Artificial Intelligence
by Melanie Mitchell
Santa Fe Institute professor Melanie Mitchell's measured critique of AI hype, exploring why machines excel at chess but mistake buses for ostriches
"Today's AI is far from general intelligence, and I don't believe that machine superintelligence is anywhere on the horizon".
Editorial Summary
Santa Fe Institute professor Melanie Mitchell delivers a sobering assessment of artificial intelligence's true capabilities in this 2019 examination of the field's turbulent history and overhyped present. Mitchell, who studied under cognitive scientist Douglas Hofstadter, argues that people consistently overestimate AI's abilities while underestimating the complexity of human intelligence. The book explores how modern systems like Watson and deep learning networks excel at narrow tasks but catastrophically fail when encountering situations outside their training data, making bizarre errors like confusing school buses with ostriches. Mitchell examines the disconnect between spectacular advances in computer vision and translation versus AI's complete lack of common sense reasoning, drawing on her meetings with Google engineers and conversations with Hofstadter, who fears AI might reduce human creativity to mere "bag of tricks."
Perspective
"Mitchell writes as someone who genuinely loves the field and is therefore credible when she says we are fooling ourselves — the result is the most honest assessment of where AI actually stands, stripped of both fear and hype. Her distinctive contribution is documenting the specific, embarrassing failure modes of current systems: not abstract alignment concerns but concrete cases where state-of-the-art AI makes mistakes a five-year-old wouldn't. Anyone trying to calibrate their sense of how capable current AI systems really are — before making decisions based on that calibration — should read this first."
Matched by concept and theme
