The Last Human Internet

The Last Human Internet argues that for roughly two decades, artificial intelligence benefited from an extraordinary historical accident: human beings digitized civilization without doing so to train machines, creating an accidental archive of books, forums, photographs, code, and reviews. But 2022 marked a turning point when generative AI became widely accessible, causing the Internet to increasingly produce its own content and creating a ground-truth bottleneck. The book contends that the next major constraint on AI will not be compute or parameters, but access to observations originating outside the model ecosystem—cameras, robots, laboratories, and sensors that generate causal data through physical action. This transforms AI development from reading the archive to participating in reality itself, where deployed systems produce training episodes no archive could contain, creating a reality premium and a competitive moat based on privileged access to the physical world.