Why Machines Learn: The Elegant Math Behind Modern AI

This impenetrable primer from science writer Ananthaswamy (Through Two Doors at Once) unsuccessfully attempts to elucidate how AI works. He explains that it learns by scanning data for patterns and then makes predictions about what kinds of data are likely to appear in sequence. Unfortunately, the excruciatingly detailed breakdown of the roles played by probability, principal component analysis (“projecting high-dimensional data onto a much smaller number of axes to find the dimensions along which the data vary the most”), and eigenvectors (which are never satisfactorily defined) will sail over the heads of anyone without an advanced math degree. Biographical background on physicist John Hopfield, electrical engineer Bernhard Boser, and other pioneering contributors to machine learning does little to alleviate the labyrinthine discussions of their advances. There are some bright spots—as when Ananthaswamy discusses how statisticians deduced the authorship of the contested Federalist Papers by analyzing whether the writing more closely reflected the vocabulary of James Madison or Alexander Hamilton—but these highlights are few and far between, surrounded by bewildering equations and dense proofs for mathematical theorems. General readers will struggle to follow this. Agent: Peter Tallack, Curious Minds Agency. (July)