Anthropic co-founder Jack Clark says a billion virtual scientists are coming. Or 10 billion.

Finding each new scientific idea tends to cost more than finding the one before it. A major economics study published in 2020 found that keeping up the pace of doubling the number of components that fit on a computer chip required more than 18 times as many researchers as it did in the early 1970s. This means researchers became about 18 times less productive at making progress in chip technology. Drug discovery shows a similar pattern. Since 1950, the number of new drugs approved for every billion dollars spent on research and development (R&D) has roughly halved every nine years. Analysts call this trend “Eroom’s Law.” Much of science faces the same problem. Many of the easiest scientific discoveries have already been made. Many more discoveries are still possible, but finding them is rarely cheap. So far, the main solution has been to hire more researchers, and training new researchers takes years.

Anthropic co-founder Jack Clark believes AI could change these costs across science. Speaking at the IMF-World Bank Annual Meetings in Bangkok on October 17, Clark said the singularity had begun. He described the singularity as a time when AI makes scientific and technical progress so fast that people struggle to keep up. He asked the audience to prepare for a world where about 10 million human scientists direct a billion virtual scientists, or even more. “How would my work change if I could make copies of myself and order them to do work under my supervision?” Clark asked. “That is what the next few years will be like for large swathes of the economy – we will be able to augment human labor in frontier disciplines, like science, in a way we’ve never been able to do before.”

Clark gave the 2026 Per Jacobsson Foundation Lecture, titled “Inside the Event Horizon: Why Everyone Should Take AI Progress Seriously.” His audience included economists, bankers and policymakers. Two days later, he published the speech text in his Import AI newsletter. He began by explaining his own background and experience.

“I am a human being. There is one of me. It has taken me almost 40 years to build the experience to give this speech.”

Clark worked as a journalist covering data centers about 20 years ago. He left journalism to join OpenAI in 2016 and co-founded Anthropic a few years later. He based his ideas on science-fiction writer Vernor Vinge, who predicted in 1993 that intelligence greater than human intelligence would make progress happen at a speed people would struggle to follow.

“They call this the singularity. I believe that this year we have gained convincing evidence that it has now begun.”

Clark said he would focus his lecture on the benefits of AI, most of which, he argued, would come through science.

“I believe that AI-driven scientific invention will change the course of the global economy, alter the ways countries plan and build their own infrastructure, make us all healthier and able to spend more time with our loved ones…”

From better coding to solving scientific problems

Clark's first example was software development. He said Anthropic researchers now write almost 10 times as much code as they did during the company's first four years. These same improvements have also made AI systems better at both defending against cyberattacks and carrying them out.

“These capabilities will not stop at coding; the same thing will happen to science and robotics. It has already begun to happen.”

Humanity’s Last Exam, a test covering subjects from chemistry to classical literature, launched in January 2025. Clark said the highest AI score at the time was 8%. Claude Opus 5.5, released on September 22, scored 67.7% when allowed to use tools. It had the highest score as of Clark's speech. He ended his examples by discussing TheoPhysEval, a paper published on October 9. Physicists created this test using 50 unpublished problems from their own research. Clark quoted co-author Suvrat Raju’s post about the results. GPT-6 Astra's score rose from the high 80s to 98%, solving 49 of the 50 problems. This happened after the AI models argued that some expert answers were wrong and the experts found mistakes in their own calculations.

“Some of the world’s scientific experts built a test out of problems an AI could never have possibly read the answer to. And when they tested the AI, not only did it score high, but it got a higher score by finding errors in the answers that the humans had worked on!”

The paper's authors describe the problems as “‘turn-the-crank’ research problems, where known techniques are used to generalize an extant result.” In other words, they involve using established methods to extend an existing result. The authors report that experts made mistakes in “several” cases. Claude Opus 5.5 solved 46 of the 50 problems. The authors also say that AI agents “do not appear to have the same ‘taste’ in problems.”

A billion virtual scientists, and the labs they would need

Clark referred to the economists' finding that “good ideas are getting harder to find.” He said training a new group of researchers can take “10 or 15 years of targeted education, and millions of dollars.” He then returned to the value of his own experience.

“What would it mean to be able to run hundreds or thousands of copies of me simultaneously? What would it mean if new versions of me could be created instantly, rather than waiting 40 years?”

Clark estimated that there are about 10 million scientists worldwide, citing UNESCO. The agency's most recent Science Report counted 8.854 million full-time-equivalent researchers in 2018. Clark said that even this number of researchers does not have enough laboratory space, biology labs or large scientific instruments to carry out all the experiments they want to do.

“What happens when instead of 10 million scientists, we have a billion virtual scientists directed by our own 10 million? Or 10 billion? That is the world that is coming right for us. We have to get ready.”

More proposed drugs, but limited ability to test them

Clark said that if scientists can make discoveries in much greater numbers, society must avoid creating new delays later in the process. Policymakers will need to help move discoveries “out of the lab and into the clinic, the factory, and the field.”

“An avalanche of new proposed drugs is of little use if only a handful can ever be tested.”

AI moves into real laboratories

Earlier in his speech, Clark said AI systems “are now increasingly able to autonomously operate laboratory equipment remotely.” He said these systems are being used to work on things such as batteries and fuel cells. In August, Anthropic released the Model Hardware Standard, an early research version of a system developed with HHMI’s Janelia Research Campus. It lets AI agents operate lab equipment that has a programmable interface (R&D World coverage). Anthropic said the system helped a Carnegie Mellon team run experiments that measure how a substance's effects change with its dose about three times faster. Genentech researchers also had to teach Claude that foaming in samples caused real physical problems, rather than software bugs. In September, Reuters reported that Anthropic had set up a wet lab in the San Francisco Bay Area. Eric Kauderer-Abrams, Anthropic's head of life sciences, told Reuters that “the final test is still and will be for a while in real lab work.”

“Just because we have an AI system that can do science, that doesn’t mean that it should be left to do science unattended.”

Clark ended by referring to Carl Sagan. In his book Pale Blue Dot, Sagan wrote that the people who eventually travel to other stars will be “a species very like us, but with more of our strengths and fewer of our weaknesses.”

“Where Sagan thought we had hundreds of thousands of years to achieve this, to become a better and wiser people, we instead have a handful of years. I believe we can do it.”