SOSV Deep Tech LIVE – Why Is the Brain So Fast and what AI Can Learn, with Nobel Laureate Dr. Jim Rothman

A fascinated full house listened in as SOSV's Sean O'Sullivan led a fireside chat with Professor James Rothman, Ph.D., winner of the 2013 Nobel Prize in Physiology or Medicine and Sterling Professor of Cell Biology at Yale University. The conversation moved from the molecular machinery of thought to what AI can learn from the brain - and the increasingly difficult business of funding scientific discovery.

Professor Rothman is one of the few scientists to have shaped both the foundations of modern biology and its commercial future. Trained as a physicist before becoming a biochemist, he currently serves as the Director of the Yale Nanobiology Institute. Beyond his landmark academic discoveries, he co-founded ARIAD Pharmaceuticals, served as Chief Scientist at GE Healthcare, and actively shapes the startup ecosystem as a board member and advisor to bleeding-edge platforms like Prellis Biologics and Celesta Capital.

Rothman shared the Nobel Prize for explaining machinery that helps cells transport and release substances. But his latest work tackles a puzzle that the earlier discovery left unresolved: How can the same molecular machinery operate thousands of times faster in the nervous system than elsewhere in the body?

His emerging answer involves twelve molecular machines, an artificial synapse, and a pressure-driven mechanism that could change how we think about communication in the brain.

"I haven't published it yet. It'll be very controversial when it's published."

The Same Machinery, an Astonishing Difference in Speed

Cells package substances into tiny membrane-bound containers called vesicles. In the pancreas, those packets store insulin for release when it is needed. At nerve endings, they carry neurotransmitters that relay signals to other cells.

Rothman's foundational research revealed how proteins called SNAREs bring membranes together so that a vesicle can release its contents. Think of a molecular zipper: proteins anchored in the two membranes assemble together, pulling the membranes into contact and driving fusion.

But the timing presents a puzzle. Rothman explained that some fusion processes elsewhere in the body take roughly a second or more. At a synapse, release can happen in as little as 100 microseconds.

The same basic machinery. A thousand- to ten-thousand-fold difference in speed.

For Rothman, that raises a question with consequences far beyond cell biology: What sets the brain's "CPU speed"?

Ten Years to Build an Artificial Synapse

To investigate, Rothman and his colleagues spent a decade building a system they could control.

They purified the necessary proteins, combined them with membrane components, and constructed artificial vesicles containing a colored dye. A second artificial membrane stood in for the cell surface. The goal was to reproduce release in less than a millisecond after adding calcium.

They achieved it.

The advantage is precision. In living brain tissue, countless processes happen simultaneously. In the reconstructed system, researchers determine which proteins are present and can watch individual molecules assemble. They also tested disease-associated versions of the proteins, reproducing defects that helped establish the system's biological relevance.

Then came a striking result: in this simplified system, each vesicle assembled exactly twelve SNARE complexes.

"12. Not 11, not 13, 12."

Rothman described two waves of assembly: six, followed by six more. In disease-associated cases he discussed, the system assembled only six, and release slowed dramatically.

The next question was whether twelve simply supplied more force-or whether the second six were doing something different.

Six to Fuse, Six to Build Pressure

According to Rothman's proposed explanation, the second six SNAREs pull on the vesicle wall, increasing pressure inside it.

His team's calculations put that pressure at approximately two to three atmospheres. When the fusion opening forms, the stored pressure helps widen it and drive the contents out rapidly.

To test the idea, the researchers used molecular pores to puncture the artificial vesicles.

Even with twelve SNAREs still present, the punctured vesicles released their contents slowly.

That experiment supports the idea that the extra SNAREs contribute something beyond additional fusion machinery: they help store energy as internal pressure.

It is a physical explanation for a biological speed problem - and, as Rothman emphasized, work that was still unpublished at the time of the conversation.

Could the Brain Carry a Second Kind of Message?

The pressure hypothesis leads to an even more provocative possibility.

When pressurized liquid escapes a vesicle, it creates flow. Could a synapse detect that flow as a signal in its own right?

Rothman pointed to proteins found at synapses that are also associated with sensing mechanical forces elsewhere in the body. He is investigating whether they could respond to the fluid released by vesicles, creating an additional communication mechanism alongside the chemical action of neurotransmitters.

