KorvaJoin the waitlist
Tech

Even Human-Level AI Must Beat the Team, Not the Average Worker

Economist Alex Imas says firms win by giving each job to the best-suited person, and AI will have to clear that higher bar.

AI image
The a16z Show · Alex Imas & Tyler Cowen | AI Is Moving Fast. Why Isn’t the Economy?

When economist Alex Imas was asked about his new paper, he started with a puzzle. AI keeps scoring higher on tests of individual skill, yet the economy looks largely unchanged.

His answer, worked out with collaborators, was this: "The relevant competitive unit is not the agent, it's the organization."

Imas argues that the usual comparison is wrong. Benchmarks pit one AI system against one person, or against the average person, but businesses do not work that way.

A firm, he says, uses what economists call Ricardian comparative advantage, meaning each job goes to the person best at it, not to the average worker. So the group's combined ability is far above the average.

He describes organizations as "a form of superintelligence" because they fit different strengths together.

That distinction changes what human-level AI would mean. Imas defines minimal AGI as AI "as good as average human capability on every single dimension," and says even that milestone would not bring major economic transformation.

Production would get cheaper, he says, and that could move wages. But in his argument, average competence is unlikely to beat a team built to exploit the best competence for each task.

The milestone that would matter more, in his account, is coordination. Today's agent swarms mostly split work into parallel pieces rather than truly dividing labor.

On coordination tests, he says, two AI agents that can communicate can do worse than one agent alone. Give the same work to people, and they communicate, specialize, and do much better than one person with twice the resources.

If AI learns to coordinate like people do, Imas says, it could break what he calls "the Coasean ceiling on firm size." That is the limit on how big a firm can get before coordination costs stop it growing. Then firms themselves could look very, very different.

Questions people ask

What does comparative advantage mean here?

It is the idea that people and firms gain by specializing in what they do relatively best, then trading or collaborating. Imas uses it to explain why a team can outperform even a talented all-rounder, because each task goes to the strongest fit.

What is minimal AGI in this argument?

Imas uses it to mean an AI system that matches average human skill on every task. His point is that average skill sounds impressive, but organizations already perform above average by matching people to tasks.

Why could wages change even without economic transformation?

If AI can do some work more cheaply, employers may pay less for that human work or shift workers elsewhere. Output can rise a little while paychecks change a lot, without reorganizing whole industries.

More in Tech