DeepMind & MIT Alumni Raise $13.5M to Build «Generative Knowledge Discovery» AI
Palo Alto- and Cambridge-based Unreasonable Labs has emerged from stealth with $13.5 million in funding to develop an AI platform designed for “generative knowledge discovery”. It is designed to go beyond current AI tools that mainly search or summarize existing knowledge, allowing the system to connect ideas across fields, propose new scientific hypotheses, test them in simulations, and outline experiments.
The funding round was led by Playground Global, with participation from AIX Ventures, E14 Fund, and MS&AD Ventures. The company was founded by Yuan Cao, formerly a senior staff research scientist at Google DeepMind, and Prof. Markus Buehler, an engineering professor at MIT known for work in computational materials science and AI-driven discovery.
In addition to the founders, the company lists several advisors including physicist Nobel Prize in Physics winner Konstantin Novoselov, MIT biotechnology researcher Robert Langer, and Hugging Face co-founder Thomas Wolf.
Unreasonable’s system combines large language models with neurosymbolic mathematical abstractions designed to represent physical and chemical principles. Instead of relying only on patterns learned during training, the company is building a general framework designed to help AI reason across different scientific fields.
This structure converts scattered research data into an organized network of entities and relationships, allowing the system to analyze information more systematically and trace how conclusions are reached. Scientists can also modify the reasoning rules, making the system’s logic easier to inspect and adjust. This allows the AI to combine ideas in creative ways, a uniquely human capability important for generating original scientific hypotheses and discoveries.
Within this framework, the system is designed to support several stages of scientific research. It can combine information from multiple disciplines into a unified knowledge model, identify connections between distant concepts to suggest new hypotheses, test ideas using physics-based simulations before experiments are conducted, and translate promising results into experimental protocols while incorporating feedback from laboratory data.
According to the company, the approach is designed to address a growing bottleneck in scientific research: the rapid expansion of technical information across disciplines.
The platform is also designed to interact with simulation environments and experimental hardware, creating a feedback loop in which hypotheses generated computationally can be evaluated through modeling and then refined based on experimental data.
Unreasonable reports that it has begun pilot collaborations with industrial partners in areas including energy technologies, materials science, and pharmaceuticals. The company plans to use the new funding to expand engineering and machine learning teams focused on simulation, scientific computing, and AI model development.
Cover image credit Unreasonable Labs
Topic: Biotech Ventures
