Radical AI's Joseph Krause on Using AI for Scientific Discovery Instead of Small Problems
EO Korea
Joseph Krause, co-founder and CEO of Radical AI, argues that the AI industry is mostly aiming too low. In his view, too many companies use a transformative technology to build software tools and incremental products, when the hardest and most consequential problems are in scientific discovery. Radical AI describes its goal as building "artificial general intelligence for scientific discovery," starting with materials science. In this interview, Krause explains how his path through the National Guard, graduate research, the Army Research Lab, and venture investing led to that thesis, and how the company's culture is built around taking on problems that may fail.
The Problem: A 150-Year-Old Discovery Process
Krause describes the company's aim as disrupting a scientific process he says is about 150 years old, and doing so alongside some of today's largest materials companies. He does not see the core challenge as improving existing systems. At Radical AI, he says, those are called optimization problems, meaning 1, 2, or 5% improvements to something that already exists. He believes the technology world is already good at those. What remains very hard is novel discovery, and he says that is where most hard problems lie.
Materials matter, in his account, because many of the world's most important industries depend directly on materials R&D. He names automotive, aerospace, manufacturing, defense, climate and energy, and semiconductors. He says materials processes typically take more than ten years and cost a very large amount of money to move from a new discovery to a material system at scale. Radical AI's premise is that AI and autonomy can change this process and remove materials as the main barrier to progress in those industries.
He gives one specific performance claim. Radical AI's AI agents index millions of scientific publications and learn what a field has already done. He says this lets the process run about 370 times faster than a human scientist operates today. The interview does not explain how that figure was measured or under what conditions.
Early Influences: His Father and the National Guard
Krause credits his father with a guiding principle: pursue something you can be the best in the world at and that you truly love. He says his father emphasized being deeply engaged in the work, not just happy with it. Krause says the question "Where can I make an impact?" has guided his decisions ever since.
As an undergraduate, he wanted to attend a U.S. military academy because of its prestige and the chance to serve. He then met someone serving part-time in the National Guard. Krause explains that the Guard is a reserve component of both the Army and the Air Force. Members drill one weekend a month and train for two weeks in the summer to stay current, and they can be deployed. He saw this as a way to keep studying while serving in his free time, so he enlisted in his junior year of college.
Basic training stood out to him because he was surrounded by strangers from very different backgrounds. He lists recruits from New York, Pennsylvania, Georgia, Maine, the Midwest and western states, San Diego, and Silicon Valley. He says they all shared one goal: building a stronger military to protect and defend the freedoms America seeks to promote around the world. Krause calls this the first time he was forced to take himself out of the equation and put the mission of his team, the battery he served in, ahead of himself. He describes it as a deeply formative lesson, and it anticipates the mission-first hiring philosophy he later describes at Radical AI.
He finished his senior year after basic training, attended further training, and then began graduate school at Rice University.
Why Rice, and the Question of What Came Next
He chose Rice mainly because of an undergraduate research symposium where he presented his work and won the award for best presentation. The prize let him meet graduate students and professors in the program. Their research had a common theme: problems that could have a large impact and that they were trying to move into real use. He says no other program he interviewed with gave him the same sense of focus on the future.
He continued serving in the Guard during graduate school. He says he still did not know what he wanted to do afterward. He was unsure about becoming a professional scientist and had considered fields such as patent law. So he used his time in graduate school to study where he could make an impact.
The Army Research Lab and the Gap Between Science and Application
While serving in the Guard, Krause also worked as a scientist at the Army Research Lab. He describes it as a corporate-style research lab where Army scientists try to push new research to higher technology readiness levels, or make the science more usable for the Army. His own work involved thinking about electrons and hydrogen atoms and developing novel 2D materials for various technologies.
Commanders regularly toured the facility, and researchers had to explain why their work mattered to the Army. Krause says the two groups understood "relevance" differently. The commanders thought about future conflict and how to strengthen the force, while the scientists thought at the level of atoms and materials. One day he had to explain how the lab's work could eventually affect the future of warfare or technologies the Army would want. He says that was when he realized what he wanted to do: move from fundamental research to commercializing science, so it could go into products and make a difference. He concluded that the best way to push technology forward in a new way was through startups, which he calls, in his opinion, the best place to do that today.
