Advancements in Causality and Machine Learning at Microsoft Research

Causality and Machine Learning

This content explores the intersection of causality and machine learning, highlighting research and developments in understanding cause-and-effect relationships within AI systems. It features various news and blog posts from Microsoft Research, showcasing advancements in causal inference, causal reasoning with large language models (LLMs), and the practical applications of these technologies.

Key Research Areas and Concepts:

• Causal Inference: The fundamental goal of understanding how changes in one variable affect another, moving beyond mere correlation to establish causation.
• Causal Reasoning with LLMs: Research into how LLMs can be used for causal reasoning, including their contributions to conferences like CHI 2025 and ICLR 2025, and their application in countering LLM jailbreak attacks.
• DoWhy Library: An open-source library that facilitates causal inference by providing a structured approach to causal modeling, estimation, and refutation. It has evolved into the PyWhy model, supporting the growth of causal inference research.
• Counterfactual Explanations: Using machine learning to provide explanations for model decisions through diverse counterfactuals, helping to understand why a particular outcome occurred (e.g., loan rejection).
• Adversarial Machine Learning and Instrumental Variables: Techniques that leverage adversarial training and instrumental variables for flexible causal modeling, addressing complex decision-making scenarios in various domains.
• AI for Decision Making: The application of AI to automate decision-making processes, such as personalized treatment recommendations or optimized business strategies, often by understanding causal relationships.

Featured Content and Publications:

• Research Focus: Week of April 21, 2025: Discusses contributions to CHI 2025 and ICLR 2025, causal reasoning with LLMs, countering LLM jailbreaks, and human vs. AI decision-making.
• What's Your Story: Emre Kiciman (Microsoft Research Podcast): Emre Kiciman shares his journey from systems and networking to computational social science and causal analysis in AI.
• AI Frontiers: The future of causal reasoning with Emre Kiciman and Amit Sharma (Microsoft Research Podcast): Explores the causal capabilities of LLMs and ongoing work with GPT-3.5 and GPT-4.
• Research Focus: Week of March 27, 2023: Highlights publications on machine translation (MT) models and their data learning processes.
• Research Focus: Week of November 28, 2022: Features papers accepted at NeurIPS 2022, focusing on neural information processing systems.
• Microsoft is teaching computers to understand cause and effect (TechRepublic): Discusses how causal machine learning and Microsoft's next-best-question model can improve business decision-making by replacing A/B testing.
• DoWhy evolves to independent PyWhy model to help causal inference grow: Details the evolution of the DoWhy library and its significance for scientific inquiry and data-driven decision-making.
• Adversarial machine learning and instrumental variables for flexible causal modeling: Explains how these techniques are used in automated decision-making across various fields.
• Open-source library provides explanation for machine learning through diverse counterfactuals: Introduces a library that offers explanations for ML decisions using counterfactuals.

Key Figures and Contributions:

• Emre Kiciman: A Senior Principal Researcher at Microsoft Research, actively involved in research on causal analysis, LLMs, and computational social science. He is featured in multiple podcasts discussing these topics.
• Amit Sharma: A Principal Researcher at Microsoft Research, contributing to research on causal inference and the DoWhy library.
• Johannes Gehrke: Co-host of the Microsoft Research Podcast episode featuring Emre Kiciman.
• Ashley Llorens: Host of the