Theoretical Impediments to Machine Learning With Seven Sparks from the Causal Revolution

By Judea Pearl. 2018

367 citations recorded by the source corpus.

Current machine learning systems operate, almost exclusively, in a statistical, or model-blind mode, which entails severe theoretical limits on their power and performance. Such systems cannot reason about interventions and retrospection and, therefore, cannot serve as the basis for strong AI. To achieve human level intelligence, learning machines need the guidance of a model of reality, similar to the ones used in causal inference. To demonstrate the essential role of such models, I will present a summary of seven tasks which are beyond reach of current machine learning systems and which have been accomplished using the tools of causal inference.

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

In Causal Inference & Discovery

Citations in the corpus

Improving the accuracy of medical diagnosis with causal machine learning (2020, methodology): “As noted by Pearl, associative inference is the simplest in a hierarchy of possible inference schemes 24 – 26 .”
Learning First-Order Symbolic Representations for Planning from the Structure of the State Space (2019, background): “Learners, and in particular deep learners, have achieved considerable success but result in black boxes that do not have the flexibility, transparency, and generality of their model-based counterparts [26, 27, 32, 12, 17].”
Generalization in anti-causal learning (2018, background): “The study of causality and its role in statistics and machine learning has a long history and recently causal reasoning has been claimed vital for solving future challenges in AI (Lake et al., 2016; Pearl, 2018).”
Relational inductive biases, deep learning, and graph networks (2018, background): “Despite deep learning’s successes, however, important critiques (Shalev-Shwartz et al., 2017; Lake et al., 2017; Lake and Baroni, 2018; Marcus, 2018; Pearl, 2018; Yuille and Liu, 2018) have highlighted key challenges it faces in complex language and scene understanding, reasoning about structured da”
Causal Interpretability for Machine Learning - Problems, Methods and Evaluation (2020, background): “nterpretable model can help us understand the real causes of decisions made by machine learning algorithms, improve their performance, and prevent them from failing in unexpected circumstances. Pearl [83] introduces diﬀerent levels of said interpretability and argues that generating counterfactual e”
COPHY: Counterfactual Learning of Physical Dynamics (2019, background): “…mainstream attention relatively recently (Lopez-Paz et al., 2017; Lopez-Paz & Oquab, 2017; Kocaoglu et al., 2018; Rojas-Carulla et al., 2018; Mooij et al., 2016; Sch¨olkopf et al., 2012), due to limitations of statistical learning becoming increasingly apparent (Pearl, 2018; Lake et al., 2017).”
A Review of Challenges and Opportunities in Machine Learning for Health. (2018, background): “The shortcomings of classical statistical learning when answering causal questions are discussed in greater detail in prior work 16 .”
Tell me why! - Explanations support learning of relational and causal structure (2021, background): “…and analogies, and generalization—are often cited as areas where deep learning has not yet achieved human-level performance (e.g. Fodor and Pylyshyn, 1988; Lake et al., 2017; Pearl, 2018; Marcus, 2020; Ichien et al., 2021; Holyoak and Lu, 2021; Puebla and Bowers, 2021; Geirhos et al., 2020).”
Causability and explainability of artificial intelligence in medicine (2019, methodology): “The main problem of the most successful current ML systems, recently emphasized by Pearl (2018), is that they work on a statistical, or model-free mode, which entails severe limitations on their performance.”
The debate over understanding in AI’s large language models (2022, background): “It has been argued that understanding of this kind may enable abilities not possible for purely statistical models: learning new concepts from limited examples, being able to generalize that knowledge to novel instances of those concepts, being able to flexibly transfer one’s knowledge to new situat”
A Survey of Learning Causality with Data: Problems and Methods (2018, methodology): “Ignorability plays a core role in the potential outcome framework as it directly leads to causal identification [102].”
FutureMapping: The Computational Structure of Spatial AI Systems (2018, background): “Judea Pearl, recently discussing efﬁcient situated learning and the need to reason about causation [43], argues that “what humans possessed that other species lacked was a mental representation, a blue-print of their environment which they could manipulate at will to imagine alternative…”
Tell me why! Explanations support learning relational and causal structure (2021, background): “It is often argued that machine learning models—and deep learning models in particular—lack the human proficiencies for forming abstractions and inferring relational or causal structure (e.g. Fodor & Pylyshyn, 1988; Lake et al., 2017; Pearl, 2018; Marcus, 2020; Ichien et al., 2021; Holyoak & Lu, 202”
Model-free, Model-based, and General Intelligence (2018, methodology): “Explanation and accountability also require models: we build explanations and make predictions using models [Darwiche, 2017; Lake et al., 2017; Pearl, 2018].”
Causal Learning and Explanation of Deep Neural Networks via Autoencoded Activations (2018, background): “Causality has a long history in AI and numerous ML efforts have focused on building realistic and accurate causal models of the world instead of statistical models [Pearl, 2018].”
MultiVerse: Causal Reasoning using Importance Sampling in Probabilistic Programming (2019, background): “Machine learning has renewed interest in causal tools to aid reasoning (Pearl, 2018).”
Principles and Practice of Explainable Machine Learning (2020, background): “Finally, as XAI matures, notions of causal analysis should be incorporated to new approaches [59, 65].”
A Survey of Learning Causality with Data (2018, background): “The ability to learn causality is considered as a significant component of human-level intelligence and can serve as the foundation of AI [114].”
CAUSAL INTERPRETATIONS OF BLACK-BOX MODELS (2019, neutral): “In comparison, despite some good efforts to introduce causal inference concepts to the machine learning community (Spirtes 2010; Peters, Janzing, and Sch¨olkopf 2017; Pearl 2018), the learning curve remains steep for researchers who are used to building black-box predictive models.”