Jun 17, 2026 · 1h 16m · latent-space
🔬 The Limits of AI in Science - Why We Need Self-Driving Labs — Joseph Krause, Radical AI
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Radical AI CEO Joseph Krause explores why pure computational AI cannot solve materials science without physical self-driving laboratories, discussing how closed-loop robotic experimentation, active learning, and concurrent engineering are revolutionizing alloy discovery and commercial manufacturing.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 5.4 Guest teaching 6.1 Guest disagreement 1.4 The hosts pushing back 3.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 45:00 Dismissing the compute-constraint narrative
Joseph forcefully dismisses the prevailing tech consensus that AI progress is compute-limited, asserting that materials science is purely experiment-constrained.
Hardest push from the hosts ▶ 41:24 Challenging low-data AI efficacy vs domain expertsBrandon pushes back against the value of Radical's AI models at small data scale, citing cheminformatics history where human experts consistently outperform models.
Biggest teaching moment ▶ 23:17 Dissecting automated vs self-driving laboratoriesJoseph clearly educates the hosts on the difference between automated throughput and closed-loop research campaigns using a Waymo autonomous vehicle analogy.
The host holds their own ▶ 14:15 Proposing parallelized clinical trial frameworksBrandon demonstrates deep domain knowledge from drug discovery to probe whether sequential regulatory qualifications can be parallelized via Operation Warp Speed methods.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| The Fundamental Limits of AI in Materials Discovery | 5 | 6 | 2 | 4 | RJ opens directly with a competitive challenge by citing specific competitors like Lila, Cusp, and Periodic in the crowded AI materials space. Joseph responds collaboratively, explaining why experimental closed-loop data is essential rather than purely relying on computation. | |
| Bridging the Gap from Discovery to Manufacturing | 6 | 5 | 1 | 5 | RJ probes the core thesis using a rule-of-thumb heuristic about problem shifts across orders of magnitude in scaling. Joseph builds directly on this by explaining the disconnect between discovery, characterization, and manufacturing in materials science. | |
| Microstructure, Lab Testing, and Radical AI's Progress | 4 | 6 | 1 | 3 | Joseph walks through characterization tools and physical metallurgy testing metrics, while Brandon lightly steps in to pause the acronym overload for listeners. Joseph details Radical AI's output of 1,200 alloys and 300 novel materials. | |
| Exploring High-Entropy Alloys and Concurrent Engineering | 5 | 5 | 1 | 4 | RJ asks whether Radical AI is merely doing combinatorial parameter optimization or expanding scientific frontiers. Joseph explains the unique properties of high-entropy alloys and cites SpaceX's concept of concurrent engineering. | |
| Overcoming Qualification Bottlenecks and Material Constraints | 7 | 6 | 2 | 5 | Brandon draws sharp analogies to pharmaceutical clinical trials and asks if materials qualification can be parallelized like Operation Warp Speed. Joseph educates on the differences in aerospace regulatory standards and critical mineral supply chain dependencies like hafnium. | |
| Human-in-the-Loop Intuition and Tool Vendor Challenges | 4 | 6 | 2 | 3 | RJ asks how humans remain in the loop and whether legacy tool vendors are trying to protect proprietary data moats. Joseph explains how PhD metallurgists train the AI intuition and describes the resistance from instrument providers lacking modern APIs. | |
| Differentiating Automated Labs from True Self-Driving Labs | 5 | 7 | 1 | 3 | Brandon notes that automation in bio-labs often fails to speed up research workflows. Joseph delivers a crisp conceptual distinction between automated labs and true self-driving labs using a hands-free driver versus Waymo comparison. | |
| Focusing on Vertical Integration and Semiconductor Opportunities | 5 | 6 | 1 | 4 | RJ and Brandon press Joseph for concrete, high-leverage application spaces and realistic commercialization timelines. Joseph highlights semiconductor interconnect barrier layers and defense applications, acknowledging a multi-year validation cycle. | |
| Active Learning Loops and Expanding Alloy Composition Space | 6 | 6 | 2 | 5 | RJ compares active learning iterations to compounding hallucination errors in AI coding, and Brandon suggests AI exploration might just be high-temperature sampling. Joseph shows how the system navigates completely unmapped elemental alloy spaces uninhibited by human cognitive biases. | |
| Experiment Constraints, Parallel AI Learning, and Small Data | 7 | 7 | 2 | 6 | Brandon compares Radical's low data volume to cheminformatics regimes where human chemists outpace models. Joseph pushes back by citing DARPA's MOCK program benchmarks and explaining that materials discovery is constrained by experimentation rather than compute. | |
