Our STAI CDT Annual Showcase took part last month on 10 June. The event was once again co-located with the King’s Department of Informatics Showcase, a London Tech Week fringe event, to maximise engagement from external attendees.
The showcase was open to industry, public and third sector organisations who were interested in collaborating with the CDT and King’s, and in meeting emerging talent for AI and technology focused roles. It was a hugely successful event with over 200 industry parters attending.
STAI CDT Panel on AI Assurance
As part of the showcase, we hosted a panel discussion on AI Assurance, which was moderated by STAI CDT co-director Dr Caitlin Bentley, and featured important insights from:
• Ben Gilburt, Data Ethics Lead, DSIT
• Chris Barnes, Head of Science of AI, National Physical Labaratory
• Tess Buckley, Senior Programme Manager in Digital Ethics and AI Safety, Tech Uk
The panel discussion was conducted under the Chatham House Rule to encourage open and candid dialogue. Panel members were invited to reflect on several key questions relating to safe and trustworthy AI and the emerging field of AI assurance. They discussed how their organisations are working to develop and deploy safe and trustworthy AI, and what AI assurance means in practice within their respective contexts. The conversation also explored the current state of AI assurance in real-world deployments, highlighting areas where organisations are already demonstrating good practice as well as the most significant practical gaps that remain. Panel members shared their perspectives on the first steps organisations deploying AI should take to demonstrate that their systems are trustworthy. Finally, they considered how, in the context of rapid AI innovation and the growth of the AI assurance ecosystem, a credible evidence base and robust standards for AI assurance can be developed.
Research Posters and Demos
The showcase was also an opportunity for current second and third year STAI CDT PhD students to showcase their research through poster presentations and live demos and engage with partners and interested researchers.
All attendees were invited to vote for their favourite poster presentation, and we were delighted to congratulate Akchunya Chanchal, who was awarded the STAI CDT Poster Prize.
Akchunya’s poster discussed one of his recent papers, ‘I Am Big, You Are Little; I Am Right, You Are Wrong’, which studies how Vision AI systems make decisions. Akchunya and his co-authors’ research show that performance metrics alone can give a misleading picture of how well a model performs, particularly when considering safety and alignment. In some cases, the highest-performing models were found to make decisions in a more rigid or narrow way. The findings suggest that decisions should not be solely made based on performance benchmarks when selecting AI models. Instead, it’s also important to consider explainability in representative real-world use cases to better understand how models arrive at their decisions and whether those decisions are appropriate, safe, and trustworthy for the intended task.
The poster presentation is an important opportunity for PhD students to explain their research to a range of audiences. Reflecting on his experience, Akchunya said, “I really enjoyed the challenge of explaining my work at different levels of abstractions, going into more technical details when conversing with a technical person, while also having to motivate the direct use-case and impact of the work in a business use-case to the more business-inclined person. It forced me to think more broadly about how to be more effective when communicating my research”.
Overall, the event was a great success, celebrating the innovative research taking place across the STAI CDT and setting the stage for another inspiring showcase next year.