Learning From Kevin Ryan and the "Bias to Action"
Krause knew he wanted to build a company but did not know how. He decided to find entrepreneurs and investors who had built companies and were now investing in new ones. He focused his search on New York and met Kevin Ryan. Krause describes Ryan as a prolific New York entrepreneur who built and ran DoubleClick, took it public, and then founded AlleyCorp, an incubation studio and early-stage investment firm.
Krause told Ryan that if he wasn't investing in materials, he wasn't investing in the future. Ryan replied that this was a big bet and invited Krause to come prove it. Krause took a leave of absence and moved to New York a week later for a six-month internship to test whether this career suited him.
The main lesson he took from Ryan is a "bias to action." Strategy can work, Krause says, but it often leads to a dead end when no action follows. In his view, the best entrepreneurs simply get things done, watch the results, and keep iterating. He argues that startups are already hard, and the way to make them easier is to do more than anyone else in your space, with more information than anyone else has, and with a thesis no one else has come up with. He calls bias to action the most important thing he learned from Ryan.
The Founding Question: "Why Is No One Using AI to Cure Cancer?"
Krause and his co-founder Jorge were both investors at AlleyCorp and sat next to each other. Jorge was studying AI at a fundamental level, reading papers on new architectures and approaches to machine learning. One day he told Krause that he was convinced AI would change the world, but he didn't understand why everyone was picking low-hanging fruit and solving small problems. He asked why no one was using AI to cure cancer.
Krause replied that he didn't know about curing cancer, but it pointed to a real opportunity. The two spent about a month and a half reading hundreds of papers across fields where they thought AI could apply. Krause says he eventually concluded that materials science, with its fragmentation, slow pace, and lack of progress over short timelines, could be a strong fit for AI.
They then read every paper they could find at the intersection of materials science, AI, and, as they discovered, robotics. That research led them to their third co-founder, Gerbrand Ceder, who had built an autonomous scientific lab at Lawrence Berkeley National Laboratory. Krause says the three shared a vision of science moving from a human-driven process to one driven by AI and autonomy. They agreed that science would go this way whether or not they started Radical AI, and that it had to be them who built the company.
How AI Changes Discovery: Indexing, AlphaGo, and Inverse Design
According to Krause, AI's main advantage in this field is its ability to index information. His favorite example is AlphaGo, the Google-built AI that played the world's best Go player. He highlights the famous move that observers first thought was a mistake. In his telling, AlphaGo had indexed so many games that it made a move no human had considered.
He applies this to science by arguing that its limits come from the human brain: how many papers a person can read, how many things they can simulate, how many experiments they can run, and how they can combine all of that into a new hypothesis. Putting AI and autonomy at the center, he says, removes this bottleneck. A system can read millions of publications, simulate billions of materials, test thousands of them, and connect all that information in real time.
The result he envisions is "inverse design." Today, he says, researchers make materials and then look for problems they can solve. With inverse design, you start from the hardest problems and design materials to address them. Krause says human scientists either cannot do this today or can only do it over very long periods, while AI and autonomy can do it quickly and efficiently.
He stresses that this is not meant to replace scientists. The goal is to give them tools to work across many experimental processes, so they spend their time deciding what material to build or what problem to solve next rather than on the monotonous parts of fundamental research.
Raising the Pre-Seed Round in About 45 Minutes
The company was incubated within AlleyCorp, and Ryan had said AlleyCorp would invest in that setting. The founders knew they needed a lot of capital. Krause says materials science is a hard business, and Radical AI is building a full-stack solution, so they wanted an unusually large pre-seed round.
They prepared a pitch deck of about 100 pages. It covered why the timing was right, where the technology stood, and why an interdisciplinary approach was essential. When they presented it to Ryan, Krause says, Ryan told them they would not raise money anywhere else because he wanted to provide all of it. AlleyCorp was the only investor, and Krause says they raised the pre-seed round in about 45 minutes. The three founders then began recruiting and building the company.
