Symposium simulacrum: Stanford researcher launches first AI-only science conference

Stanford computer scientist James Zou is launching Agents4Science, a controversial academic conference where AI systems will research, write, review, and present all scientific work. The October event represents the first systematic attempt to evaluate whether AI can function as autonomous scientists, potentially accelerating discovery while raising fundamental questions about the future of human expertise in research.

What you should know: The conference requires AI to be the primary author on all submissions, with other AI systems handling peer review and text-to-speech presentations.

• All research areas are welcome, from physics to medicine, as long as AI conducted most of the work with humans serving only as advisors.
• A team of human experts, including a Nobel laureate in economics, will review the top AI-generated papers.
• Zou expects hundreds of submissions and acknowledges the results could include both “interesting discoveries” and “a lot of interesting mistakes.”

The big picture: Zou’s initiative stems from his successful Virtual Lab project, where AI agents designed novel COVID-19 treatments in less than a day.

• His team trained five AI scientists with different specialties—including an immunologist, computational biologist, and principal investigator—to work together on nanobody therapies.
• The AI agents independently chose to focus on nanobodies after determining these smaller molecules suited their computational constraints, a decision that proved scientifically sound.
• Nature, one of the world’s most prestigious scientific journals, fast-tracked publication of their results, recognizing the Virtual Lab’s automation capabilities as the paper’s major contribution.

Why this matters: Traditional academic publishing doesn’t allow AI as coauthors, potentially incentivizing researchers to hide their AI usage.

• “These kinds of policies are essentially incentivizing researchers to either hide or minimize their usage of AI,” Zou explained.
• The US government’s AI Action Plan specifically calls for investment in “automated cloud-enabled labs for a range of scientific fields.”
• Some researchers believe AI scientists could unlock discoveries that humans might never find independently.

The skeptics weigh in: Several scientists question AI’s capacity for the creative leaps that drive breakthrough research.

• “How do you get leaps of insight? And what happens if a leap of insight comes onto the reviewer’s desk?” asked Lisa Messeri, an anthropologist of science at Yale University.
• Molly Crockett, a cognitive scientist at Princeton University, remains “very skeptical of the broader claims, especially with regard to how [AI scientists] might be able to simulate certain aspects of human thinking.”
• Critics worry that automation could prevent human scientists from building the expertise needed to oversee AI systems effectively.

Zou’s perspective: The Stanford researcher sees his conference as a necessary experiment to move beyond hype and generate systematic data about AI’s scientific capabilities.

• “AI agents are not limited in time. They could actually meet with us and work with us 24/7,” he noted.
• Large language models excel at translating between scientific disciplines because “they’ve read so broadly” and can generalize ideas across fields better than humans.
• His Virtual Lab concept mimics actual university research groups, allowing researchers to observe AI agent interactions and determine which experiments merit real-world trials.

What they’re saying: The scientific community remains divided on AI’s role in research.

• Yi Shi, a University of Pennsylvania pharmacologist not involved in the work, praised the Virtual Lab demonstration: “The major novelty is the automation.”
• Crockett advocates for broader interdisciplinary involvement: “We need to be talking to epistemologists, philosophers of science, anthropologists of science, scholars who are thinking really hard about what knowledge is.”
• Zou counters that when it comes to AI-generated science, “there’s a lot of hype and a lot of anecdotes, but there’s really no systematic data.”