Today’s life science automation is only capable of achieving a high volume of experiments, not a high variety. Current lab robots require extensive programming by specialized engineers to replicate protocols, resulting in systems that resemble assembly lines instead of an assistant to scientists. As a result, the applications for these tools remain limited and ineffective for many of the tasks scientists are burdened with.
Lab robots cannot understand their environment, experimental concepts, data, or even scientists’ experimental intent. AI has the potential to transform these robots into tools that communicate more like human scientists, unlocking automation as an indispensable asset in research. While the robotic body already exists, an AI scientist must also have a mind that understands intent and context. Building such a mind is now becoming more achievable.
The development of agents capable of understanding and executing experimental workflows allows for more reproducible science to be done than ever, freeing up human scientists to focus on analyzing data and developing new hypotheses. Creating agents that can understand experimental intent is a foundational step towards AI scientists that can generate and act on their own intent. This once science fiction vision could work to automate the scientific method and exponentially accelerate progress towards new medicines, therapies, and organisms.
Scientists often communicate using high-level, ambiguous instructions that require expertise to translate into explicit actions. Current robotic systems lack this ability. Recent advances in natural language processing, combined with linear programming and rules-based systems, bridge the gap between experimental intent and robot execution. This two-way translation allows scientists to interact with lab robots as seamlessly as with each other, opening the door for automation to be used as more than a brute force tool.
Lab robots require skilled engineers for operation, as they operate with a lack of closed-loop controls. By enhancing these robots with automated calibration, liquid class characterization, and improved collision detection, Tetsuwan enables their reliable use by scientists. These improvements enable lab robots to handle complex, high-differentiation tasks and operate autonomously.
When we were kids, we both watched Astro Boy, a show about a robotic kid who protected humanity from mortal threats. In Japanese, the show is known as “Tetsuwan Atom”. Our parents and the shows of our childhood imbued a sense of wonder that led us to fall in love with science and technology. We chased that sense of wonder to Caltech, ETH, and MIT, looking for how we could best impact the world through technology. Our work at Tetsuwan has given us an answer to this question.
In the short to medium term, AI scientists will resemble assistants that empower human scientists to conduct more experiments with more reproducibility, enabling more breakthroughs in areas like cancer therapeutics, precision medicine, and drug discovery. In the long term, AI scientists will resemble scientists capable of not only executing based on a human’s implicit scientific intent, but creating and executing based on their own intent. This will enable the automation of the scientific method, which will catalyze an exponential rate of discovery and serve as one of the most consequential developments in human history.
While this problem is incredibly multifaceted and varies significantly across disciplines (chemistry, materials, biology, etc.) an increasing number of the associated technical challenges are becoming solvable. There is nothing more important that we could currently be working towards, as the automation of the scientific method reframes everything we ever have worked towards.
