The use of gameplay data to inform high-level AI decision making
Commercial digital games pose unique challenges for AI systems. Not only must a game’s AI system provide the player with the impression of intelligence, but it must do so with heavily limited computational resources. Often, the AI system must be able to cede control to hard-coded scripts or behaviours on demand, in order to help tell a game’s story. The system should be accessible: it should be simple to add new behaviours, for example, and a designer with limited programming experience should be able to tweak the AI to their satisfaction. And if the AI does something unexpected, it should be able to explain itself, so that the cause of that behaviour can be investigated and changed if necessary.
These requirements have historically been met by techniques like behaviour trees. Unfortunately, such techniques are unable to generalise to new situations. As games become ever more complex, designers are forced to fall back to “generic” behaviours in unforeseen situations, which, by their nature, do not always appear to be particularly intelligent to a player. Occasionally, the consequences of a hard-coded rule will lead to AI failures that negatively affect a player’s experience.
My research focuses on how AI techniques such as neural networks and search algorithms can be used in games to create better, more believable NPCs, while staying within computational budgets and respecting the needs of game designers.
My undergraduate degree was in Theoretical Physics at the University of York, with my MPhys project conducted at Dyson Ltd. After graduating in 2015, I spent a year at the University of Edinburgh studying for a MSc in Artificial Intelligence. Now back at York, I’m looking forward to applying my skills to create better games.
Home institution: York
Ready to apply?
Once you have identified your potential supervisor, we would encourage you to contact them to discuss your research proposal.Learn More