No. 244 of 500 No. 42 in Machine Learning

Elements of Causal Inference

Foundations and Learning Algorithms

By Jonas Peters, Dominik Janzing and Bernhard Schölkopf

Print length
288 pages
Publisher
The MIT Press
Reading level
Advanced
ISBN-13
978-0262037310
Published
November 29, 2017
Written for
Graduate students, ML researchers and data scientists who want a formal grounding in causal models.

Synopsis

Jonas Peters, Dominik Janzing and Bernhard Schölkopf wrote this book to give machine learning and statistics readers a self-contained route into causal reasoning. They begin by explaining why prediction from observational data cannot answer questions about interventions, then introduce structural causal models as the central formalism. The book treats the two-variable case first, where classic conditional-independence methods fail, and shows how asymmetries between cause and effect, such as additive noise and the independence of mechanisms, can still reveal causal direction. It then extends the same ideas to multivariate models, covering graphs, interventions, counterfactuals, identifiability and algorithms for learning causal structure from observational and interventional data.

Later chapters connect causality to standard machine learning problems, including semi-supervised learning, covariate shift, domain adaptation and robustness across changing environments, and address hidden variables and time series. Exercises, code snippets and an appendix summarizing key probability and statistics concepts support classroom use. The writing is concise and mathematical, so readers gain precise definitions and a research-level view of where causal ideas improve generalization. Published in MIT Press's Adaptive Computation and Machine Learning series in 2017, the book is open access, and its focus on invariance and distribution shift anticipates problems that still challenge deployed AI systems.

Scorecard

Editorial score

How it scored

Editorial score: 75.0 / 1005 weighted criteria

  • Influence30%

    23.4/30
  • Authority25%

    23.3/25
  • Practical value20%

    12.0/20
  • Currency15%

    11.3/15
  • Readability10%

    5.0/10
  • Editorial score

    75.0/100

How the rank was computed

Rank No. 244 of 500 comes from a composite of 55.2: observable signals 21.4 of 55 points plus editorial review 33.8 of 45 points.

All 11 inputs
InputPoints
Observable signals21.4/55
Reach (Amazon ratings volume)6.6/16.5
Consensus across best-of lists2.8/13.8
Awards and bestseller evidence5.0/8.3
Reader rating, sample-adjusted6.4/8.3
Amazon sales rank0.6/5.5
Recency0.0/2.8
Editorial review33.8/45
Author authority10.5/11.3
Influence10.5/13.5
Practical value5.4/9.0
Readability2.3/4.5
Currency5.1/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A rigorous, compact bridge between causal modeling and machine learning, best suited to graduate students and researchers comfortable with probability.

Strengths

  • Free open-access edition from MIT Press
  • Distinctive treatment of two-variable causal discovery
  • Explicit links to machine learning problems

Keep in mind

Dense, mathematical exposition assumes graduate-level probability; practitioners seeking applied, code-first tools for A/B testing or econometric causal analysis will find little hands-on guidance.

Key takeaways

  1. Predicting outcomes from observational data differs from predicting the effects of interventions, which requires an explicit causal model.
  2. With only two variables, conditional independence tests cannot reveal causal direction, but assumptions such as additive noise or independent mechanisms sometimes can.
  3. Structural causal models unify observational distributions, intervention distributions and counterfactuals in one framework, with graphs encoding the assumed causal structure.
  4. Causal knowledge can improve machine learning under distribution shift, semi-supervised learning and transfer, because causal mechanisms tend to stay invariant across environments.

Notable facts

  • MIT Press released the book as an open-access title that can be downloaded free, as Jonas Peters notes in his research summary. Source (opens in a new tab)
  • Published in 2017 in MIT Press's Adaptive Computation and Machine Learning series, at 288 pages. Source (opens in a new tab)

About the author

Jonas Peters, Dominik Janzing and Bernhard Schölkopf

Jonas Peters is professor of statistics at ETH Zurich, previously professor at the University of Copenhagen and leader of the causality group at the Max Planck Institute for Intelligent Systems. Co-author Bernhard Schölkopf directs that institute's Empirical Inference department and is known for kernel methods; Dominik Janzing is a principal research scientist at Amazon Web Services working on causal inference.

Questions

What is Elements of Causal Inference about?
Elements of Causal Inference is a concise graduate-level introduction to causal models written from a machine learning perspective. Peters, Janzing and Schölkopf explain structural causal models, interventions and counterfactuals, show how causal structure can be learned from data, including the difficult two-variable case, and connect causal ideas to machine learning problems such as semi-supervised learning and distribution shift.
Is Elements of Causal Inference free to read?
Yes. MIT Press published the book in 2017 as an open-access title, and Jonas Peters notes that the PDF can be downloaded free from the MIT Press website; a copy is also hosted by the OAPEN Library under a Creative Commons noncommercial license. Print copies remain available for readers who prefer a physical edition.
Is Elements of Causal Inference worth reading in 2026?
For readers with solid probability and statistics, yes. Its treatment of invariance, cause-effect asymmetries and distribution shift bears directly on current questions about why AI models fail outside their training data. The book received the American Statistical Association's 2018 Causality in Statistics Education Award. Newcomers wanting applied, code-first causal inference should start with a gentler text and return to this one afterward.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.