Hiring for Mission, Not Jobs
Krause says the company cares deeply about its culture. Every candidate gets the same message: if you are looking for a job, this is not the place for you; you are very intelligent and can find work elsewhere, but if you are looking for a mission, you should come here. He says everyone at Radical AI deeply believes in the mission, and culture is what keeps that alignment.
A central part of that culture is not fearing failure. Krause says failure is part of discovery and learning, and the company pushes aggressively and relentlessly to rethink problems from first principles and build a connected system that can discover new materials.
The 51% Rule
Radical AI uses what Krause calls the 51% rule, which he believes SpaceX started. When you are 51% confident in a decision, you make it. His reasoning is that debating decisions that could be made quickly wastes time and reduces the company's effectiveness, and it is better to decide and see the result.
To judge whether they are at 51%, the team considers two things: how big the decision is, and what happens if it is wrong. Large decisions that could define a year or a decade still get extensive research, conversation, and debate. Everyday decisions can be made with confidence more quickly. Krause says asking what the worst case would be usually shows whether you are already at 51% or nowhere near it, and the company uses this as a checkpoint for everyone.
He adds that the rule is not about speed for its own sake. It means making decisions you were going to make anyway, just earlier, and accepting that you will not always be right. In his view, whether you deliberate for two weeks or two years, there is still a chance the decision is wrong.
Failure as Learning
Because the company works this way, Krause says failure is built in. Anyone who joins Radical AI will fail at something. The company does not label people or projects as failures. It treats these outcomes as learning and a chance to rework a process they thought they should build. He says this pushes them further into asking "why" and reasoning from first principles, and that trying things that don't work, learning, and trying again is simply normal there.
Connecting the Dots
Krause cites one of his favorite quotes, Steve Jobs's idea that you can only connect the dots looking backward. He applies it to his own path, especially his time in materials research. He did not know then where that experience would lead. Looking back, he considers it essential that he lived through the frustrations of materials research as it works today.
He says he did not care what came next, whether it would work, or whether he would end up a scientist. What mattered was continuing to push forward and search for an answer, which he believes is how he found it. He reminded himself every day that the work would lead to something, and he says you must never give up in any situation.
A Company Meant to Outlast Its Founders
Krause closes with the company's long-term aim. Radical AI wants to remove materials as the barrier to innovation in the industries that depend on them. For that reason, he says, the founders do not see an end point for the company. They expect it to last 100 or 200 years and to far outlive him and his two co-founders. If the technology succeeds, he says, it could create a world that is not considered possible today.
AI is so incredible. This technology is going to change the world. What I don't understand is why everyone is picking low-hanging fruit and solving small problems. We saw a bunch of AI companies in software or, you know, normal SaaS-based businesses. Why is no one using AI to cure cancer?
We are trying to disrupt the process that is 150 years old with some of the biggest companies today in materials. We want to reinvent the way we do the scientific process. We don't struggle with driving one, two or 5% improvements in a current system. What is incredibly challenging is novel discovery. The most important areas are the most challenging. I think those are the things people should work on because if you succeed, you will fundamentally reshape human trajectory.
Every single person that we hire into Radical AI, we ask the same question. If you are looking for a job, this is not the place for you. You're incredibly intelligent. You can go get employed other places. If you are looking for a mission, then you should come work here.
My name is Joseph Krause. I'm the co-founder and CEO of Radical AI. At Radical AI, we are building artificial general intelligence for scientific discovery starting in material science. The material discovery process is challenging today. Processes in materials just take an incredibly long time, typically 10 plus years. Our AI agents will index millions of scientific publications and learn what a field has done already. And so this process can happen at an incredible speed in comparison to a human scientist, about 370 times the speed we operate as a human scientist today.
I just always was taught and really wanted to be like my dad in pursuing something that I knew I was the best in the world at and that I truly love to do. My father, he always instilled that it's not just about being happy about what you're doing, but much more engaged in what you're doing. Find the thing that you can be better at than anyone else in the world. Where can I make an impact? That for me has always been this guiding light into what I want to do and how I want to do it.
