The Book of Why

By Robert L. Shook. 1983

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning stable and predictive structures in kinetic systems (2018, background): “This is why causal models are expected to work more reliably when predicting under distributional shifts (26–28).”
Deconfounding Reinforcement Learning in Observational Settings (2018, background): “The term “confounding” originally meant “mixing” in English, which describes that the true causal eﬀect T → R is “mixed” with the spurious correlation between T and R induced by the fork T ← Z → R (Pearl and Mackenzie, 2018).”
Deconfounded Image Captioning: A Causal Retrospect (2020, background): “0 as well as pointing us to a new future.”
TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second (2022, methodology): “Predictions based on ideas from causal reasoning Previous works have applied causal reasoning to predict observations on unseen data by using causal inference, a method which seeks to identify causal relations between the components of a system by the use of interventions and observational data (Pea”
On Pearl’s Hierarchy and the Foundations of Causal Inference (2022, methodology): “Each SCM naturally defines a qualitative hierarchy of concepts, described as the “ladder of causation” in Pearl and Mackenzie [2018], which we have been call ing the PCH (Figure 27.1).”
Perception Test: A Diagnostic Benchmark for Multimodal Video Models (2023, background): “Flamingo struggles the most on counterfactual questions, where its accuracy is below random, pointing to an important ability that our models clearly lack [43].”
D’ya Like DAGs? A Survey on Structure Learning and Causal Discovery (2021, methodology): “The ultimate benefit of the graphical and structural model frameworks is that they, at least in principle and under some strong assumptions, enable us to use observational data to answer scientific questions such as ‘how?’”
Explainable AI Methods - A Brief Overview (2020, background): “It can be seen as a new name for a very old quest in science to help to provide answers to questions of why [66].”
Connecting the Dots in Trustworthy Artificial Intelligence: From AI Principles, Ethics, and Key Requirements to Responsible AI Systems and Regulation (2023, background): “Causality can help debugging algorithmic bias mitigation or explaining models [87], e.”
AdaRL: What, Where, and How to Adapt in Transfer Reinforcement Learning (2021, background): “In contrast, humans are usually good at transferring acquired knowledge to new environments and tasks both efﬁciently and effectively (Pearl and Mackenzie, 2018), thanks to the ability to understand the environments.”
Causability and explainability of artificial intelligence in medicine (2019, methodology): “A very important direction is research towards structural causal models (Pearl, 2009; Pearl & Mackenzie, 2018; Peters et al., 2017).”