Dunia Innovations is an AI materials discovery startup partnering with corporations and governments to accelerate the journey from idea to real-world application.
What gaps do you have to bridge to bring AI into real life applications?
One of the big gaps right now is the simulation-to-reality gap. How do you go from a computational material to an actual physical material that is validated for an industrial application? At the end of the day, a material by itself is useless–it is always part of a product or a system with a function we want. A second gap between the laboratory and factory is more hidden: not everything that works in the laboratory works in industrial settings. Some things are not scalable or practical to manufacture. These are the two major bridges we have to build–getting from the computer to the lab and from the lab to real products people can use.
What changes is your robotics-driven Intelligent Research and Innovation System (IRIS) bringing to material science?
There has been a quiet revolution over the last 10 years, enabled by advances and reduced cost curves for robotics teams contributing to some of the most advanced systems in the world. Back in my graduate school days, I was in a program funded by the U.S. Defense Advanced Research Projects Agency, trying to build an artificial chemist–a generative workflow with AI as the brain and robots as the hands. The systems were transformer-based already and we had all of these great algorithms. We tried and failed. Some great papers came out of the process but there was still that gap to making it real for industry and products.
IRIS, our Intelligent Research and Innovation System, is a third-generation platform that takes us a step further toward industry. A combination of AI-orchestrated self-driving labs and automated research systems, IRIS is able to take prompts and deliver new and improved physical outputs. IRIS today is addressing what went wrong in previous iterations and has already discovered a few exciting materials. Some of these optimize existing processes. Others are truly novel with state-of-the-art performance, moving complex objectives forward and removing constraints with drop-ins to existing manufacturing tools.
Why kind of problems does Dunia Innovations address?
We like to unlock bottlenecks to market take-off, and have been working with large, established companies and scaleups alike to validate that IRIS really helps move the needle. We have projects with Valterra Platinum, a mining company in South Africa, and Hitachi Hi-Tech in Japan, who see that materials innovation is about to change dramatically. One collaboration is with Matteco, a Spanish scaleup working on anode materials for electrolyzers.
Electrolyzers contain electrocatalysts that undergo electrolysis, the conversion of free electrons into bound electrons–chemical energy. Water electrolysis splits H₂O into hydrogen and oxygen. The hydrogen generation is easy and has received a lot of attention for industrial decarbonization potential. Oxygen evolution is the sluggish, expensive, difficult and inhibiting part, similar to how renewable energy used to be too expensive and inefficient for mass market EVs until a massive drop in lithium-ion technology costs that also unlocked new use cases. One third of this effect came from scaling–two thirds was the aggregate of many small advances in chemistry. So, while electrolysis might not seem to be right for aviation or container ships right now, we do not know that. Materials innovation might make it cost-effective to decarbonize these massive parts of the economy with one fundamental process.
What would be the ultimate unlock for Dunia Innovations?
Artificial photosynthesis is the dream. Nature took 2.5 billion years to evolve the leaf, using a process every single carbon atom in our bodies has gone through many times. The question is whether we can replicate and magnify this with semiconductors, catalysts, membranes and electrolytes. Some people are really excited about superconducting magnets for fusion but we also have a big fusion reactor in the sky: the sun. Learning how to harvest this energy would have civilization-scale impacts.
We have a waste problem right now. Our whole economy is a one-way street that just generates heat and emits CO₂, neither of which is sustainable. The most sci-fi solution would be a molecular 3D printer powered by photons, and I think technology will eventually get us to this North Star.
We are just still in the 2020s–maybe AI will deliver 2050 technology in 2030.
What is the vision behind the proposed multimillion-Euro GigaLab initiative with the European Commission, and the overall outlook for the future of AI materials discovery?
Dunia has built a strong reputation for results in a hyped-up field, and companies see our technologies as real. Many of our partners are unconventional players, not necessarily large companies already dominating the industry. Some are equipment companies that need better materials. Others are mining companies looking for higher-value applications. We also work with conglomerates that see AI-driven materials discovery as the key to growing faster than ever before. Our algorithms are working in industrial applications. Scaling up comes next because AI systems improve phenomenally with larger. better datasets. Existing simulation-only approaches have had limited success because they lacked feedback from the real world. GigaLab aims to provide large-scale datasets from self-driving laboratories connected to the physical world. This will potentially require billions of trials but yield massive downstream volume, fundamentally changing the way materials science is done.
The need for experimentation is only increasing. We can tackle harder industrial problems instead of toy problems and suddenly have many more ideas worth testing. Bringing a new material to market takes decades and hundreds of millions of dollars. Our technology could dramatically accelerate the process. With all this pioneering and equipment, we still need highly skilled scientists and engineers to build, operate and guide these systems, so I do not believe humans will be replaced anytime soon. The goal is to give researchers superpowers so one scientist has the productivity of a hundred.
