Artificial Intelligence: A Guide for Thinking Humans

The book opens by challenging the hype surrounding artificial intelligence, arguing that many claims about AI—especially about deep learning—are exaggerated. Mitchell, drawing on her experience as a researcher in analogical reasoning and complex systems, shows that today's AI systems are brittle, data-hungry, and lack the robust understanding that even a child possesses. She uses examples like AlphaGo and GPT-3 to illustrate both achievements and failures, concluding that we are far from achieving artificial general intelligence.

The book is structured in three parts: a history of AI, a deep dive into modern techniques (neural networks, reinforcement learning, and natural language processing), and a critical look at the societal and philosophical implications. Mitchell explains technical concepts clearly, including backpropagation, convolutional neural networks, and the Turing Test.

Upon release, the book was widely praised for its clarity and balance. Nature called it “a much-needed reality check”, while Science noted its “lucid explanations”. It has been recommended by Bill Gates and featured in lists of best popular science books. The book has been translated into multiple languages and is frequently used in undergraduate courses on AI.

Mitchell includes personal anecdotes from her time at Santa Fe Institute working with Douglas Hofstadter, whose influence is evident in the book's emphasis on analogy and perception. The title itself is a nod to Hofstadter's Gödel, Escher, Bach. A lesser-known fact: Mitchell originally planned to write the book as a collaboration with Hofstadter, but the project evolved into a solo work. The book also contains a detailed critique of Noam Chomsky's views on AI, a debate rarely covered in popular AI books.

Deep learning

A subset of machine learning using multi-layered neural networks to model complex patterns in data.

Reinforcement learning

A training method where an agent learns to make decisions by receiving rewards or penalties.

Artificial general intelligence

A hypothetical AI that can perform any intellectual task a human being can.

Analogical reasoning

The cognitive process of transferring knowledge from one situation to another based on similarity.