Using Living Brain Cells To Train AI, With TBC's Alex Ksendzovsky
When a neurosurgeon startup founder hooks up neurons to tiny electrodes, you get a novel way to train AI image models, and a brave new world of bioethics.
It seems surreal, but there it is right in front of me: a palm-sized array that looks like a cross between a big silicon chip and a petri dish, with what looks like a thin pink film of gunk at the top.
It’s April, and I’m scrubbed up in the San Francisco lab of The Biological Computing Co., a startup founded by two neurosurgeons in 2022.
Co-founder and CEO Alex Ksendzovksy is demonstrating how, by firing about 4,000 micro electrodes in the middle of the dish, he can make the pink stuff — a mass of living neurons, like the ones in your brain — fire responses on a grid on a big screen above.
The brain cells obviously don’t have eyes. They can’t see. (And, importantly, they’re not sentient, without any blood vessels or ability to feel pain.)
But just like your computer translating zeros and ones into visual images you can see, TBC can feed the neurons sequences that represent the image, then create algorithms from the responses.
Sci-fi in real life. Still, you’re probably wondering: why?
On this week’s episode of The Upstarts Podcast, Ksendzovsky explains. The brain, he argues, is a more efficient, cheaper image processor than the best computer. Technologists know this — they’ve talked about chips designed to more closely represent the brain for decades.
But even better than brain-inspired AI, Ksendzovsky, is biologically-derived AI.
“I have about 500 million years of evidence behind me,” Ksendzovsky says. “So our moat is evolution.”
TBC has started by training AI image and video models improved by the algorithms derived from its neurons that it can sell and use to demonstrate that it’s more than a research project. With one, it’s already demonstrated a 27-fold efficiency gain.
While investors have poured more than $25 million into TBC, at least one told Ksendzovsky to leave the neurons out of their pitch, at least for now. He won’t listen.
On this week’s show, we discuss why brain cells are such an overpowered training tool; the ethics of using living neurons; and the implications of this technology going mainstream.
Plus, Ksendzovksy shares his Upstart Moment: giving up, alongside his neurosurgeon co-founder, promising careers in academia to pursue a startup.
If you’re enjoying the show, please help us grow by tossing us a like and review on YouTube, Apple, Spotify, and forward this to a friend.
Our three big takeaways from the conversation are below.
A MESSAGE FROM OUR SPONSOR
Uncover insights. Take action.
Rippling AI unlocks deep business insights, builds reports instantly, and runs complex actions across your startup — all powered by your live workforce data and permissions.
Startup lesson 1: Turning obsession into product
Giving up a promising career in neuroscience academia to build a startup wasn’t an easy call for Ksendzovsky and his co-founder, Jon Pomeraniec.
But for Ksendzovsky, TBC represents the culmination of a 20-year-old “obsession.”
“ I worked in a lab, one of the first labs growing brain cells on electrodes, and immediately thought that we should be using it to predict the stock market. That led me onto this journey, and at the time, I kind of pitched it to my professor, and he said it was a bad idea, and he was right.”
The scientist didn’t forget the idea, however. “This has been in the back of my mind ever since,” he says.
Ksendzovsky went on to research growing neurons to help study epilepsy and disease, and found his way back.
“These concepts come together and converge in a dish of neurons where we can abstract from the neural signals, just like we do from epilepsy patients; [we can] keep the cells alive for a long time, just like I had to do in my lab.”
They’d worked on a startup before, and met VCs, so that part wasn’t as scary. The key was to feel passionate enough about the idea to drop their careers and move from Baltimore to San Francisco.
“So [TBC] was really born out of this idea that we were obsessed with this concept, and it had to exist. We were finding the best way to get it out to the world, and try to understand it, and try to create it. And it just so happens that Silicon Valley, San Francisco, is where crazy ideas are born.”
2. Winning with today’s tech first
If TBC’s technology takes off, it’s easy to imagine biological computing at a bigger scale: in data centers. As noted by Upstarts reader and my father, a former IP lawyer with expertise in chips, the concept of “biochips” dates back at least to the 1980s.
If you’re following the possibilities that far, it’s only natural to imagine such chips directly in data centers, operating at huge scale.
Ksendzovsky doesn’t disagree. He says TBC is building the “foundation” first.
“ But to get there, it's going to require probably five, 10 years. And a lot needs to be figured out. What we realized is, in the interim, as we're building this foundation, we can really incorporate our findings into current day AI systems, right? And so we can make significant strides on transformer models, for example.”
In that context, TBC’s approach is a smart one for a startup to take: prove your value in the current generation of tech, and build real revenue, while working toward the deep tech, bigger goal.
“I think the pie in the sky, the North Star is real-time biological computing, where we're actually plugging biological networks or neurons into data centers for real-time inference, for things like intuition, for a personalized AI that's constantly learning, just like the brain does. So that's the vision.
It's certainly not something that's possible today, but with the tools that we're building on the way to improving software and, and AI systems,, we're learning more and more towards realizing this world of real-time biological computing.”
3. Drawing the ethical redline
One question I knew I needed to ask Ksendzovsky as soon as I’d seen that demo: does TBC employ staff bioethicists? How does it think about the weighty implications of its technology?
“We talk about this a lot. We think about this a lot, and we interface with bioethicists a lot on this,” he replied.
TBC made some important distinctions with its technology that keep its neurons from “feeling” or functioning like a brain in a sentient way.
“Just to be very clear, what we have in the dish is not sentient, never going to be sentient, does not feel pain, does not have any of the things that, humans, animals, et cetera, have.
It's a tiny, tiny fraction of the brain, what we have in there. They're neurons. Not only is it a tiny fraction, we also remove all the connections from them and reconnect them to serve our purpose, right? There's no blood vessels. There's no three-dimensional architecture to them, right? If you think of what has to happen for you to have the emerging properties in your brain, like consciousness, awareness, sentience, pain, et cetera, it's so different than what we have in the dish that it's never gonna get to that point.”
Limits, Ksendzovsky says, are critical for this area of technology.
“As we continue to drive this technology, we will-- we, we have very clear guardrails of not getting to that point and making sure it doesn't.”
Thanks for clarifying. I had assumed the write-up summarized the interview. I’ll be sure to watch!
This is very troubling; I’m not surprised their founders urged them not to discuss their use of neurons. Using their incredible education and training not to cure human disease and suffering, but to strengthen AI. Wow. A big question I had- you never said what kind of neurons they’re using. Human or animal? And where did they get them? Thanks for reporting on this, Alex!