Human Intelligence,
Amplified

In 2008, Google announced something extraordinary. Their engineers had built a system that could predict flu outbreaks faster than the Centers for Disease Control — sometimes weeks faster.

The system, called Google Flu Trends, worked by tracking what people searched for. When searches for “flu symptoms” or “how long does the flu last” spiked in a region, Google could predict that flu cases would soon spike too.

It seemed like magic. It also seemed like the future: a world where massive datasets could reveal patterns invisible to traditional methods.

There was just one problem. Before I tell you what went wrong, take a moment. If you were building a system that predicts flu from search queries, what could go wrong?

By 2013, Google Flu Trends was overshooting actual flu rates by nearly 100 percent. The system had found real patterns — but those patterns reflected search behavior, not flu biology.

The Google Flu Trends story illustrates a truth this book keeps returning to: data and thinking are complements, not substitutes. More data doesn't replace the need for better questions — it amplifies that need.

Why do smart people make these mistakes? Part of the answer is that data feels certain. When someone says “the data shows…” it sounds authoritative, scientific, final.

But data never speaks for itself. Data only speaks through the questions we ask of it — and those questions reflect our assumptions, our blind spots, what we chose to measure and what we didn't.