We are entering a new epoch. The Digital Turn gave us cheap information and global networks. Now, we are building something new on top of that foundation.
Above the stack of devices, networks, and data, a new layer has emerged: Models. These aren't just static code; they are systems that learn from the data they process.
This changes the nature of capital. It becomes "active." Data flows up, models improve, and decisions flow down—automating credit scores, logistics, and recommendations. Capital is no longer just a tool; it's a system that improves with experience.
To understand AI's economic impact, we must look at what it actually does. Every decision has components: Data, Prediction, Judgment, and Action. Historically, humans did the prediction.
AI drops the cost of prediction. As Agrawal, Gans, and Goldfarb argue, when the cost of an input falls, we use more of it. We start using prediction for problems we never thought of as prediction problems—like driving or translation.
While the cost of prediction plummets, the value of the complements—Judgment and Data—rises. The bottleneck shifts from "knowing what will happen" to "knowing what to do about it."
This forces a redesign of workflows. Some tasks become fully automated. Others become "human-in-the-loop," where AI predicts and humans judge. The boundary between labor and capital is being redrawn in real-time.
Automation doesn't hit everywhere at once. It lands in specific places. Industrial robots, for instance, concentrated in manufacturing hubs.
In regions with high exposure to robots, we see a clear trend: employment growth slows. The machines substitute for routine labor.
Wages follow a similar pattern. In highly exposed areas, wages for low- and medium-skill workers stagnate. The productivity gains don't automatically trickle down locally.
Aggregate statistics hide these local realities. A tech hub might boom while a manufacturing town struggles. Exposure is local, and the pain is concentrated.
The composition of capital has shifted decisively. Tangible assets (machines, buildings) are being overtaken by intangibles. And within intangibles, a new category is exploding: Data & Model Capital.
This new capital is different. It is Scalable, Automated, and Self-Improving. It doesn't just sit there; it acts.
Think of the modern firm as a stack: Data feeds Models, Models drive Applications, Applications generate more Data. It's a feedback loop that creates a "data moat."
This creates a divide. Firms that master this "active capital" pull away from the pack. We see a bifurcation between AI-native firms and traditional incumbents struggling to adapt.
Will AI take my job? That's the wrong question. Jobs are bundles of tasks. The right question is: Which tasks are suitable for machine learning (SML)?
We find high SML scores not just in low-wage work, but in high-wage, cognitive professions. Radiologists, coders, and writers face high exposure to task substitution.
Take a single occupation. It's a mix of information processing, social interaction, and manual dexterity. AI might be great at the first, but terrible at the others.
The future of work isn't about mass unemployment. It's about rebundling. We strip away the automatable tasks and augment the human-centric ones. The job description changes, but the job remains—if we adapt.
Erik Brynjolfsson warns of the "Turing Trap": the obsession with making machines that mimic humans. This leads to a Substitution path: high productivity, but stagnant wages and falling labor share.
There is another way: Augmentation. Building machines that do what humans cannot do. This path boosts productivity and wages, keeping labor relevant.
The choice determines the distribution. Substitution funnels gains to capital owners. Augmentation spreads them to workers. It's not a technological inevitability; it's a choice.
We are at a crossroads. Policy, tax incentives, and design choices will determine which path we take. The future is not written.
From the Invisible Hand to the Hydraulic Machine, from the Digital Turn to the Intelligence Age. Economics has always evolved to match the technology of its time.
Today, we must model learning systems, active capital, and algorithmic institutions. The old concepts—markets, prices, incentives—remain, but they are now embedded in a digital, intelligent substrate.
The Intelligence Age is just beginning. Whether it leads to shared prosperity or concentrated power depends on how we design the institutions that govern it.