He was explicit about the uncertainty:

"I don't know if it's right, or I don't know if it's wrong, but I do know that we're pursuing it."

If the idea holds up, it could offer new ways to investigate memory and develop mechanically targeted interventions, including focused ultrasound.

During the audience discussion, Rothman went further, suggesting that there might even be vesicles that carry no neurotransmitter but still deliver a pressure-driven signal. That remains a hypothesis, not an established discovery. It illustrates how a carefully controlled experiment can open a question much larger than the one that started it.

What AI Could Learn From a Twenty-Watt Brain

For an audience of deep tech founders and investors, the conversation naturally turned to computing, and what AI can learn from the brain.

Rothman sees enormous promise in connectomics - the mapping of neural connections - and in efforts to emulate the behavior of those networks. Better wiring diagrams, combined with a deeper understanding of how synapses work, could give engineers more useful principles for designing computing systems.

The long-term attraction is energy efficiency. As Rothman noted, the human brain operates on roughly twenty watts.

He sees an opportunity for biology to inform AI hardware as researchers learn more about the systems they are trying to emulate. His own molecular research does not yet provide a chip design. But it underscores how much remains to be understood about biological information processing.

"I personally think the other way is going to come from biology."

For companies confronting AI's growing energy demands, Rothman's challenge was straightforward: invest in understanding the biological machinery that already processes information so efficiently.

A Nobel-Winning Research Program, Built With Someone Else's Waste

One of the evening's best stories reached back to Rothman's early research at Stanford.

His small laboratory needed enormous quantities of cultured cells to isolate an enzyme. Producing enough material themselves was out of the question.

A connection to Genentech provided a solution: leftover cell pellets from manufacturing its tPA drug.

The arrangement, Rothman recalled, was settled with a handshake. Material that was a byproduct for one organization became a resource for fundamental discovery in another.

"So one guy's garbage, turns out to be another guy's pearl."

That experience sits alongside a substantial career in industry, including co-founding ARIAD Pharmaceuticals and serving as chief scientist at GE Healthcare. Rothman said those roles changed how he ran his laboratory, coordinated multidisciplinary work, and thought about translating discoveries into applications.

For founders, one lesson stood out: recognize when the research plan needs to change. When questioning the plan starts to feel almost antisocial, he suggested, that may be precisely when the company needs to reconsider it.

Who Pays for the Next Decade of Discovery?

A synthetic synapse took ten years to build. The discoveries it enabled emerged only after that investment.

That creates a difficult funding problem. Rothman described the challenge of sustaining patient, long-term research within systems that increasingly expect results on shorter, project-based schedules.

He also resisted an easy answer: venture capital cannot simply replace public funding for basic science. Investors need a credible path to returns; research that establishes entirely new concepts may be far from an identifiable product.

Asked about the medical impact of his work, Rothman offered a different way to understand its value: basic science as a lantern.

A discovery illuminates an area that was previously dark. Researchers working on different diseases can then use that understanding to identify their own targets and treatments. The impact may be broad and profound, even when it is delayed and difficult to measure.

For founders and investors, that is the larger question behind the evening's molecular surprises: How do we sustain the research that makes entirely new fields possible?

Rothman remained optimistic about discovery itself - even as he warned that the places benefiting from it can change.

"Science and discovery will always win out. It's only a question of which culture embraces it."

Watch the full conversation:

Broadcast live from our SF and NY event spaces, SOSV Deep Tech LIVE is our video podcast series exploring the frontiers of deep tech with leading founders, researchers, and technologists. Check out some of our other popular videos:

The 100-Year Quest for Fusion Energy with Sir Steven Cowley, Director of PPPLIt's a popular concern that conventional "share-holder primacy" governance is catastrophic for stewarding AI labs and LLMs. OpenAI, Anthropic, Google, Meta, Cohere, and Palantir have all established some form of alternative oversight or voting power as a guardrail.

Navigating the Realities of Energy Scale Up with Barbara Burger, the "Godmother of Energy Investing"

Scaling Moore's Law & Physical AI with Dr. Rick Gottscho, Retired CTO of Lam Research