ALICE

Developer(s)	Microsoft Research Lab - New England (Subsidiary of Microsoft)
Type	Economics Artificial intelligence Decision-Making
Website	www.microsoft.com/en-us/research/group/alice/

ALICE is an Artificial Intelligence project initiated by Microsoft Research, known as Automated Learning and Intelligence for Causation and Economics. The project focuses on leveraging state-of-the-art machine learning techniques combined with econometrics to enhance economic decision-making processes.^[1]

The primary goal of ALICE is to measure causation in economic systems, which is crucial for making informed policy decisions. This involves understanding the reasons behind the movements within complex economies. The project builds on Microsoft's long history of integrating Economics and Computer Science, bringing together researchers from various fields such as Social Science, AI, and Machine Learning. The ALICE team aims to scale up the adaptation of existing ML technologies for economic applications and develop new deep learning architectures for causal inference.^[1] Their research addresses practical policy-relevant applications, including demand estimation, price optimization, effectiveness of advertising, sales strategies, and designing incentives for desirable healthcare and education outcomes. This endeavor aims to democratize economic research using AI, while simultaneously advancing the frontier of AI through economic theory.^[2]

One notable aspect of the ALICE project is the collaboration with TripAdvisor. A case study between Microsoft Research and TripAdvisor explored the use of causal AI for customer segmentation,^[3] This partnership emerged from a chance encounter between data scientists from both organizations, leading to a joint effort in understanding the impact of a membership model on user engagement. By leveraging an A/B test, the ALICE team developed a new statistical method to measure the direct effects of membership on engagement. This approach, which builds upon traditional instrumental variables techniques, revealed significant variation in user engagement based on the platform used and pages visited by the user.^[3]

The collaboration resulted in valuable insights for TripAdvisor. The ALICE team found that membership positively affects user engagement, with significant variation among users. The major drivers of this variation included the platform from which the user accessed TripAdvisor and the pages they visited before the experiment. The key innovation was developing an ML-based method for estimating heterogeneous causal effects in A/B tests with non-compliance, which allows for complex individual-level differences in both compliance and the intervention's effect.^[3]

The methodology has been implemented in the EconML software package, an open-source Python library developed by the ALICE team. EconML applies machine learning techniques to estimate individualized causal responses from observational or experimental data.^[2]

Principal Economist Eleanor Dillon currently leads the ALICE project at the Microsoft Research Lab - New England.^[1]