The Robots Are Coming!

In the future, machines such as robots will be able to do all kinds of complicated tasks, such as building furniture, driving cars, playing games, and even working together with other robots. Reinforcement learning is a key approach in AI which enables machines to learn such tasks by trial and error - they try things, observe the outcomes, and learn from mistakes. In this talk, I will explain how reinforcement learning works along with some impressive results that scientists have been able to achieve - with lots of fun videos of reinforcement learning in action!

Find out how AI and robots have been used to improve the experience of patients in a real hospital memory clinic! Using generative AI and the social robot ARI, patients and their companions can have conversations with the system while waiting for their next appointment. They can ask for hospital information like directions and the cafe menu, but also enjoy jokes and quizzes, providing distraction from their otherwise stressful day. This multi-party setting in the clinic is extremely challenging, requiring computer vision, speech understanding, and gesture generation (arm, head, and eye movements). The Large Language Model-based dialogue system can decide when it should take its turn, and asks for clarification when patients pause mid-speech, improving accessibility for people with dementia. The system answers in-domain questions about the hospital, but also respond to other requests (like generating jokes or quizzes), which is particularly remarkable, as real patients often utter unexpected sentences that could not be handled previously.