Artificial Intelligence (AI) is beginning to accelerate scientific discovery at a pace that could reshape the global economy, according to Anthropic co-founder Jack Clark, who urged governments to prepare for a period of technological change that could outstrip current expectations.
Speaking at the 2026 Per Jacobsson Lecture during the IMF-World Bank Meetings in Bangkok, Clark said recent advances in AI systems provide convincing evidence that what researchers have called the technological singularity has begun.
The term describes a period when technological progress accelerates so rapidly, potentially through increasingly capable artificial intelligence, that conventional expectations about the pace of change become unreliable.
Clark cited improvements in AI performance on difficult mathematics and scientific tests, alongside applications in software development, drug discovery and laboratory automation, as evidence that AI is moving beyond routine assistance towards work at the frontiers of knowledge.
At Anthropic, researchers now produce almost 10 times as much code as they did during the company’s first four years, Clark said, attributing the increase to advances in AI capabilities.
He also pointed to performance on Humanity’s Last Exam, a test covering disciplines including mathematics, chemistry, computer science and ecology. According to Clark, the highest score had risen from 8% when the test launched in January 2025 to 67.7% by the week of his lecture.
On FrontierMath, a benchmark designed to test advanced mathematical reasoning, he said the best-performing systems had improved from 2% to 94%.
Clark also cited results from TheoPhysEval, a test involving 50 unpublished theoretical physics problems. He said an AI system scored 98%, or 49 out of 50, after researchers revisited several answers initially considered incorrect.
The broader economic implication, he argued, is that AI could increase the supply of researchers able to investigate complex problems, reducing the time required to generate new ideas and develop technologies.
Clark said AI systems could eventually allow individual scientists to supervise hundreds or thousands of digital research assistants, expanding research capacity without requiring each assistant to undergo years of human training.
He cited Anthropic modelling suggesting that, under an extreme scenario of rapid AI advancement and adoption, US gross domestic product could reach $44 trillion by 2030. The projection depends on assumptions about future AI capabilities and should not be treated as a baseline economic forecast.
Clark acknowledged that faster scientific progress could bring substantial risks, including more capable cyberattacks and the misuse of biological research. He called for governments and technology companies to establish stronger safeguards while investing in the infrastructure needed to turn scientific discoveries into practical benefits.
His central argument was that the economic consequences of AI could extend far beyond the technology industry, affecting the pace of innovation across the wider economy.