| Why Materials Lack an AlphaFold Equivalent | 6 | 6 | 1 | 4 | Brandon cites Heather Kulik's assertion that there is no AlphaFold for materials and articulates the microstructure mapping challenge. Joseph agrees and explains why downstream physical processing makes one-shot generative predictions impossible. | |
| Capturing Tacit Knowledge and Lab Engineering War Stories | 4 | 7 | 1 | 2 | Joseph shares vivid war stories about reverse-engineering proprietary hardware software and learning tacit knowledge from a 35-year 3M veteran. He highlights how interdisciplinary mechatronics and software form a massive defensive moat. | |
| Why Now for SDLs and Competing with China's R&D | 6 | 6 | 2 | 5 | Brandon brings up China's state-backed speed in end-to-end materials manufacturing. Joseph acknowledges the competitive pressure and outlines a defense strategy centered on public-private partnerships, national lab data, and high-throughput autonomy. | |
| Transforming Tool Infrastructure and Advice for MLEs | 5 | 7 | 1 | 2 | Joseph advocates for rebuilding lab instrument software around autonomous agents rather than human GUIs, and encourages ML engineers to bring their specialized domain knowledge to science. He also details Radical's open-source thesis. |
Statements from this episode (36)
Krause: Physical synthesis and testing are the only ground truth in materials
“In materials, the ground truth is the material itself. You have to be able to make it, you have to be able to test it and characterize it, and then you have to really, at one point, be able to see if it can go into a real application if you're going to have it…”
Krause: Alloy performance is primarily determined by manufacturing and post-processing
“So much of what dictates the performance of those alloys is actually in processing. How do you manufacture it? What techniques are you using post processing and manufacturing that push performance or change performance?”
Krause: AI models cannot qualify new aerospace alloys without physical experiments
“A model can't figure out your way through the qualification pipeline for a new alloy for a jet turbine. You have to do experiments to do that”
Krause: The real opportunity in materials AI is closed-loop manufacturing integration
“That is what we think the true opportunity is for AI and autonomy in materials is linking those two together in a fully closed loop system.”
Krause: Radical AI made 1,200 alloys in six months, with 300 novel
“We probably made 1200 alloys in the last five or six months. 300 of those alloys are new novel, never before seen in literature. And I'd say probably 10 of those alloys had performance that Has got us very excited on where they're going to be in the industry.”
Krause: Heavy industries have used the same alloys for 50 years
“And there has been, you know, for the past 50 years, the same alloys used in all these industries. And the reason why is these long discovery timelines we talked about.”
Krause: Commercial airplanes still use alloys from the 1950s to 1970s
“The alloys that are in the plane I flew here on, the 19 fifties, 19 sixties, 19 seventies, they might be coated with some CMZs from the late 19 nineties.”
Krause: Aerospace alloy qualification is typically a ten-year process
“Typically a ten-year process today, you have to make a number of different ingots of material and run these kind of standardized tests on them to prove your material is usable in those systems.”
Krause: Hafnium prices surged 10x to 15x due to Chinese supply dominance
“Hafnium, 10 to 15 X in price, because China owns a majority of the supply chain, things like refractories, tampium, niobium.”
Krause: Common aerospace alloy C-103 contains roughly 10% hafnium by weight
“It's about 10% weight or weight percentage of hafnium in C one or three, which is a very common aerospace and space alloy that's used today.”
Krause: Radical AI developed a hafnium-free alternative to C-103 aerospace alloy
“We've worked on that problem specifically, and we have successfully done that.”
Krause: No AI model can one-shot materials for iPhones or Starship
“And this is what's so hard is there is no one model that can one shot a new material that ends up in your iPhone or that ends up on Starship.”
Major instrument vendors previously refused to give Radical AI data access
“A few very big tool vendors were Not too excited about self-driving labs two years ago. They were not jumping to give us, even with payment, access to the software and pulling the data. That tone has now changed.”
DeepMind, Microsoft, and Meta are building or using physical science labs
“You see people like Google DeepMind, Microsoft, other places like Meta, either building their own lab or running experiments at someone else's lab to get that data back.”
