Andrew Huberman says AI labs are biotech firms but the tech is sellable
By Nikhil Raghavan · Reporting from San Francisco ·
The most arresting prediction on the latest episode of Invest Like the Best is about the inevitable transformation of the Silicon Valley elite into neurotech giants. Andrew Huberman, the Stanford neuroscientist, told host Patrick O'Shaughnessy that Meta, OpenAI, and Anthropic are not merely large language model companies. Instead, he argued, they are biotech companies in waiting. Their ultimate objective is to achieve the ability to non-invasively read from and write to the human brain.
The fantasy of the non-invasive write-head
On the podcast, Huberman argued that the current AI arms race is a quest for dynamic modulation of neural activity. He envisions a world where we no longer rely on blunt tools like caffeine to manage our internal states. Instead, we would use non-invasive technologies like ultrasound or light to dial up motivation with surgical precision. Huberman suggests that within 7 to 12 months, we will see the release of technologies that can measure rapid eye movement directly.
From an infrastructure perspective, the read side of this proposal is already shipping. We have been comfortable with indirect readouts like heart rate variability for years. But the write side—the ability to selectively quiet brain areas—is where the spec falls apart. Huberman cites his colleague Eddie Chang, whose work allows locked-in patients to speak, but that requires breaching the skull. Huberman’s alternative is a cap that directs light or ultrasound through the bone to neurons tagged with a viral vector. As someone who has spent years on platform integrity, I have to ask: who gets paged at three in the morning when the viral ANDgate fails? The latency between a lab-scale proof of concept and a shippable consumer product is not 12 months.
When the biological model is a black box
Huberman admitted a striking gap in our understanding of the system these companies intend to regulate. He noted that while we understand the periphery, we still do not know what a thought actually is. More troubling is a study he cited from the memory field. It turns out the temporal sequence of neuronal firing might not matter for memory expression as much as the textbooks claim. If you activate the right neurons in the wrong order, you still get the behavior.
This is a massive implementation problem. If the sequence does not matter, the writing might be easier. But the specificity required to avoid sending the system into haywire becomes impossible to calculate. Huberman noted that neurons are repurposed for multiple circuits. A cluster used for a tennis serve might also be tied to a childhood memory. The strongest opposing case is that the brain is a plastic system that will simply figure it out. But in capacity engineering, we do not build for the best-case scenario. We build for the edge case where the non-specific stimulus triggers a rage response. Huberman hopes for safeguards built into these devices, but benevolence is not a technical specification.
The theatrics of the optimization era
Huberman was refreshingly blunt about the current state of the longevity industry. He called the public discourse a disaster led by clowns focused on theatrics. He predicts a downturn in the over-optimization trend as people tire of complex morning routines. Yet, his own vision of a future involving viral injections to enable brain-writing hats feels like the ultimate evolution of that same theatrics. He argues that technologists like Sam Altman, Mark Zuckerberg, and Elon Musk are the athletes of brain exploration.
This is a classic Silicon Valley move. It treats biology as a stack that can be optimized with better code. But biology has no documentation and no version control. While Huberman is right that the intimacy of technology will increase, the idea that we can safely write to the brain non-invasively in the near term is sellable but not shippable. We are moving toward more specificity. But the biological hurdles mean the cyborg line is much further away than a 12 month product roadmap suggests. Huberman’s prediction that AI labs will become biotech firms is likely correct. But they will become biotech firms because they will be mired in the same grueling validation cycles that define the industry. They will discover that the human brain is the one platform that cannot be disrupted by a weekend sprint.