Just three weeks after being named a winner of the 2025 Nobel Prize in Chemistry, Omar M. Yaghi stepped onto a new kind of stage — one where chemistry and artificial intelligence collide.
Yaghi, a professor at the University of California, Berkeley best known as the “father of MOF,” delivered his first public lecture since the Nobel announcement at the AI Accelerating Science (AIAS) Conference, hosted by the Tianqiao and Chrissy Chen Institute.
But rather than revisit the breakthroughs that secured his prize — Metal-Organic Frameworks (MOFs) and Covalent Organic Frameworks (COFs) — Yaghi used the moment to unveil what he called an entirely new scientific paradigm: “chemistry that can think, reason, and evolve by itself.”
“Science has always been a dialogue between nature and the human mind,” he told the packed audience. “Now, with the help of artificial intelligence, we are endowing science itself with the capacity to think.”
Yaghi describes his new framework as “From Molecule to Society” — a closed-loop ecosystem that unifies molecule design, laboratory experimentation, industrial scale-up, and societal deployment. The engine behind it: generative AI, robotics, automated labs, and self-improving intelligent agents.
The lecture marked a conceptual transformation for Yaghi — from chemist building new materials to scientist designing new forms of intelligence that will, in turn, build materials for humanity.
Making AI a Chemist
His first example was deceptively simple.
He asked: Can a large language model understand chemistry well enough to reason like a scientist?
Yaghi’s team trained ChatGPT on thousands of synthesis reports, teaching it to extract reaction parameters, predict outcomes, and determine whether a crystallization experiment would yield single-crystal or polycrystalline materials.
The results, he said, surpassed many traditional expert-based heuristics. In doing so, the AI system evolved from a chatbot into a scientific reasoning engine — translating unstructured natural-language records into actionable experimental logic.
“We no longer need to ask what AI can do for science,” Yaghi said. “Now we need to ask what science will become when it is driven by AI.”
It’s not autonomous discovery — yet. But it represents the first time that chemical intuition, once locked inside human minds and handwritten lab notebooks, has been digitized and made machine-learnable.
“AI doesn’t just help scientists,” he said. “It gives science a new way of thinking.”
From AI to Water in the Desert
In the second case — one he called the “Death Valley Experiment” — the team pushed chemical intelligence into the real world.
MOFs have long been explored as materials capable of capturing water from dry air. But optimizing their molecular structure is a trial-and-error process that can take years.
Instead, Yaghi’s group used ChatGPT-assisted molecular editing to fine-tune water-harvesting MOFs. The AI-driven design enabled a portable, zero-energy device that extracts drinking water from desert air at just 15% humidity — among the driest climates on Earth.
Field tests in the Mojave matched the predictions ChatGPT had helped generate in the lab.
“For the first time, AI-designed materials are directly improving human life,” Yaghi said. “AI doesn’t replace chemists — it amplifies their creativity.”
Seven AI Agents, One Breakthrough
His third example sounded like something from science fiction.
Yaghi described a virtual laboratory staffed by seven AI agents, all based on ChatGPT, each with a different scientific role: experiment planner, literature researcher, data analyst, safety reviewer, algorithm developer, robot controller, and Bayesian optimizer.
Working together, these agents designed and ran hundreds of COF-323 crystallization experiments autonomously, using robotic synthesis platforms. COF-323 — notoriously difficult to produce in crystalline form — was transformed from amorphous powder into a fully crystallized material within days.
This proof-of-concept, Yaghi said, represents the beginning of “AI self-collaborative science” — where digital scientists interact not only with humans, but also with each other and with automated laboratories.
The implications extend well beyond reticular chemistry: a world in which experiments generate themselves, 24 hours a day, improving continuously.
A Startup to Scale “Molecule-to-Society” Science
Yaghi closed with a look toward commercialization.
A Berkeley-born company called AIMATX is building software and robotics that operationalize his entire Molecule-to-Society architecture, consisting of three AI-powered layers:
Design Layer — algorithms generate and predict new molecular frameworks
Synthesis Layer — automated agents perform and analyze experiments
Scaling Layer — AI evaluates markets, production logistics, and regulatory pathways
Every step feeds data back into the system, accelerating discovery with each iteration.
“We are building a living, never-resting system of discovery,” Yaghi said.
He displayed newly identified ZIF and LZIF crystal structures discovered using this cycle — the discovery rate is already twice that of traditional human-only exploration.
The presentation amounted to more than a showcase of breakthroughs. It was a declaration that the nature of scientific research is changing — and that chemistry may be the field most transformed.
“We’re not just accelerating experiments,” Yaghi said. “We’re accelerating humanity’s ability to solve problems.”
The remark drew long applause, along with a standing ovation when a group photo of Yaghi and his students appeared behind him — symbolizing the human-AI collaboration at the heart of this new era.
Yaghi’s lecture — his first as a Nobel laureate — may ultimately be remembered less for MOFs and COFs than for his audacious vision of what comes next.
A scientist who once turned metal ions and organic molecules into sponges that capture CO₂ and water is now trying to engineer a thinking chemistry — one capable of discovering the materials that do not yet exist.
“This is just the beginning,” he said. “Chemistry is learning to think.”
And with that, the audience witnessed something rare: the moment a founder of a field announced he was reinventing it — again.
