The Four Waves of AI

Definition

Deep learning does not arrive as one undifferentiated force; it washes over the economy in four waves, each defined by which kind of data it exploits: internet AI (mining clicks and behavior to recommend content), business AI (mining existing corporate records for hidden correlations), perception AI (digitizing the physical world through vision and voice), and autonomous AI (fusing perception with the ability to act — cars, drones, robots). Each wave requires progressively harder-to-obtain data, so they unlock roughly in that order.

In the Book

Lee grounds internet AI in Jinri Toutiao, whose recommendation algorithms serve as "AI editors" that curate and even write news, growing to a $30 billion valuation on the back of automatically-labeled clicks and reading time. Business AI is illustrated through Palantir- and 4th Paradigm-style firms mining banks' and hospitals' decades of structured records for "weak features" no human analyst would notice. Perception AI and autonomous AI are previewed through speech and image recognition firms like iFlyTek, which built a near-perfect voice clone of Donald Trump speaking Mandarin. Lee uses the framework to score US-China competition wave by wave: China favored in internet and perception AI (more users, less privacy friction, denser data), the US still ahead in business AI (more structured enterprise data), and autonomous AI still open.

Why It Matters

The framework replaces "is AI here yet?" with a sharper question: which specific kind of data does this application need, and has a mechanism emerged to collect and label it? That turns AI adoption forecasting into a data-supply-chain question rather than a capability question — useful anywhere a general technology's real-world rollout depends on which inputs become available first, not just on what the technology can theoretically do.