And so when I was an undergraduate, I actually bumped into someone who was serving part-time in the National Guard. I really wanted to go to a military academy in the United States. They're incredibly prestigious. You have the opportunity to serve the country. So I thought that this would be for me. And the US National Guard is a reserve component both of the Army and the Air Force. You pretty much go to drill one weekend a month and then you do two weeks in the summer of training to actually make sure you're up to code with your military training, and if you ever get deployed down range. I was like, well, this is amazing. I can continue to study, but I can also serve in the military in my free time, is the way I looked at it. So, I enlisted in the military my junior year of college.
Basic training is an amazing experience, right? Because you're with a bunch of people you've never met in your life. Meeting people from so many different walks of life. I met people from New York, Pennsylvania, Georgia, Maine, and the East Coast through to the Midwestern or western states all the way through to California, coming from San Diego or Silicon Valley. And everyone was there for the same mission. They were there for the same goal of building a stronger military to protect and defend the freedoms that America likes to propagate throughout the world. That was not only inspirational but was the first time I was forced to remove myself from the equation and put the mission of the team, which was the battery that I was serving in, before yourself, and that was an incredibly insightful lesson. And so I went away to basic training and then came back and finished my senior year and then went away to training again before starting graduate school at Rice.
So for Rice there was one specific reason. I had gone to this undergraduate research symposium, which I presented my work at, and I won that symposium, and in winning the symposium, the best presentation, I got to sit with a bunch of both graduate students and professors inside the program and ask them about their research and what they were working on, and there was a concurrent theme through that. They were all working on problems that could have a big impact, actual things that they were trying to transition at one point or another. I interviewed at a lot of graduate programs when I was thinking about where to go to school, and none had felt as strong as I did when I met the faculty and students at Rice and this kind of focus on the future and what they were trying to build. So, I was doing them concurrently, graduate school and the military at the same time. But I didn't know what I wanted to do after graduate school. I didn't know if I wanted to be a professional scientist. I had thought about fields like law, patent law. So in graduate school, I did this deep dive in understanding where could I make an impact.
While I was serving in the National Guard, I was also a scientist at the Army Research Lab. It is this corporate research lab that the US Army has, a bunch of scientists working in it to try to push novel research up to higher technology readiness levels or make that science more approachable for the army. We used to have these commanders who would come through and tour the facility. We would have to tell them why the work we were doing was relevant to the army, and their relevance was not the same relevance to us. They were thinking about future conflict, about how to bolster the force or make it stronger. We were thinking about electrons and hydrogen atoms and thinking about making new novel 2D materials for different technologies. And so there was this disconnect between leadership at the army and the scientific perspective that we were driving. And there was this day where I had to actually explain why what we were doing in the lab could eventually impact the future of warfare or future technology that the army would want to use. And that's when I realized that that is exactly what I wanted to do. I wanted to transition from this fundamental area of research driving our understanding of science to actually commercialize science. Science that could actually go into products and make a difference. And that was where things started to come together, that the only way to really do this is to actually push technology in a novel way, in which case startups are, in my opinion, the best place to do that today.
Okay. I know one day I want to build a company, but I have no idea how to build it. And so I said, let me find entrepreneurs and investors who have built a company in the past and now are investing in the companies of the future. And so I looked up a bunch of venture capitalists, entrepreneurs, and really wanted to be in New York. Actually, I tailored my search to New York and I bumped into Kevin Ryan. Kevin Ryan is a prolific entrepreneur in New York City. He built and ran DoubleClick before taking that company public. And then he came out of that and started AlleyCorp, which is both an incubation studio as well as an early-stage investment firm. And when I talked to Kevin, I had told him, if you're not investing in materials, you're not going to invest in the future. And he said, "That's a big bet. Why don't you come prove that out?" And so I took a leave of absence. I moved to New York a week later, and I had a six-month internship to see if this was the field for me, if this was the career for me.
