Sargent on the AI New Economy: AI Is Still in the 'Kepler Stage'; Facing the Unknown, the Most Important Thing Is Humility

"The U.S. is pouring enormous investment into AI, built on questionable models. No one truly knows what these models will ultimately produce."

On September 9, Thomas Sargent, 2011 Nobel laureate in economics and professor at New York University, said at the 2026 Inclusion·Bund Summit insights forum that AI is becoming a major force driving economic growth, but investment, decision-making, and policy-making around AI are also facing enormous uncertainty. The expanding scale of capital investment does not mean future returns are more certain. This was a major intellectual discussion ahead of the official opening of the 2026 Inclusion·Bund Summit. Economists, industry leaders, and investors from home and abroad discussed AI and economic growth, productivity, investment decisions, and the future shape of the economy.

In Sargent's view, understanding the economic changes AI is bringing first requires acknowledging a fact: we are entering a stage where "we know AI is important, but we still do not know where exactly it will take the economy."

He compared this to Kepler and Newton in the 17th century.

Kepler summarized the laws of planetary motion through extensive astronomical observation data, but at the time did not know the physical reasons behind these laws; Newton further explained the basic principles behind them. "Today's AI is a bit like Kepler," Sargent said.

Current AI is already very good at identifying patterns from massive data, fitting regularities, and making predictions, but that does not mean it truly understands why the world operates as it does. The progress AI has made so far largely comes from unprecedented data scale, computing power, and more complex models.

But in Sargent's view, the real frontier of AI is whether it can move from the "Kepler stage" to the "Newton stage" — not only discovering correlations, but also further understanding and inferring the structure behind things, achieving better generalization beyond training data, and recognizing the boundaries of its own knowledge.

"We still do not know when AI will enter the 'Newton stage,'" Sargent said.

This also means that the arrival of the new AI economy does not mean the future is already determined.

On the contrary, as AI goes deeper into areas such as financial investment, business operations, scientific research and innovation, and public policy, how to face "uncertainty" may become an increasingly important economic issue in the AI era.

Sargent reminded that many people understand neural networks as a "model-free" learning method, but in fact, every AI algorithm contains specific parametric models and many implicit assumptions about the world. A model may perform very well on familiar data and in familiar environments, but when the environment changes and new situations outside the training data arise, the model may fail.

This is also the difference between "risk" and "uncertainty" in the economic sense.

Risk means that although outcomes are unknown, probabilities can be roughly estimated; uncertainty means that people cannot even accurately know what may appear in the future or the probabilities of various outcomes. Many of the changes brought by AI fall precisely into the latter category.

Sargent's long-term research on "robust control" theory is precisely an attempt to solve such decision-making problems.

Under this approach, the goal of decision-making is not to find an "absolutely correct" model, but to make decisions that remain relatively reliable even when the model deviates or is wrong. In other words, one should not only ask "if the model is right, what is the best choice," but also "if the model is wrong, can this choice still be tolerated."

This is particularly important for the emerging AI new economy.

AI is changing productivity, as well as corporate R&D, financial investment, education and research, and consumer services. But how much productivity will increase, where new value will be created, which industries will be restructured, and how this added value will ultimately be distributed — none of these have definitive answers yet.

"We are at the core of the unknown," Sargent said. Therefore, he repeatedly emphasized "humility."

In his view, in the face of the enormous changes brought by AI, enterprises, investors, and regulators alike cannot simply rely on an idealized model to predict the future. They need to acknowledge the limitations of models, leave room for the unknown and for errors, and establish decision-making mechanisms that can cope with different possibilities.

This may also be another side of understanding the AI new economy: what AI brings is not only more powerful predictive capabilities, but also a renewed understanding of the boundaries of the unknown.

For today's AI, what is truly scarce may not be more predictions, but, beyond prediction, acknowledging the unknown, leaving room, and making more robust choices amid uncertainty. AI can become increasingly intelligent, but in facing the AI new economy, humans may first need to learn to admit: we still do not know.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.