Christian Ponce was wearing an Indiana Jones costume when he met co-founder Theo Schaefer. That was at the 2023 Halloween Party hosted by Entrepreneur First, a startup program that introduces founders to each other before launching their ideas.
The two hit it off, Ponce recalls. Schaefer earned a master's degree in underwater autonomous robotics from MIT and worked at NASA's Jet Propulsion Laboratory searching for extraterrestrial life on Jupiter's moons. “That's crazy,” Ponce says with a laugh. “I came from Caltech and did bioengineering,” where he worked on E. coli.
The two bonded over stories about their monotonous jobs as laboratory technicians. Ponce (pictured above, left) specifically complained about all the manual labor involved in genetic engineering. Low-level research technicians can spend hours manually transferring liquids from tube to tube using scientific syringes called pipettes.
Attempts to automate this process have not gained momentum because the robots that can do it are specialized, expensive, and require special programming skills. Every time a scientist needs to change the parameters of an experiment (which happens all the time), they have to wait for a programmer to program and debug the bot. In most cases, using humans is easier, cheaper, and more accurate.
The company they founded, Tetsuwan Scientific, set out to address this problem by modifying low-cost, white-label laboratory robots.
But then, in May 2024, the co-founders turned their attention to OpenAI's multi-model product announcement (the one that made Scarlett Johansson sound like one). OpenAI showed people talking to the model.
It was the missing link that Astronomical Science needed. “We are witnessing incredible advances in large-scale language models and their scientific inference capabilities,” Ponce said.
After the demo, Ponce fired up GPT 4 and showed an image of the DNA gel. Not only did the model correctly interpret what the image was, it actually identified the problem: unintended DNA fragments known as primer-dimers. We then provided very detailed scientific suggestions about what causes it and how to change conditions to prevent it.
Ponce said it was a “lightbulb moment” where LLM models were already able to diagnose scientific results, but “we didn't have the physical institutions to actually implement the recommendations they were making.” said.
Tetsuwan Scientific's robot AI scientist resembles a glass cube. Image credit: Tetsuwan Scientific
The co-founders are not alone in exploring the use of AI in scientific discovery. The origins of robot AI scientists can be traced back to 1999 with Ross King's robot “Adam and Eve,” but it began with a series of academic papers starting in 2023.
But the problem, according to Tetsuwan's research, was that there was no software to “translate” the scientific intent, or what the experiment was asking for, into execution by a robot. For example, robots have no way of understanding the physical properties of the liquid they are pipetting.
“That robot doesn't have the context it needs to know. It might be a viscous liquid. It might…crystallize. So we have to tell it,” he said. Ta. An audio LLM with hallucinations suppressed by RAG can handle “things that are difficult to hard code.”
Astronomical robots are not humanoid. As you can see in the photo, it has a square glass structure. But it's built to evaluate results and make changes on its own, just like humans do. This includes building software and sensors that allow the robot to understand things like calibration, fluid class characterization, and other properties.
Tetsuwan Scientific currently has an alpha customer, La Jolla labs, a biotechnology company researching RNA therapeutics. Robots help measure and determine dosage effectiveness. The company also raised $2.7 million in an oversubscribed pre-seed round led by 2048 Ventures with participation from Carbon Silicon, Everywhere Ventures, and some influential biotech angel investors.
Ponce's eyes light up when he talks about the ultimate destination of this research: an independent AI scientist that can be used to automate the entire scientific process, from hypothesis to reproducible results.
“This is the craziest thing we can do. Technology that automates the scientific method is a catalyst for hyperbolic growth,” he says.
He's not alone in thinking this way. Other companies working on AI scientists include the for-profit organization FutureHouse and Seattle-based Potato.