CHIA conference: 11 June, 2024

Prof Mateja Jamnik, University of Cambridge

How can we make trustworthy AI?

Not too long ago most headlines talked about our fear of AI. Today, AI is ubiquitous, and the conversation has moved on from whether we should use AI to how we can make the AI systems that we use in our daily lives trustworthy. In this talk I look at some key technical ingredients that help us build confidence and trust in using intelligent technology. I argue that intuitiveness, adaptability, explainability and inclusion of human domain knowledge are essential in building this trust. I present some of the techniques and methods we are building for making AI systems that think and interact with humans in more insightful and personalised ways, enabling humans to better understand the solutions produced by machines, and enabling machines to incorporate human domain knowledge in their reasoning and learning processes.

Mr Simon Knowles, FRS, Graphcore

Costs of scaling AI

AI is emerging as a general-purpose technology which will have great impact on many areas of human life. In its current neural-network form, AI requires a lot of computation. This coincides with the collapse of historical cost and energy scaling trends for computation in silicon. So realising the potential of AI might be rather expensive -- what can we do about that?

Prof Cecilia Mascolo, University of Cambridge

AI for wearable data

Wearable devices are becoming pervasive in our lives, from smart watches measuring our physiology to wearables for the ear accompanying us in every run or virtual meeting. The monitoring of our health and fitness through sensors and wearables is the focus of much research in the community, however, despite advances in AI on wearable data, many challenges still remain before truly scalable, trustworthy and affordable wellness monitoring becomes a reality. In this talk I will discuss where commercial systems have gotten to today and highlight the open challenges that these technologies still face before they can be trusted health measurement proxies. Namely, the ability to work in the wild and to cope with the variability of uses; the trade offs that we need to consider with respect to the sensitivity of the data and the use of constrained on device resources; the uncertainty of the prediction over the data and the crucial need for robust, open AI frameworks to benchmark and assess performance in the context of critical applications such as health and fitness. I will mostly use examples from my team’s ongoing research on AI for health and fitness, on-device machine learning and “hearable” sensing to explore the current and future path of AI for wearable data.

Prof Yi Zeng, Chinese Academy of Sciences

From Brain-inspired AI to Sustainable Symbiotic Society

I will firstly raise some concerns for current AI and discuss the difference between data-driven AI and mechanism-driven AI. then I will extend the discussion from the perspective of Brain-inspired AI, and how it can contribute to a ecological-centric sustainable symbiotic society. To realize this vision, I will investigate on the current status of AI for Sustainable Development Goals, and introduce the effort of AI for Biodiversity and AI for Cultural Interactions from Center for Long-term AI and Chinese Academy of Sciences.