The seven tools of causal inference, with reflections on machine learning
By Judea Pearl. 2019
660 citations recorded by the source corpus.
The kind of causal inference seen in natural human thought can be "algorithmitized" to help produce human-level machine intelligence.
Subfields: Artificial Intelligence, Causal Inference & Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yoshua Bengio
- Sergey Levine
- Mohit Bansal
- Percy Liang
- Jianfeng Gao
- Quoc V. Le
- Tat-seng Chua
- Dacheng Tao
In Causal Inference & Discovery
- Yoshua Bengio
- Simon Lacoste-Julien
- Alexandre Lacoste
- Rémi Le Priol
- Pau Rodr'iguez L'opez
- Sébastien Lachapelle
- José Miguel Hernández-Lobato
- Fernando Martínez-Plumed
Citations in the corpus
- Disentanglement via Mechanism Sparsity Regularization: A New Principle for Nonlinear ICA (2021, background): “It has been proposed that causal reasoning will be central to move modern machine learning algorithms beyond their current shortcomings, such as their lack of robustness, transferability and interpretability (Pearl, 2019; Schölkopf, 2019; Schölkopf et al., 2021; Goyal and Bengio, 2021).”
- Gradient-Based Neural DAG Learning (2019, background): “Causal graphical models (CGM) [12] are BNs which support interventional queries like: What will happen if someone external to the system intervene on variable X? Recent work suggests that causality could partially solve challenges faced by current machine learning systems such as robustness to out-o”
- On the Binding Problem in Artificial Neural Networks (2020, background): “…highly relevant for composition, due to their robustness under intervention and utility for reasoning about hypothetical or unobserved scenarios (Pearl, 2019; Schölkopf, 2019). of several layers that each produce an updated set of node representations by applying graph-convolutions to a local…”
- Getting a CLUE: A Method for Explaining Uncertainty Estimates (2020, background): “1) Those in causal inference make causal assumptions about interdependencies among variables and use these assumptions to incorporate consequential adjustments when particular variables are set to new values (Kusner et al., 2017; Pearl, 2019).”
- CRISP-DM Twenty Years Later: From Data Mining Processes to Data Science Trajectories (2021, methodology): “Causal inference from data focuses on answering questions of the type “what if” and relies on methods that incorporate causal knowledge (such as the Structural Causal Models [46], the Potential Outcomes Framework [47] or the Linear non-Gaussian acyclic models [48]).”
- Deep Structural Causal Models for Tractable Counterfactual Inference (2020, background): “Second, we compare our work to recent progress in deep causal learning in light of Pearl’s ladder of causation [19].”
- Neurosymbolic AI: the 3rd wave (2020, background): “In Pearl (2019), Pearl proposes a hierarchy consisting of three levels: association, intervention and counterfactual reasoning, and claims that ML is only capable of achieving association.”
- Tell me why! Explanations support learning relational and causal structure (2021, methodology): “The ability of deep learning to infer causality is sometimes questioned (e.g. Pearl, 2019), but while theoretical limitations hold for passive learners, RL agents can intervene and can therefore identify causal structure.”
- Machine learning and deep learning—A review for ecologists (2022, positive): “Moreover, a pervasive concern is that ML models are trained for prediction, but the best predictive model does not necessarily correspond to the causal model (Breiman, 2001b; Pearl, 2019 , 2021; Box 4).”
- Tell me why! - Explanations support learning of relational and causal structure (2021, background): “The ability of deep learning to learn causal structure is sometimes questioned (e.g. Pearl, 2019), but while theoretical limitations hold for passive learners, RL agents can intervene and can therefore learn causal structure.”
- Generative causal explanations of black-box classifiers (2020, background): “Because explanation methods seek to answer “why” and “how” questions that use the language of cause and effect [3, 4], causal reasoning has played an increasingly important role in designing explanation frameworks [5].”
- Omnidirectional Transfer for Quasilinear Lifelong Learning (2021, background): “Pearl [19] introduced the transfer benefit ratio, which builds directly off relative efficiency from classical statistics [22].”
- Learning to Imagine: Integrating Counterfactual Thinking in Neural Discrete Reasoning (2022, background): “Taking the hypothetical question in Figure 1 as an example, an ideal L2I should recognize the target variable ( finished goods in 2019 ), identify the corresponding fact ( $133,682 ), and replace the fact with the assumed value ( $132,935 ).”
- Tackling Climate Change with Machine Learning (2019, background): “ML can play a role in causal inference [760–762], including in the context of policy problems [763, 764] and in climate-relevant scenarios such as estimating the effects of temperature on human mortality [765] and the effects of World Bank projects on vegetative cover [766].”
- Explainable AI Methods - A Brief Overview (2020, methodology): “Therefore, it will be important to measure the quality of explanations so that an xAI method achieves a certain level of causal understanding [65] for a user with effectiveness, efficiency and satisfaction in a given context of use”
- A general approach to progressive learning (2020, background): “Pearl [17] introduced the transfer benefit ratio (TBR), which builds directly off relative efficiency from classical statistics [24].”