“We’ll Need More Scientists”: An Essay on the Human Role in the Expanding Machinery of Science

Scientific AI agents will broaden the range of questions we can investigate, so we will need more scientists, according to Josh Bloom

On August 26, Josh Bloom, cofounder of Valency, a company that develops AI tools for science, published an essay on scientific AI agents, programs that perform part of the research process. He argues that their wider use will require more people to choose research questions, scrutinize answers, and take responsibility for decisions.

Scientific agents can already perform some research tasks. Bloom argues that new ways to observe, calculate, and compare data increase the range of questions that researchers can test. Laboratories must therefore decide more often where to direct their time, money, and attention.

“No amount of filling the volume can replace choosing a direction worth pursuing.”

In July, mathematician Terence Tao wrote that AI can propose proofs faster than mathematicians can understand how those proofs fit into the broader theory. Bloom extends this gap to the entire research cycle. Before seeking an answer, laboratories must decide whether the question warrants their time, money, and attention.

Once an answer has been obtained, a different kind of human work begins. Researchers must identify which of many plausible results looks too good to be true, then test whether it withstands careful scrutiny.

“When we encounter new results, the skeptic in each of us becomes more valuable as the number of seemingly plausible results grows.”

Finally, the conclusion must belong to a person or organization that decides what to do next and accepts responsibility for the consequences.

“Responsibility grows with the number of people, not with computing power.”