Kevin had taught me early in my career that you have to have a bias to action and just get things done. Strategy can be effective but most of the time will also lead to a dead end with no action involved. The best entrepreneurs have an immense bias to action. They just get things done, and in doing so watch the results unfold in front of them, and then they continue to go through their process. Startups are already hard enough, and the way to make them easier is to do more than anyone in your space will do, with more information than anyone in your space has, with a thesis that no one else has been able to come up with. That is truly what can drive good value. And I think a bias to action is the most important lesson I've learned from him.
One of my other co-founders and I were both investors at AlleyCorp. And Jorge was looking into AI technology as every good investor was, but really at a fundamental level. He was reading the publications coming out on novel architectures and novel approaches to machine learning and really what the impact of the technology was. And he came over to me one day, we sat next to each other in the office, and said, you know, AI is so incredible. I am actually convinced this technology is going to change the world. What I don't understand is why everyone is picking low-hanging fruit and solving small problems. Why is no one using AI to cure cancer, was the question he asked me. And I said, well, I don't know about curing cancer, but that's a really good opportunity. And so we spent the next month and a half reading hundreds of research papers on all the different fields that we think AI could be put into. And finally it dawned on me that, well, material science has problems with fragmentation, slow-moving ability, you know, lack of progress over short timelines. This might be a great area for AI to be implemented in. And so we took another deep dive and we read every publication we could find at the intersection of material science, AI and, as we found out, robotics, and that led us to our third co-founder, Gerbrand Ceder, who had built an autonomous scientific lab at Lawrence Berkeley National Lab. And together with him, us three really saw and came up with this vision for where science is going to go, where AI and autonomy are going to build a new paradigm of scientific discovery, driving us from a human-driven to an AI and autonomy-driven process. And we were all aligned that whether we started Radical AI or not, this is the way science was going to go, and it had to be us who were going to build this company. And so all three of us agreed, and that was the formation story for Radical AI.
The most important areas are the most challenging. I think those are the things people should work on. I think if you look at technology today, we don't struggle with what we call optimization problems at Radical AI, driving one, two, or 5% improvements in a current system. We're actually quite good at that. What is incredibly challenging is novel discovery. And that's where most hard problems lie. And when us three wanted to form the company and we were going to build Radical AI, we had this opinion that if you want to impact the most important industries in the world, automotive and aerospace, manufacturing and defense, climate and energy, semiconductors, all of them are a direct result of materials R&D. But these processes in materials just take an incredibly long time, typically 10 plus years, and an exorbitant amount of cost to go from a novel discovery to a scaled material system. And this was the exact place that we could make a massive impact. We think AI and autonomy are going to drive change in the process, and in doing so can actually unlock and remove materials as our biggest barrier to some of our most important industries. And I think if you are aligned on that mission, then the problems that you want to solve are naturally going to be hard, because if you succeed, you will fundamentally reshape human trajectory. And that is an incredibly impactful thing to be able to do and also incredibly important to the future of the world.
We're going to leave AlleyCorp and we're going to raise money, of course, to start the company, and AlleyCorp, as I mentioned, likes to incubate companies, and we were incubating this company inside the AlleyCorp umbrella. So we knew, Kevin had told us, they are going to put some money in in this incubation setting and then we'll roll out from there. And we knew we needed a lot of capital. Material science is a hard business to build in, and we are building a full-stack solution, which requires a lot of capital. So we knew we wanted to raise a fairly large pre-seed round. And so we went to Kevin and said, "Look, we're going to start this company." We put together a whole 100-page pitch deck on why we think the opportunity is now, where the technology is today, why an interdisciplinary approach is incredibly important to solving the problem.
The reason AI is so impactful in this space is its ability to index information. The best example that we always like to give is AlphaGo. It was this AI that was built by the Google team, and it played the best player in the world in a game of Go. And the best part of the story is this infamous move it makes where everyone watching thinks the AI makes a mistake, and in reality it has indexed so many games of Go, it made a move the human brain has never thought of before. And when we take that to science, all of our limitations in science come from the human brain. How many papers can we read? How many things can we simulate? How many experiments can we run? And then how do we tie all those together to make a new hypothesis on what we want to build? But if you bring AI and autonomy to the center of that, you unlock this indexing problem that a human scientist feels. You can read millions of publications, simulate billions of materials and test thousands of them, connecting all this information in real time. And so if you can do that, the world that we think we can really build is one that is what we call inversely designed, where we are no longer making materials and then looking for a problem to solve. We are actually taking our hardest problems and inversely designing materials from that. And human scientists can't do that today, or if they can, can do so only in very, very long time frames, whereas AI and autonomy can do that incredibly fast and incredibly efficiently. What we actually think AI can do is not replace that scientist but rather give them the tools and the capabilities to index across a multitude of different experimental processes. So they are spending their time thinking about what next material to build or what next problem to solve and not on going through the monotonous process of fundamental research. That is really where we see the impact of AI for science and how we are building the future of the scientific process.
