STAI CDT PhD students Alex Goodall and Omar Adalat took part in the Great Exhibition Road Festival on Saturday 7 and Sunday 8 June, joining researchers, artists and members of the public for a weekend of hands-on activities across Imperial’s South Kensington campus.
Alex and Omar are part of the Formal Methods in AI (FMAI) Lab at Imperial College London. The Lab’s exhibit, ‘Can You Train AI to Stay Safe?’, invited visitors to take part in a live Reinforcement Learning challenge. Participants chose between Pac-Man and MetaDrive, adjusted key training parameters through an interactive quiz, watched their AI agent learn in real time, and compared their performance on a public leaderboard.
The activity introduced visitors to key concepts in Constrained Reinforcement Learning, including balancing reward maximisation with safety requirements. Through choices such as future focus, the cost of safety mistakes, learning speed, and safety-first versus reward-seeking policies, participants explored some of the same challenges researchers face when designing trustworthy AI systems.
Alongside fellow lab member Terrence Fernandes, Alex and Omar spoke with visitors about the demonstration, their research, and the role of formal methods in improving the safety and reliability of AI systems.
The exhibit proved popular with festival-goers, with 218 visitors taking part in the challenge. Of these, 118 chose to play MetaDrive, while 100 selected Pac-Man, giving participants the chance to experience AI safety concepts through two different environments. Many visitors returned for multiple attempts to see how different training choices influenced outcomes and to try the alternative environment.
Reflecting on the experience, Omar said: “The public reaction was overwhelmingly positive. We hope that the public started thinking in more detail about constrained optimisation and how safety has to be a design principle of AI solutions.”
For Omar, the festival provided an opportunity to develop his communications skills, focusing on distilling complex research themes into their simple essence for people of all backgrounds. “As researchers, technical terms and jargon naturally become ingrained within us, and we mostly interact with colleagues who understand these terms.”
The experience reinforced the importance of public engagement in research. Omar added: “Scholarly articles are often written by academics, for academics, making them inaccessible to many people outside a specialist field. Public engagement helps bridge that gap, supports the real-world application of research, and enables people to better understand the work being carried out with public funding and its wider implications.”
Following the strong interest and engagement from the public, the researchers hope to bring an expanded version of the demo to future events, with even more opportunities for visitors to explore the challenges of building safe and trustworthy AI.
We’re proud of Omar and Alex for their public engagement and agree that it is essential for helping to make complex research more accessible and fostering meaningful dialogue between researchers and society.