sensory fusion to map sensory inputs to rich semantic meanings. If successful, the result will be the ability to build AI systems that are more reliable, more open and explainable, and able to ‘learn by being told’ rather than relying on huge quantities of data. 
 
The ideal candidate for this PhD would have experience in building machine learning models, good programming skills and an interest in how machines and humans learn. A degree (BSc. Or MSc.) in a relevant subject such as Computing Science, Artificial Intelligence or Mathematics is required. You will be based in the Division of Computing Science and Mathematics but also work closely with colleagues in Psychology, drawing on expertise in human knowledge representation and sensory processing.

Accessible and inclusive extended reality 

This project will investigate the physiological and cognitive effects of extended use of virtual reality (VR) and augmented reality (AR), focusing on visual strain and its impact on visual health, mental performance, and operational effectiveness. In these extended reality (XR) systems, a conflict between accommodation (focusing the eye’s lens) and vergence (coordinated eye movement for binocular vision) can occur when virtual objects are perceived at a different distance than the display surface, leading to visual discomfort, eye strain, and cognitive fatigue.

These problems are amplified when properties of the display are not well matched to the individual user, for example assuming an incorrect interocular separation. Extended use of XR also creates a risk of visual stress. Symptoms of these problems include visual fatigue, reduced clarity of vision, and increased cognitive load, all of which can impair performance and situational awareness. While XR enhances situational awareness through real-time information overlays, prolonged use of digital augmentations remains an under-researched aspect of XR. There is also a pressing need to understand how these systems should be designed to maximise their accessibility to neurodiverse populations, and to accommodate individual differences.

The study aims to combine subjective, behavioural, and neuroscientific research methods and to produce a comprehensive understanding of visual strain in extended use of XR. This will be used to evaluate optimal respite times needed to reduce visual stress and eye strain, to provide evidence-based recommendations for reducing overall discomfort, and to maximise accessibility, inclusivity and usability. 
 
A degree in psychology or a related subject is required. The ideal candidate would have experience in psychophysical and/or physiological methods (such as EEG or fNIRS) and an interest in vision science and its application in XR.

Trust and reliability in AI

AI provides huge potential for augmenting human decision making. It can out-perform humans on some tasks, such as object and face recognition, provide results in real time and, unlike people, is not limited by a single attentional spotlight, thus supporting multitasking. However, it is not always accurate, and can be susceptible to incorrect decisions and hallucinations. In addition, in many situations it remains important that identifiable humans take ownership for their decision making. AI creates a challenge in learning how to trust a potentially unreliable information source, and not to be overly reliant on technology for decision making.

This project will assess decision making and confidence when sharing information and judgements between teams of humans, and humans interacting with AI systems. We will identify current limitations of AI, but also experimentally model and manipulate its accuracy and reliability. We will use these approaches to assess human ability to calibrate their confidence in unreliable information sources. 
  
A degree in psychology or a related subject is required. The ideal candidate would have experience in experimental design and data analysis, and an interest in applying these to interdisciplinary research. You will be based in the Division of Psychology but also work closely with colleagues in Computing Science and Mathematics, drawing on expertise in AI.

Intuitive extended reality for three-dimensional scene understanding

Extended reality (XR) systems allow users to monitor information from a broad range of sources, including cameras, drones, AI systems, and other humans. These can be used to share information and to support joint decision making, but create a risk of overwhelming the user with visual clutter and cognitive overload.

This project will develop novel ways of embedding this information perceptually in the three-dimensional environment, with the goals of reducing cognitive load and visual clutter, and providing critical spatial information directly without the need for cognitive remapping.

The project will also evaluate the effectiveness of this approach in applications such as search and rescue, particularly under conditions of high cognitive load.  
 
A degree in psychology or a related subject is required. The ideal candidate would have experience in psychophysical methods and an interest in vision science and its application in XR.