And to Kevin's credit, he said, you're not going to raise money anywhere else. I want to give you all of it. And so he was the only investor. We raised our pre-seed round in about 45 minutes, which was great, and we were able to get started and start building, and us three started recruiting from that time on and built the company that we have today.
We are incredibly passionate about the culture that we build at the company. Every single person that we hire into Radical AI, we ask the same question. If you are looking for a job, this is not the place for you. You're incredibly intelligent. You can go get employed other places. If you are looking for a mission, then you should come work here. Every person that comes to Radical AI deeply, deeply believes in the mission that we are going after. And the way that we keep that alignment is through culture.
We are never afraid to fail. We actually know failure is a part of the discovery and learning process. And so we will push aggressively, relentlessly to drive technology, to rethink from first principles, and to build a connected system that can truly discover novel materials.
So we adopted a rule very early in the company, I believe SpaceX started it, called the 51% rule, where when you are at 51% confidence on a decision, you just make that decision. And the reason why is you will actually cause more time and less efficacy across the company by debating on decisions that you could otherwise quickly decide on and see what the result of that decision is going to be. When it comes to 51%, there are two different things that we think about: first, how big is the decision, and second, what are the risks if the decision is wrong, and those
two together allow us to actually identify when we are at 51% or not. Of course, we do a bunch of research and have a bunch of conversations and debate around getting to that confidence interval of 51%. But when it is a decision in day-to-day, for example, it's not year-defining or really decade-defining in what we're trying to go after. Well, then we can have confidence to make it quickly.
And when we think about what the other side of that decision is, what is the worst-case scenario that happens, that usually alerts you to whether you are already at 51% or you are nowhere near 51%. And that is a great checkpoint that we ask ourselves and everyone in the company, constantly pushing towards getting at quick decision-making but effective decision-making as well.
You know, 51% is not about just making decisions fast for fast's sake. 51% is making decisions that you are already going to make, just making them earlier, and then dealing with the effect of not always being 100% correct. Because no matter if you take a bad decision for 2 weeks or 2 years, there's still a likelihood that you're not going to be correct.
And because we operate the company that way, failure is inherent. You will automatically fail if you come work at Radical AI in something that you do. We don't view someone as failing or a project as failing. We view it as learning and an opportunity to recreate the process that we just thought that we should build. And so for us, I actually think we go deeper into the essence of asking why and first principles. And we don't view things as failure. It is just normal to try things that don't work, learn from them, and try again at Radical AI.
One of my favorite quotes in the world is, you know, Steve Jobs's connecting the dots. You can never connect the dots looking forward. You can only connect them looking backwards. This is an exact moment where I didn't know where this dot was going to lead. But now when I look back, it was imperative that I went through that experience and lived through the frustrations of materials research today.
You need dots to be able to connect. And while you might not know where this dot is going to make an impact, you never know where it is going to or how it is going to impact what comes next. And so I didn't care that I didn't know what was coming next. I didn't care if it was going to work or not. I didn't care if I was going to be a scientist or not in the real retrospect. What I cared about was continuing to push forward and searching for the answer. And I think that's truly how I ended up getting to the answer. So I just reminded myself every day that this will lead to something. You have to never ever give up, no matter the scenario.
We want to remove materials as the blocker to those innovations. And that's why we don't think there's an end to the company. We think this company will be 100, 200 years old. It will way outlive myself and the two co-founders. This is a company and a technology that, when it succeeds, will truly be able to create a world not today thought possible.
Article published