Practical AI avatar

Dexa/Practical AI

Learn more
Published Apr 25, 2023

Causal inference

Delve into the world of causal AI with insights from Paul Hünermund, as he distinguishes it from traditional AI methods and showcases its transformative potential in industry through practical applications and advanced inference techniques, essential for data-driven decision-making.
Episode Highlights
Practical AI logo

Topics covered

Popular Clips

Episode Highlights

  • Experimental Methods

    Experimental methods in causal inference, such as A/B testing, are often seen as the "big hammer" for tech companies due to their straightforward application and ease of understanding. explains that while these methods are effective in controlled environments, they can be costly and sometimes unethical in complex scenarios like vaccine testing during the COVID-19 pandemic 1. He suggests that in cases where businesses can control variables, like a candle manufacturing business deciding on packaging, experimental approaches are beneficial 2.

    Experiments are easy in many domains to set up, easy to understand, and you don't need a lot of background knowledge for it.

    ---

    However, the challenge lies in determining the sample size and understanding the selection process to draw robust conclusions.

       

    Observational Approaches

    Observational approaches in causal inference are crucial when active experimentation isn't feasible. highlights that these methods involve passively observing data to deduce cause and effect, often facing challenges like self-selection bias 1. He notes that tools like causal random forests are developed to efficiently partition populations and reduce costs while maintaining robust conclusions 3.

    We don't actively intervene ourselves, but we passively observe the data and still want to get cause and effect out of it.

    ---

    These methods are particularly useful in scenarios where random assignment isn't possible, such as assessing the returns to college education.

       

    Emerging Techniques

    Emerging techniques in causal inference, like causal discovery and root cause analysis, are gaining traction in modern data science. shares that these methods are promising for extracting causality from observational data, though they require good benchmarking datasets 4. He also mentions the importance of integrating background knowledge into causal workflows, highlighting tools like Microsoft's DoWhy package that streamline the causal inference process 5.

    How far can we go with just simply trying to get out causality from observational data?

    ---

    These advancements are essential for addressing complex problems and enhancing the practical application of causal inference in various domains.

Related Episodes