Interpretability and explainability from a causal lens

Judea Pearl
University of California, Los Angeles (UCLA)
Computer Science

I will describe the task of interpreting and explaining data as seen through the science of cause and effect, and distinguish it from the task of interpreting algorithmic systems. The former calls for a mapping between data and the ropes of reality, the latter between data and the intentions of the system builder.
    Reference: J. Pearl "The Limitations of Opaque Learning Machines," 

          Chapter 2 in John Brockman (Ed.), Possible Minds:

Presentation (PDF File)

Back to Workshop II: Interpretable Learning in Physical Sciences