AI’s Value Is Collective—Not a Private Secretary
From Michael I. Jordan on Machine Learning Street Talk: today’s AI aggregates input from billions and should serve billions. Chasing a always-on private secretary misses the bigger economic problem.
Much of the AI conversation sounds personal: your assistant, your coach, your always-on secretary. UC Berkeley and Inria scholar Michael I. Jordan pushes the opposite frame. Today’s systems already sit on data from billions of people. The hard question is how they serve billions of people—not how they whisper in one user’s ear all day.
A collectivist view, in plain terms
Jordan has argued for what he calls a collectivist economic perspective on AI. The word can sound abstract. His meaning is concrete: modern AI is already a vast network that takes collective input and returns collective output. Understanding that network needs computer science plus social science—especially economics.
People cooperate and compete. Markets have done this for thousands of years. The point is not to chant “collectivism” as a slogan. It is to turn those social interactions into workable mathematical frames—so systems can allocate value, reduce harm, and improve coordination at scale.
Why the private-secretary dream is a weak business
Jordan is blunt about one popular product fantasy: a large language model as your personal secretary, offering advice and handling chores all day. He calls it a terrible business model. People still want to think for themselves. Few want an intelligent presence hovering through every hour. An end-of-day summary can help; constant interruption usually does not.
Meanwhile, healthcare, transport, and finance already move data among billions of participants. Machine learning already runs inside those systems. What is often missing is an economic lens: what do participants want, how do they cooperate and compete, and how should technology improve that market—not just chat with one user.
Multi-agent systems do not magically create markets
Some Silicon Valley voices suggest that wiring large language models into multi-agent systems will somehow produce economic value on its own. Jordan rejects that as poor engineering thinking. Mixing chemicals without theory once led to explosions and harm; stacking agents without economic design can also create real damage—from platform harms to careless talk about jobs vanishing and “new ones appearing.”
- Collective input from many people is already the fuel of modern AI
- Collective service—markets, logistics, care, finance—is where large value lives
- Economics turns “many people interacting” into design constraints, not slogans
Not against AI—against the wrong target
Jordan says he is not trying to tear AI down. He wants it better, safer, and more valuable. That means treating humans as both producers and consumers whose roles should be respected and amplified—not covered over by AGI marketing. The future he points to is not a sci-fi super-mind, but a collective tool that makes ordinary life work more fairly and more reliably.