No. 93 of 500 No. 6 in Machine Learning

Why Machines Learn

The Elegant Math Behind Modern AI

By Anil Ananthaswamy

Published
July 16, 2024
Print length
480 pages
Publisher
Dutton
Reading level
Intermediate
ISBN-13
978-0593185742
Written for
Curious professionals, students and engineers who want to understand the math behind machine learning.

Synopsis

Science writer Anil Ananthaswamy, a former software engineer, argues that anyone who wants to judge what AI can and cannot do should understand the mathematics underneath it, and he sets out to teach that mathematics to motivated nonspecialists. The book moves roughly chronologically, from the McCulloch-Pitts neuron and Frank Rosenblatt's perceptron through Bayesian reasoning, nearest-neighbor methods, principal component analysis, support vector machines and kernel methods, Hopfield networks, backpropagation, convolutional networks and the puzzles of overparameterized deep learning. Each idea arrives with its inventors, the problem that motivated them, and the vectors, probability, calculus and linear algebra it requires.

Unlike most popular science books, this one includes equations, and Ananthaswamy works through them step by step, recapping often so readers without recent math training can follow the argument. Physics World's reviewer noted that the demands are real while praising the gradual build-up of understanding. The payoff is a working grasp of why learning algorithms succeed, the origins of their assumptions, and the distance that still separates them from human intelligence. The 2025 paperback adds an afterword on the transformer architecture, but the core of the book concerns the foundations beneath large language models rather than the models themselves.

Scorecard

Editorial score

How it scored

Editorial score: 74.8 / 1005 weighted criteria

  • Influence30%

    22.5/30
  • Authority25%

    18.8/25
  • Practical value20%

    14.0/20
  • Currency15%

    12.0/15
  • Readability10%

    7.5/10
  • Editorial score

    74.8/100

How the rank was computed

Rank No. 93 of 500 comes from a composite of 63.5: observable signals 29.9 of 55 points plus editorial review 33.6 of 45 points.

All 11 inputs
InputPoints
Observable signals29.9/55
Reach (Amazon ratings volume)10.8/16.5
Consensus across best-of lists8.3/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted7.4/8.3
Amazon sales rank1.7/5.5
Recency1.7/2.8
Editorial review33.6/45
Author authority8.4/11.3
Influence10.1/13.5
Practical value6.3/9.0
Readability3.4/4.5
Currency5.4/6.8

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 rare bridge between popular AI writing and textbooks that teaches the actual mathematics of machine learning through its history.

Strengths

  • Teaches real equations with patient recaps
  • Weaves history and personalities into math
  • Clear historical context for each idea

Keep in mind

Readers without high school algebra and some calculus may find the mathematical chapters hard going, and coverage of large language models is brief.

Key takeaways

  1. Much of machine learning rests on centuries-old mathematics, including linear algebra, calculus and probability, applied at a scale that modern hardware made possible.
  2. Representing data as vectors in high-dimensional space lets algorithms find patterns geometrically, which is why techniques like principal component analysis and kernel methods matter.
  3. Backpropagation and gradient descent let multilayer networks learn from errors, overcoming the limits that stalled single-layer perceptrons in the late 1960s.
  4. Deep networks generalize well even with far more parameters than classical statistical theory considers safe, a puzzle still open at the research frontier.

Notable facts

About the author

Anil Ananthaswamy

Anil Ananthaswamy is a science writer and former staff writer and deputy news editor at New Scientist who writes for Quanta, Scientific American and Nature. Trained in electronics and computer engineering at IIT Madras and the University of Washington, he worked as a software engineer before journalism. He was a 2019-20 MIT Knight Science Journalism Fellow and wrote The Edge of Physics and Through Two Doors at Once.

Questions

How much math do you need for Why Machines Learn?
Comfort with high school algebra helps, and some exposure to calculus and probability helps more. Ananthaswamy introduces vectors, matrices, derivatives and probability as needed and recaps often, so determined readers can follow without a degree in math. Readers who skip the equations still get the history and intuition, though they miss part of the book's purpose.
Does Why Machines Learn cover ChatGPT and large language models?
Only briefly. The book concentrates on the foundations beneath large language models: perceptrons, probability, kernel methods, backpropagation and deep networks. The 2025 paperback adds an afterword on the transformer architecture behind ChatGPT. Readers who want a detailed treatment of LLMs should follow it with a dedicated book, but this one supplies the groundwork that makes such books easier to follow.
Is Why Machines Learn worth reading in 2026?
Yes, for readers willing to engage with equations. The mathematics it teaches has not changed, and understanding it makes claims about AI capabilities easier to evaluate. Its 2024 publication date and the 2025 afterword on transformers keep it closer to the current field than most explanatory books, and Geoffrey Hinton and Melanie Mitchell are among the researchers who have endorsed it.

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.