Cristian Ponce wore an Indiana Jones costume while meeting his co-founder Théo Schäfer. A 2023 Halloween party hosted by Entrepreneur First, a startup program where founders are introduced to each other before pitching an idea.
Both were off, Ponce remembers. Schäfer studied with an expert on underwater autonomous robots at MIT and explored Jupiter's moons for alien life at NASA's Jet Propulsion Lab. “Just fools,” Ponce smiled. “I came from Cal Tech, doing bioengineering,” he says, working on E. coli.
The two bond over stories about the brutality of being a lab technician. Ponce (pictured above left) argued against all the manual labor involved in genetic engineering, especially genetic engineering. Minimal laboratory technique involves manually moving liquids from tube to tube with a scientific tube “pipette”.
specialized Attempts to automate the process have so far been unsuccessful because it is expensive and requires special programming skills. Whenever scientists need to change the parameters of an experiment; They allow the programmer to program the bot; Debugging etc will have to wait. In most cases, Easier for one person to use; Cheaper and more accurate.
The company he founded, Tetsuwan Scientific, set out to solve this problem by developing low-cost white-label lab robots.
But in May 2024, the co-founders announced OpenAI's multi-model product launch (they're looking at it). Scarlett Johansson ticked. with the same sound). OpenAI showed people talking to the model.
Tetsuwan Scientific Missing required links. “We're watching the stupidity of language models grow before our eyes, with their capacity for scientific reasoning,” Ponce said.
After the demonstration, Ponce took out GPT 4 and showed an image of a DNA gel. Not only does the model successfully define what the image is; This identified a problem, which was actually an unintended piece of DNA called a primer dimer. It then gives very detailed scientific advice on what causes it and how to change conditions to prevent it.
Ponce stated that while LLM models are already capable of detecting scientific results, “they don't have the physical agency to act on the recommendations they're making.”
The co-founders are not alone in exploring the use of AI in scientific discovery. Robotic AI scientists can be traced back. 1999 with Ross King's robot “Adam & Eve”;But It really kicks ass. off and a series of Academic papers Starting in 2023.
But the problem, according to Tetsuwan's research, is the lack of software that “translates” the scientific intent the experiment is looking for into robotic execution. For example, The robot has no way of understanding the physical properties of the fluid being pumped.
“This robot has no reason to know. If the liquid is viscous. If possible… it will crystallize. So we have to say,” he said. With shocks from RAG, audio LLMs can work with “hard code” stuff.
Tetsuwan Scientific's robots are not humanoid. As shown in Fig. They are square shaped. But they are built to evaluate results and make adjustments on their own, just like a human. It involves building software and sensors so robots can adjust, It helps to understand things like fluid classification and other properties.
Tetsuwan Scientific has La Jolla laboratories, a biotech working on RNA therapeutic drugs. Robots are helping to measure and determine the effectiveness of doses. 2048 Ventures; Carbon Silicon, It raised $2.7 million in a pre-booked seed round led by 2048 Ventures, which included Everywhere Ventures and influential biotech angel investors.
When Ponce spoke of the ultimate destination of this work; An independent AI that can be used to automate the entire scientific method will lighten the eyes of scientists through repeated results from imagination.
“It's the craziest thing we can do. Any technology that automates the scientific method; This is contributing to hyperbolic growth,” he said.
He wasn't the only one who thought so. There are other for-profit organizations working on AI scientists. FutureHouse and based in Seattle Potatoes.