The Age of Implementation

Definition

After a rare foundational breakthrough (deep learning, in AI's case), the economically decisive period is not further scientific discovery but the "nuts and bolts" work of implementation — gathering data, tweaking formulas, prototyping products, iterating business models. Popular attention gravitates to speculative frontier questions (artificial general intelligence, superintelligence) that are actually decades or centuries away, while the mundane engineering phase that is already funded, already staffed, and already happening goes underrated.

In the Book

Lee contrasts the utopian and dystopian camps — Ray Kurzweil's 2045 singularity, Demis Hassabis's superintelligence-solves-everything vision, Elon Musk's "summoning the demon" warnings, Nick Bostrom's control problem — with what he argues is actually underway: the same deep learning technique from Geoffrey Hinton's 2006 breakthrough being implemented across industry after industry, with no comparable scientific leap in the twelve years since. He cites his own miscalibration at Apple in the late 1980s, when he predicted mainstream speech recognition within five years and was off by twenty, as a caution against extrapolating research demonstrations into near-term timelines. His conclusion: AGI remains many decades away, but the ordinary implementation of already-proven deep learning is close enough to cause a real, near-term jobs and inequality crisis regardless.

Why It Matters

This separates two questions that get collapsed in public discourse about any frontier technology: "will there be a bigger breakthrough?" and "what will the current breakthrough actually do once fully deployed?" The second question is answerable and often more consequential, because implementation compounds across an entire economy while the next scientific leap is, by definition, unscheduled. It is a general corrective for reading technological risk and opportunity off the loudest speculative claims rather than off the deployment curve of what already works.