Tetsuwan Scientific, a startup based in San Francisco, is developing artificial intelligence (AI) robotics designed to take on the responsibilities of a scientist. Co-founders CEO Cristian Ponce and Chief Technology Officer Théo Schäfer revealed the company after completing a successful seed funding round in November. The firm’s mission is to create intelligent software that integrates with laboratory robotics, facilitating the full spectrum of scientific discovery—from hypothesis formulation to experimentation and conclusion drawing.
Developing AI-Driven Robotic Scientists
Founded in 2023, Tetsuwan Scientific spent the past year and a half in stealth mode working on its inaugural product: an AI scientist capable of conducting experiments. The startup is now collaborating with La Jolla Labs on RNA therapeutic drug development. More information about the company’s vision and its initial product can be found on its website, though it has yet to release any products for public use.
The company has identified a significant issue within the realm of scientific automation—it currently emphasizes a high volume of experiments rather than a high variety. Existing lab robots often require extensive programming to follow specific protocols, resulting in a production line-like system rather than truly assisting scientists. Tetsuwan Scientific points out that robots lack the ability to grasp scientific intent, which hinders their capacity to independently conduct experiments.
To address this, Tetsuwan Scientific is leveraging advancements in generative AI models, positing that these technologies can bridge the communication gap and enable robots to emulate a scientific approach. The company sees the solution as a dual challenge, necessitating both intelligent software and versatile robotic hardware.
In an interview with TechCrunch, Ponce emphasized the potential of large language models (LLMs) to overcome software hurdles. These models allow developers to convey scientific intent to robots without having to write extensive lines of code. He also mentioned the Retrieval-Augmented Generation (RAG) framework as a means to reduce instances of AI hallucination.
Tetsuwan Scientific is developing non-humanoid robotic entities, which have been showcased on its website. These robots feature a large, square, glass-like structure, designed to analyze results and modify scientific experiments autonomously. Equipped with AI software and sensors, these machines are programmed to understand technical parameters such as calibration and liquid classification.
Currently, the startup is in the early phases of its ambitious journey towards creating fully autonomous robotic AI scientists capable of streamlining the entire scientific process and driving innovation.
