Nobel Laureate Sargent: AI Remains in 'Kepler Stage,' Huge Investment Built on Questionable Models

The discussion was held as a major forum ahead of the opening of the 2026 Inclusion Bund Conference, bringing together economists, industry executives, and investors to discuss AI's impact on economic growth, productivity, investment decisions, and the future shape of the economy.

Sargent described AI as being in a "Kepler stage," referring to Johannes Kepler, who used extensive astronomical data to describe planetary motion without understanding the physical causes behind it. Isaac Newton later explained those underlying principles. AI today, Sargent said, is good at recognizing patterns, fitting regularities, and making predictions from large data sets, but that does not mean it truly understands why the world operates as it does.

Whether AI can move from the "Kepler stage" to a "Newton stage" remains an open question, Sargent said. This would involve not just finding correlations but also understanding and inferring the structure behind things, generalizing beyond training data, and recognizing the limits of one's own knowledge. "We do not yet know when AI will enter the Newton stage," he said.

The arrival of the AI-driven economy does not mean the future is already determined, Sargent said. As AI increasingly enters fields such as financial investment, business management, scientific research, and public policy, dealing with uncertainty may become a more important economic issue.

Sargent also cautioned that many people treat neural networks as a "model-free" way of learning. In fact, every AI algorithm contains parameterized models and implicit assumptions about the world, he said. A model may perform well in familiar data and environments, but it can fail when conditions shift and new situations emerge beyond the training data.

This relates to the distinction in economics between risk and uncertainty, he noted. With risk, outcomes are unknown but probabilities can be roughly estimated. With uncertainty, people cannot even know what future outcomes are possible or what probabilities apply. Many of the changes brought by AI fall into the latter category, he said.

Sargent pointed to his own research on "robust control" as a way to address such decision problems. Instead of finding a single "absolutely correct" model, the goal is to make decisions that remain relatively reliable even when models are biased or wrong. The question is not only what the best choice is if the model is correct, but also whether that choice can be tolerated if the model is wrong.

AI is changing productivity and reshaping corporate research and development, financial investment, education, science, and consumer services, Sargent said. But there are no definitive answers as to how much productivity will rise, where new value will be created, which industries will be restructured, or how the added value will be distributed.

"We are in the heart of the unknown," Sargent said. He repeatedly stressed the need for humility. In his view, companies, investors, and regulators cannot simply rely on an idealized model to predict the future. They must acknowledge that models have limitations, leave room for mistakes, and establish decision mechanisms that can handle different possibilities.

AI may become increasingly intelligent, but when facing the new AI economy, humans may first need to admit that they still do not know, Sargent concluded. The real scarcity for AI today is not more prediction, but rather acknowledging the unknown, reserving leeway, and making robust choices amid uncertainty.