What’s hype and what’s real - the future of AI safety

Can we trust AI?

A packed audience joined Senior Lecturer Dr Richard Hawkins and Research Fellow Dr Philippa Ryan at York Explore at the York Explore Library for their presentation ‘Can we trust AI to keep us safe?’ where they discussed what current AI technologies can do and also what the limitations of such systems are. Dr Ryan and Dr Hawkins then took the audience through examples of where and why AI-enabled autonomous systems, like cars, might go wrong and why safety assurance has a big role to play in the safe rollout of such systems across society. 

Dr Hawkins said: “Everyone is hearing a lot about AI at the moment, and it can be quite confusing. It is also difficult sometimes to know what is just hype and what its real capabilities are. Members of the public also perhaps don't understand how AI will affect them and may be concerned about whether this new AI is safe. 

“Our audience seemed particularly interested in our discussion about whether AI in a self-driving car would do a good job when driving on the streets of York. I think this was particularly engaging because it is something that they think could happen at some point in the not too distant future, and they are concerned about how it could impact them as York residents. 

“We were able to show examples of why somewhere like York is a real challenge for AI technology and describe how our research may be able to help address some of these challenges.”

Seeing safe AI in action

The following weekend visitors had a chance to see this in action at the Centre’s base in the Institute for Safe Autonomy, where Dr Eng. Paulina Lewinska and Dr Calum Imrie ran an aerial drone demonstration in our large lab, Test Space 1. The basis of this demonstration was the ASPEN project run by Professor Radu Calinescu and Dr Imrie. Our team showed the public how an unmanned aerial vehicle (UAV) with camera and embedded object detection algorithm works with our wheeled robot, nicknamed Husky, to autonomously detect a diseased plant or tree and retrieve a sample. In a real-life scenario, trekking through forests and dense woodlands presents multiple dangers for humans, such as needing to climb to retrieve a sample. By doing this autonomously, we can reduce the risk to humans.

Dr Eng. Lewinska said: “This demonstration presented a good opportunity to show the public both the capabilities of various autonomous systems and their limitations. In the scenario we presented at the Festival of Ideas the drone could fly autonomously and identify the sick plant, then the operator can send the Husky, also autonomously, to retrieve the sample. Once the Husky is at the location a human operator can take over manual control of the Husky to cut and retrieve the sample. The Husky can then resume autonomous operation and return the sample to the waiting operator.”

Watch a video of the demonstration from the day of the drone and Husky in action below (video by Dr Eng. Lewinska and Dr Imrie ).