Agents and Decision Systems
Decision systems play a fundamental role in informing human decision-making in critical areas where outcomes bear significant consequences, such as in medicine, economics and government policy. Similarly, investing in AI agents that can learn to act in synthetic environments, such as in games, is crucial for creating immersive and dynamic gaming experiences. Whether in critical real-world scenarios or the realm of entertainment, decision systems and agents shape the landscape of intelligent technology, underscoring their importance in enhancing user experiences and driving critical outcomes.
Central topics within this theme in the Centre for Multimodal AI are:
- General Game AI: develop deep learning, and deep statistical search using a forward model (including Monte Carlo Tree Search and Rolling Horizon Evolution), and more recently integrating foundation models (LLMs) into Game AI and Multi Agent Systems.
- AI decision making algorithms for automatic testing processes, game-playing opponents or companions to play alongside humans in games. Creative applications, include informing and in some cases automating the content creation and game design processes.
- Game AI and RL frameworks: including Table top Board Games (complex strategic many-player games), Griddly, General Video Game AI and Colosseum (Measuring hardness in Markov Decision Processes) Real-world applications of game AI.
- Methods from machine learning, statistics, probability theory, causality and psychology: We use these methods to solve problems and challenges presented by scale, complexity and variability in the context of decision making.
- Predictive and explainable decision models: We develop these models that often combine machine learning and data with human knowledge, with application to a wide range of domains including medical, legal, engineering, bioinformatics, security, sports, economics, risk and safety.
