Prediction Machines: The Simple Economics of Artificial Intelligence — A Prophecy Written in 2018 for 2026

Having just finished reading Prediction Machines: The Simple Economics of Artificial Intelligence, my immediate reaction was: This feels like a time-travel novel!

If you open this book today in 2026, you will find that the various phenomena, industrial transformations, and societal shifts it discusses are almost exactly what we are experiencing as our daily reality. But what shocks me most is that this book was actually published back in 2018!

The authors, Ajay Agrawal and his colleagues—all top-tier economists—succeeded because they were not hijacked by the technological details of the time (such as basic image recognition). Instead, they analyzed AI using fundamental economic logic.

1. The Essence of AI is the Ultimate Reduction in the Cost of Prediction¶

The book presents a very clear economic chain:
Data -> Prediction -> Judgment -> Action & Strategy

What exactly is AI? Viewed through an economic lens, it is the dramatic cheapening of "prediction."

• When prediction becomes cheap, it acts as a "complement" to enhance the value of human "judgment": Much like X-rays or image recognition, AI identifies anomalies (prediction), and hands them over to doctors to make the final diagnosis and care decisions (judgment). Instead of being replaced, doctors become a hundred times more efficient.
• When prediction becomes accurate enough, it acts as a "substitute" that packages the entire system: Take autonomous driving and drones, for example. Once the costs of road prediction and steering execution are driven to zero, traditional occupations are directly replaced.

This is much like how automobiles replaced horse-drawn carriages. Technology penetrates society layer by layer—evolving from a minor tool into a new strategic capability (like unmanned transport/combat vehicles), and ultimately reshaping societal institutions. Faced with this transition, those who remain passive will truly be left behind.

2. Spot-on Insights on the State-Level AI Race and the Balance of Trade-offs¶

As early as 2018, the authors used economic trade-offs to dissect several core propositions of AI development:

• The Trade-off between Performance and Privacy: AI feeds on data. If a nation can easily access citizen data for training, it will sacrifice privacy but yield more accurate predictions. Conversely, regions that strictly protect privacy face higher data acquisition costs.
• Economies of Scale and Natural Monopoly: The classic "data flywheel"—more data -> better predictions -> more users -> more data. Once scale is achieved, lower costs will lead to monopolies, which are a double-edged sword for markets and society alike.
• Redistribution and Export of Productivity: AI is essentially a redistribution of productivity. Just as China once exported cheap steel and cars, it can export cheap AI applications and technical ecosystems to stimulate competition. While this drives efficiency globally, it also causes productivity outflows in competing nations.

Looking back at the US-China competition, the evolution of open-source vs. closed-source models, and the balancing act of "privacy vs. performance" and "data scale vs. compute" from the vantage point of 2026, the authors' predictions have proven entirely correct.

Conclusion: Why is this book still worth reading today?¶

Many books on technology become obsolete and useless within a couple of years because of rapid technological changes. However, Prediction Machines is anchored in economic thinking chains. Whether it's the early days of image recognition, or the subsequent explosion of Large Language Models (LLMs) and generative AI, the framework of "prediction, judgment, complements, and substitutes" perfectly explains it all.

If you want to clearly understand your career positioning, industry trends, and even national-level competitive dynamics in the current AI wave, this is absolutely a must-read book to build your underlying logic.