Trust and Artificial Intelligence – Qumodo Collaboration

Current Projects

Face Matching Within Human-AI Teams

Face matching is used in a variety settings which involves the verification of a person’s identity. Despite being widely used, it a surprisingly error prone task. There have been major gains in the accuracy of automated facial recognition algorithms and research has found that fusing algorithm scores with the ratings made by humans can provide almost perfect accuracy on a challenging face matching task. However, further research is required to examine how to best form these human-machine teams to carry out face matching tasks.

Current projects aims to identify benchmark levels of human and algorithm face matching performance on a dataset of image pairs and explore variations in types of errors made by human and AI. We will be investigating how we can calibrate trust between human and AI to facilitate optimal human-machine team performance. Understanding the effect on trust can help people make quicker, more effective and more accurate decisions with the help of AI and help minimise the risk of errors and misidentification in applied settings. 

Angie Ng | Funding: ERSC | Email: 2600988N@student.gla.ac.uk
Morgan Bailey | Funding: UKRI SOCIAL CDT | Email: m.bailey.1@research.gla.ac.uk

Social Intelligence towards Human-AI Teambuilding

Visions of the workplace-of-the-future include applications of machine learning and artificial intelligence embedded in nearly every aspect (Brynjolfsson & Mitchell, 2017). This “digital transformation” holds promise to broadly increase effectiveness and efficiency. A challenge to realising this transformation is that the workplace is substantially a human social environment and machines are not intrinsically social. Imbuing machines with social intelligence holds promise to help build human-AI teams and current approaches to teaming one human and one machine appear reasonably straightforward to design. 

However, if there are more than one human and more than one system that are working together we can see that the complexity of social interactions increases and we need to understand the society of human-AI teams. This research proposes to take a first step in this direction to consider the interaction of triads containing humans and machines.

How Can Computer Vision be Used to Geolocate Images of Indoor Spaces?

The project involves developing a robust and interpretable deep learning technique for analysing indoor images and other data related to human trafficking. The methods and tools will be created in collaboration with Qumodo Ltd and evaluated with their end users.

Ogechi Onuoha | Funding: Qumodo/Innovate UK | Email: ogechi.onuoha@glasgow.ac.uk
Martin Ingram | Funding: ESRC/SGSSS | Email: m.ingram.1@research.gla.ac.uk

How Is Trust Towards Technology Characterised by Users?

My research focusses on how to calibrate trust between human users and artificial intelligent systems. Optimal calibration of trust occurs when the trust from human users accurately reflects the performance of the system, so that the user does not mistrust a faulty system, nor do they distrust a well-functioning system. While users primarily base their trust on the perceived performance of these autonomous systems, this calibration can be aided by the presentation of various cues which can provide the user with a more nuanced understanding of the decision making of the system. 

Therefore my research sought to explore these factors, in experiments  participants worked with autonomous image classifier systems, which are technologies that can independently identify the contents of image data. Within these experiments, we explored how classifier performance, interface transparency, and users biases all contribute towards trust. The first of our experiments has been recently published, and we are currently working on a follow-up to this experiment. 

 

References

Ingram, M., Moreton, R., Gancz, B., & Pollick, F. (2021). Calibrating Trust Toward an Autonomous Image Classifier. Technology, Mind, and Behavior, 2(1). https://doi.org/10.1037/tmb0000032