Racing With Machines

Definition

The book's answer to automation anxiety is a reframe: not "race against the machine" (a head-to-head contest humans will eventually lose in any domain where computers surpass them) but "race with machines" — finding the combination of human and machine ability that outperforms either alone. Losing a head-to-head contest in a domain does not mean humans become worthless in that domain; it means the valuable role shifts from competing to complementing.

In the Book

The proof case is freestyle chess. After Deep Blue beat world champion Garry Kasparov in 1997, human-versus-computer chess contests were considered settled. But in freestyle tournaments allowing any combination of human and machine players, teams of human-plus-machine beat the strongest standalone computers, and the 2005 event was won not by a grandmaster with a top-end PC but by two amateur players skilled at "coaching" three ordinary computers at once — described by Kasparov as "weak human + machine + better process" beating "strong human + machine + inferior process." The book draws out that humans contribute strategic guidance and, especially, ideation (generating a genuinely new idea, as Kasparov describes it) where computers remain weak, while machines supply tactical calculation. Chapter 12 extends the frame beyond chess to Amazon's Mechanical Turk (Jeff Bezos's "artificial artificial intelligence," where human judgment plugs the gaps automated systems can't resolve) and to Twitter's search algorithm, where human raters instantly disambiguate context (a "Big Bird" surge meaning a political debate reference, not Sesame Street) faster than the algorithm alone.

Why It Matters

This is a strategic response to displacement risk that doesn't depend on a task staying permanently unautomatable: even after a machine exceeds human performance solo, there can be a durable niche in improving the joint system — supplying the judgment, context, or ideation the machine lacks, or "coaching" the machine's process. It generalizes to any domain where an AI or automated system now outperforms unaided humans: the relevant question shifts from "can I still beat it" to "what does the best paired process look like, and what's my role in it."