Genie in a bottle meets Agent Anderson, Agent Smith, Morpheus, and Trinity. “What could go wrong,” you ask? WHAT COULD GO RIGHT?!
__________
Andrew Akbashev
The Stanford team created a Virtual Lab with AI scientists, achieving a 90% success rate in real-world validation studies.
📍 Nature News highlights a groundbreaking study:
1. Three ‘AI scientists’ were trained in distinct discipline: immunology, computational biology or machine learning.
These AI experts collaborated to “achieve goals set by human researchers,” leveraging their complementary skills.
2. Human scientist held regular ‘team meetings’ with AI agents to ‘evaluate their progress’ (5-10 mins per meeting).
“The virtual lab is designed to be mostly autonomous, so the agents discuss with each other. They decide what problem to solve and what approach to take and how to implement those approaches,” says Zou. “The human researchers focus on providing more high-level feedback to guide the direction of the virtual lab.” [Nature]
⭕utcome:
The Virtual Lab designed antibody fragments capable of binding to a virus, generating around 100 structures in a short time.
In real-world experiments, over 90% of these structures successfully bound to the target virus, validating their quality.
“We designed it to be a very versatile platform. So in principle, we can use these virtual lab agents and ask them to solve different scientific problems,” – James Zou, co-author.
The experiment “represents a new paradigm of taking AI as collaborators, not just tools”, says Yanjun Gao [Nature].
❗ My advice to PhD students / postdocs:
In the next decade, AI can completely transform how we work in labs.
Try to get experience in ‘AI scientists’ and automated labs. These are NOT just ‘hot topics’. This is the future. Make yourself an expert in it.
