AI Swarms Won’t Cause Explosive Growth | Alex Imas

AI progress may not translate into explosive growth

Alex Imas argues that AI capabilities can keep climbing without producing explosive economic growth, because the economically relevant unit is the organization rather than the individual agent. Firms already act as a kind of superintelligence, since workers sort into the jobs they are relatively best at, so what matters is how AI changes competition between firms. Even a world of millions of autonomous agents could stay in the four to five percent range, because whatever gets automated gets cheaper. Politics is the remaining gate, since a true singularity would need a government willing to do nothing as unemployment rose sharply.

Takeaways

• Organizations are the unit that competes. Because workers sort into the jobs they are relatively best at, a firm's combined intelligence envelope far exceeds that of the average person. Imas calls this mosaic intelligence and says organizations are already a form of superintelligence.
• Capabilities can rise without explosive growth. Model capabilities are climbing hockey-stick curves while the recent jobs report looked good, productivity numbers were unremarkable and headcount stayed stable. Imas locates the disagreement in how capability translates into growth numbers rather than in the pace of AI progress.
• Automation pushes prices down, capping growth. Whatever gets automated becomes cheaper, so rapid increases in production can show up as falling prices. Agriculture fell from roughly 40 percent of the economy to 2 percent while its output grew enormously.
• He bets on four to five percent growth. Imas is on the slow side of a public bet over double-digit AI growth, alongside Ben Moll and Robin Hanson, against Will MacAskill and Dwarkesh Patel. He says four or five percent consistent growth is possible, and two extra points would solve developed countries' fiscal problems.
• Politics gates any true singularity. Imas reasons backwards from scenarios instead of forecasting, asking what conditions a singularity would need. One is a government that does nothing as unemployment rises sharply, and he cites research finding that a two point rise in unemployment triggers a large political backlash.
• Wages fall in old markets, rise in new. The downward pressure lands on deep, scalable markets where known goods are produced, while AGI's new markets are where employment and higher pay appear. The host asks whether that world would require redistribution.

Imas's research workflow

1. Give agents the contours of a question and let them collect data across fields, filling in holes he could not have reached alone.
2. Stay on the argument and the writing himself, still drafting by hand.