Please note, this demonstration is only for King's students and staff.

Humans can feel unease if something doesn’t look quite right, but what about an AI? 

In complex models where accuracy is never guaranteed, eg weather forecasting, providing a measure of certainty is useful. Knowing there is a 90% chance of rain, you might well carry an umbrella. If it’s only 20%, then you might choose to risk it and save carrying the brolly. AI systems are being developed for image-based tasks in medical, policing, and autonomous vehicle settings. These systems can be very accurate but are never perfect and errors do occur. The effect of errors in these high-risk settings can be very significant, eg a patient undergoing an unnecessary operation. AI answers are often presented without additional context such as (un)certainty, which could allow for better informed decisions in high-risk settings.

This tabletop activity explores these ideas via an interactive demonstration. Would uncertainty aware AI could be safer, more trustworthy, and better at undertaking their tasks?

This demonstration is being run by Sam Goring, a PhD student on the UKRI CDT in Safe and Trusted AI.

This event is in-person only. Registration is not required

This demo is part of the King's Festival of Artificial Intelligence, running from Tuesday 21 to Saturday 25 May 2024.