Photo: Getty Images
Even major advances in artificial intelligence may not guarantee breakthroughs across every field, as humans will still remain essential for many tasks. Economist, author and Marginal Revolution co-founder Tyler Cowen argues that the development of AI could be constrained by a lack of experimental data, regulation and public distrust.
In an interview with Sana founder Joel Hellermark, Cowen said that systems with significantly higher levels of intelligence would not necessarily produce proportionally greater progress. He is particularly skeptical of the assumption that advanced AI will inevitably develop a theory unifying quantum mechanics and general relativity.
According to Cowen, the obstacle may not be intelligence itself, but the lack of necessary experimental data. He cited physicist David Deutsch, whom he described as an exceptionally intelligent person, while noting that even extraordinary human intelligence does not guarantee solutions to fundamental scientific problems. Some questions, he suggested, may simply remain unanswered.
“I think that in many — though not all — fields, intelligence really does face diminishing returns,” Cowen said. He added that many areas have inherent limits to perfection, although AI could allow humanity to reach those limits more frequently.
At the same time, Cowen believes fields that do not depend heavily on experiments could have much greater potential for further development. One key question, he said, is whether scientific problems can increasingly be explored through simulations. However, he is not convinced that simulations will be able to cover most scientific disciplines within the next 20–30 years.
Discussing the famous “Millennium Problems,” Cowen suggested that AI could solve many of them, assuming they are solvable in the first place. But obtaining those solutions would not necessarily lead to a significant improvement in people’s living standards.
The reason is that new knowledge and ideas must ultimately be turned into real-world products and technologies. At that stage, progress can be slowed by legislation, regulatory requirements, testing procedures and the need to convince people that new technologies are safe and worthwhile.
Cowen pointed to nuclear energy as an example, arguing that public distrust remains widespread in many countries and that convincing people of its benefits is extremely difficult.
He also highlighted the extensive experimentation and testing required before new technologies can be deployed. Among the institutions involved in this process, Cowen mentioned the FDA, which he said seeks to repeatedly verify and re-evaluate relevant decisions.
As a result, he believes technological progress in many fields will remain relatively slow regardless of how intelligent AI systems become.
Moreover, Cowen expects people could become increasingly hostile toward AI as its capabilities grow and begin to perceive the technology as a threat.
“I once said — and it was only half a joke — that we’ll know the AI revolution is going well when people start to hate it,” he said.
Earlier in the same interview, Cowen argued that AI would initially add only around 0.5 percentage points to U.S. productivity growth. In his view, the main obstacle to a rapid macroeconomic acceleration is not the quality of AI models themselves, but the slow adaptation of people and the inertia of conservative institutions, including government, higher education and healthcare. These sectors account for a significant share of the economy and could take decades to fully transform.
