Harnessing the power of automation to optimize actions is the cornerstone of artificial intelligence. Project Causica is at the forefront of expanding large-scale machine learning models for decision optimization. We delve into models that comprehend the repercussions of actions through intervention and counterfactual prediction, moving beyond mere association. By constructing a veridical world model that mirrors true-world dynamics, we aim to elevate performance across various domains.

Our research is forging a path to real-world influence in diverse fields. It converges on the nexus of action decision-making and embodied systems. These systems encompass not only real-world human actions but also artificial intelligence agents that engage in substantial interactions, whether as robots or as virtual entities within physical simulations, in a mixed-reality environment. Our research has proven effective in large-scale, real-world applications, particularly in sales and marketing, where precise intervention predictions have significantly increased revenue through targeted strategies. Our team continues to work with AI agents and explore other domains to advance the fidelity of the veridical world model.

Our codebase is openly available at https://github.com/microsoft/causica (opens in new tab), providing an easy-to-use toolkit for comprehensive end-to-end causal inference and its latest advancements. To discover more about our most recent research breakthroughs and insights in foundation models, please visit our publication page.