Further Reading — Chapter 3: How Machines Learn

Tier 1: Accessible and Essential

Domingos, Pedro. The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. Basic Books, 2015. A sweeping tour of the five major "tribes" of machine learning — each with a different philosophy about how machines should learn. Domingos writes accessibly and makes the intellectual stakes of different machine learning approaches vivid. An excellent companion to this chapter's overview of supervised, unsupervised, and reinforcement learning.

Mitchell, Tom. "The Discipline of Machine Learning." Carnegie Mellon University Machine Learning Department Technical Report, 2006. A brief, clear essay by one of the field's founders explaining what machine learning is and why it matters. At just a few pages, it's an efficient way to see how a leading researcher defines the field. Freely available online.

Alpaydin, Ethem. Machine Learning: The New AI. MIT Press Essential Knowledge Series, 2016. A compact, non-technical introduction to machine learning concepts. At under 200 pages, it covers supervised learning, unsupervised learning, reinforcement learning, and neural networks with minimal jargon. The Essential Knowledge series is designed for general readers.

3Blue1Brown. "But What Is a Neural Network?" YouTube, 2017. A visual, intuitive explanation of neural networks using animation. If the neural network section of this chapter left you wanting a more visual presentation, this video series is outstanding. The creator, Grant Sanderson, has a gift for making mathematical concepts visually intuitive without dumbing them down.

Tier 2: Deeper Exploration

Burkov, Andriy. The Hundred-Page Machine Learning Book. Self-published, 2019. Exactly what the title promises — a concise overview of machine learning that manages to be both rigorous and brief. Some light math is used but always explained clearly. Excellent for readers who want more technical depth without committing to a full textbook.

Chollet, Francois. "What Worries Me About AI." Medium, 2018. A thoughtful essay by the creator of the Keras deep learning library, examining what deep learning can and cannot do. Chollet's arguments about the gap between pattern recognition and genuine intelligence are directly relevant to this chapter's threshold concept.

Sutton, Richard, and Andrew Barto. Reinforcement Learning: An Introduction. 2nd Edition. MIT Press, 2018. The definitive textbook on reinforcement learning, written by two of the field's pioneers. While it contains mathematical formalism, the introductory chapters and examples are accessible and provide much deeper coverage of reinforcement learning than this chapter could offer. Freely available online from the authors.

O'Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown, 2016. While focused on the social consequences of machine learning rather than the technical details, O'Neil's book provides numerous real-world examples of how machine learning systems can overfit to biased data, optimize for the wrong metrics, and cause harm. Her examples of opaque models making consequential decisions about people's lives are a powerful complement to this chapter's technical framework.

Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016. The standard textbook on deep learning. The first few chapters provide an accessible (if mathematically flavored) introduction to machine learning foundations. For readers who want to understand neural networks at a deeper technical level, chapters 6–9 are authoritative. Freely available online.

Strogatz, Steven. "One Giant Step for a Chess-Playing Machine." The New York Times, December 2018. A mathematician's elegant explanation of how AlphaZero (the successor to AlphaGo) learned to play chess through reinforcement learning, developing strategies that surprised human grandmasters. Accessible, well-written, and captures the genuine wonder of reinforcement learning without overhyping it.

Sculley, D., et al. "Hidden Technical Debt in Machine Learning Systems." Advances in Neural Information Processing Systems, 2015. A landmark paper from Google researchers about the practical challenges of deploying and maintaining machine learning systems in production. While somewhat technical, its core insights — that model training is a small fraction of the work, and that real-world ML systems accumulate "technical debt" rapidly — are accessible and directly relevant to the gap between demonstration and deployment.