Krause: Automated Labs Run Experiments, Self-Driving Labs Run Entire Campaigns
“There's a difference between an automated lab and a self-driving lab, right? An automated lab does experiments for you, automated, without humans, and that high throughput, and that, that can be very effective. A self-driving lab runs research campaigns for yo…”
Krause: Radical AI's Robots Autonomously Transfer Samples Across Lab Tools
“The same way a human scientist would come in, you know, look at the results, take the sample out and go to the next one. Our robots do that today.”
Krause: Radical AI Custom-Built a High-Throughput Alloy Synthesis Tool
“We custom build a tool with a third party to do alloy synthesis and high throughput. That, that's, Built to do alloys. That's not built to do ceramics or polymers or any other material system today.”
Krause: New semiconductor interconnect materials could deliver 2x to 10x+ efficiency gains
“I think you could in, in the near future with some of the systems that have been recommended today, you would see like a two to five X generally, especially when you think about integration in these materials, you need barrier layers and there's all these inte…”
Krause: Novel alloys will reach defense and space systems in 3-5 years
“I would say something on the alloy side, like aerospace, we feel good opportunity three to five year timeline. Yeah, correct. I think it'll be an application, not like manned flight. Like, I don't think it'll be a jet turbine because of the constraints there w…”
Krause: Radical AI generates all candidate materials via AI
“For generation, all of our materials are generated by our AI scientists today.”
Krause: Radical AI models explore previously unpublished alloy families
“And then we have a second overlay on that chart of where our AI scientist has gone. And it's moved into elemental families or alloy families. No one has ever published on before.”
Krause: DARPA and GE Aerospace synthesized 500 alloys in 12 months
“The largest alloys program was the mock program. It was run by DARPA NGE Aerospace. They did 500 alloys in about 12 months. They did a bunch of kind of AI and simulations on the front end of that, and then they synthesized 500 new alloys in that whole year.”
Krause: Materials industry is constrained by experiments, not compute
“We're not compute constraint in the materials industry. We're experiment constraint.”
Krause: Understanding material microstructure does not guarantee manufacturability
“Just because you understand the microstructure, just because you see and can predict crack propagation does not mean you're necessarily going to perfectly nail manufacturing.”
Krause: Lab synthesis is not material discovery until deployed in products
“The second you synthesize it, milestone.
The second you characterize it, milestone.
That is not a new discovery.
We count a new discovery when you pick up your phone and there's a new material sitting inside of it.
That I think is a fair claim on new discovery…”
Krause: Large Alloy Manufacturers Are Bearish on AI for Science Hype
“We do talk to a lot of companies in our field who make materials at scale in, in the alloy space who are thinking about this, and they look at it from a different lens. You know, they're not all hype on AI for science, and actually I'd say a lot of them are ki…”
Krause: Autonomous Labs Require Deep Integration, Not Just Robotic Arms
“It's not about a robot in front of a tool. Go ahead, put a robotic arm in front of a tool and then watch what happens. Everything else I just talked about will come the second you do that. Now we feel so much farther ahead from the industry on really running s…”
Krause: Science will experience a foundation model breakthrough within 2-3 years
“I think we're going to have that for science over the next two to three years. I do. Once discoveries start coming out and this field continues to mature even more.”
Krause: Radical AI enables one PhD to run 10 simultaneous alloy campaigns
“We can have one PhD in metallurgy or alloys run 10 campaigns at a time. When I was in a PhD, we had 10 scientists focused on one campaign. One research problem at a time.”
Krause: China will beat the US in R&D without automated self-driving labs
“That's how I think we can compete. That's the only way we can compete. I think if we want to move forward, if we do not do that, then they will continue to win because they will outpace us on cost and they will outpace us on people.”
Krause: Scientific tool makers should rebuild their stacks for robots
“Their tools are built for humans. They should build them for agents and robots.”
Krause: ML engineers in science should specialize instead of switching fields
“Bring the specialization and lean into your expertise. Don't shy away from it. Don't try to become a material scientist. Be an MLE that works in material science.”
Krause: Radical AI's Matrix VLM boosts scientific reasoning by 5-16%
“Now we have actually seen this, and you can go read the publication, it's on archive, where the public data set that we used actually is showing improvements, like five to 16%, I believe, on general scientific reasoning.”
Krause: Most AI models will be open source in five years
“We actually think in five years, most models will be open source.”