AI Needs More Why

The father of Bayesian networks and probabilistic reasoning, Judea Pearl, published his Book of WHY: The New Science of Cause and Effect last year to suggest that the future of AI depends on building systems with notions of causality. It may seem obvious that intelligent machines would grok cause and effect — after all they’re driving autonomous vehicles. But while machine learning methods excel in describing the real world, they’re often lacking in understanding the world — simple perturbations hardly noticed by humans can cause state-of-art deep learning systems to misclassify road signs.

The formal modeling and logic to support seemingly fundamental causal reasoning has been lacking in data science and AI, a need Pearl is adamantly advocating for. His recent writings have motivated much work (and debate) in the community of AI researchers, mainly with the characterization of deep learning as merely “curve-fitting”.

That is, deep learning, and most machine learning (ML) methods for that matter, learn patterns or associations from data. On its own, observational data can only possibly convey associations between variables — the familiar adage correlation does not imply causation. Sure there may be causal signatures hidden within the data, but they are ambiguous, and more often than not corrupted by missing variables and observations, noise and bias, making it non-trivial to precisely identify cause and effect. ML systems excel in learning connections between input data and output predictions, but lack in reasoning about cause-effect relations or environment changes. These and similar critiques of today’s ML toolbox are not a matter of speculation or personal opinion, but rather grounded truths in the underlying mathematics.

As an example of the intrinsic limitations of data-centric systems that are not guided by explicit models of reality, consider a risk-estimation model for those hospitalized with pneumonia. From the data, the model learned that asthmatics are less likely to die from pneumonia. Counterintuitive? Indeed. The researchers traced the strange result back to an existing policy underlying the observed data: asthmatics with pneumonia were directly admitted to the intensive care unit (ICU), therefore receiving more aggressive treatment, and thus less likely to die than patients not given the same attention.

In the field of causal inference, Professor Judea Pearl is a pioneer for developing a theory of causal and counterfactual inference based on structural models. In 2011, Professor Pearl also received the Turing award from Association for Computing Machinery (ACM), which is the highest distinction in computer science, “for fundamental contributions to artificial intelligence through the development of a calculus for probabilistic and causal reasoning”. In 2020, Professor Pearl is also awarded as World Leader in AI World Society (AIWS.net) by Michael Dukakis Institute for Leadership and Innovation (MDI) and Boston Global Forum (BGF).

The original article can be found here.