Predicting the Earth with Josh Goldman: How KoBold Uses AI to Find Critical Minerals

Hi, listeners, and welcome back to No Priors. Today we're speaking with Josh Goldman, co-founder of KoBold Metals. KoBold is building the world's largest collection of geoscience data and using its AI tools to better identify mineral deposits like lithium and copper, making it a better explorer. KoBold invests over $100 million annually across 70 projects on 4 continents. Josh, welcome to No Priors.

It's a pleasure. Thanks so much for having me.

This is a super interesting real-world business. You run an intelligent mining company. What does that mean? What does KoBold do?

1. The Exploration Opportunity

We explore for minerals. We're looking for lithium, copper, and the other metals that we need to build businesses powered by batteries and AI. We develop AI technologies, and we combine AI with human intelligence to be better explorers and more successful at finding the sources of minerals that we need for these businesses.

Are you both finding them as well as actually doing the mining, or is it only a tool to find these sorts of assets or resources?

That's a central question. Our business is focused on exploration, and it's focused on exploration for a couple of reasons. One is because there's way more value to be created there, and the second is that's where technology can be really differentiating.

The economics of exploration are really quite extraordinary. With a few million dollars of capital, you can create a 100- to 1,000-times return. Exploration's a very old business. Think about gold miners back in the middle of the 19th century. If you can get the right claims, you can strike it rich if you can dig in the right places. It's about where you look and how effectively you can look.

The unit economics of discovery are really extraordinary. The problem with exploration as a business is that the success rate's really low. You have to try many, many different places before you can find something, and the problem keeps getting harder. But that's also the reason why technology is so differentiating.

We're looking for things that are harder and harder to find. It used to be that you could find minerals literally with your eyeballs by walking across the ground and prospecting. A lot of the copper ore minerals that form at the surface are modified by the air and water in the surface environment, turning blue and green, like the patina on the Statue of Liberty.

Anything you can find by traipsing across the ground with your eyes has been found by now, and we need more intelligent ways of looking for minerals in places that are concealed. They're literally underground and concealed by the rocks. Technology is a way to create differentiation and be a much better explorer.

Once we find things, there's a continuum from having a good idea and collecting some rock samples, to finding something underground, to having many different holes and establishing that you've got something continuous, to determining that it's going to be economic to mine, to designing the mine, to building the mine. There's a whole spectrum, and the technology that we use to find resources and define those resources helps set a project up to be a more economical mine as well. So we continue to contribute technology and stay involved in projects as they evolve.

What sort of data are you using in order to actually identify a mine site or a potential site?

2. The Geoscience Data Advantage

There's a huge amount of data. Humans have been collecting data about the Earth for as long as humans have been looking at rocks, and an enormous amount of data is actually in the public domain.

The length scales are very different. Start with the global length scale. What can you know about the entire Earth? You can look at satellite imagery in different colors, and so you can get a sense of the rocks that are exposed at the surface. There are data sets that tell you about the structure of the continents and the ancient continents that collided, where the ancient protocontinents were, and where those crashed into each other a long time ago and formed mountain ranges.

You zoom in and go to another length scale, and you can fly airborne surveys with sensors that can detect the magnetic properties, density, and electrical conductivity of the rocks. You can go out and collect rock samples and measure what they're made out of—all the concentrations of different chemical elements—and likewise for soil samples.

These are standard types of data that are used in the industry, and there's a huge number of these old data sets in the public domain. Most private companies have to disclose their data to regulators. Any place you look, typically a number of other companies have looked there before and haven't yet found anything.

Even when this data is in structured form, it's spread out over tens of thousands of different repositories. There's nowhere you can go where it's all aggregated in one place. You have to do a lot of really hard technical work to get it together, and you have to do a lot of scientific work to use judgment about what this data actually means and whether or not it's fit for purpose. There are all kinds of messy problems with the data.

A lot of this data is unstructured as well. Geologists use a lot of words. There's a very rich lexicon of geological vocabulary for rocks and time periods. There's a lot of text data and reports that are filed by companies, often with regulators, that become public after a period of time. And there's an enormous amount of data in maps of various kinds.

One of my favorite data sets that we use around the world is a set of maps from Zambia from almost 100 years ago. The originals are hand-painted on linen. We got a tip from an elderly geologist about which drawer in the state archives to look in that had this particular collection of maps.

You could never collect data like this again. It's incredibly labor-intensive, and now there are lots of farms and people living there. You can't go traipsing across their ground looking at the rocks. But these observations were made by skilled geologists, and the rocks haven't moved, so there's no expiration date on the data.

You can take data sets like this that provide ground truth and use them for training machine learning models based on modern airborne geophysical surveys and modern satellite imagery. It's the combination of all these many different data sets, different types of data, and the systematic use of structured and unstructured data that's really powerful.

In a pre-KoBold world—or maybe you can just tell us who the largest couple of other explorers are out there—how do you go look for lithium?

Again, you've got this different set of length scales. You start with the Earth and say, "Okay, I'm interested in lithium. What's the recipe for making a lithium deposit?"

What is an ore deposit in the first place? There's an enormous amount of lithium in the Earth's crust. The central problem is that the lithium that's in your driveway is in very low concentration. The lithium that's in the granites that you can see out your window isn't economical to extract. It's too dilute.

A lot of the minerals or metals we're looking for have concentrations in the crust of a few tens of parts per million. The crust is really big, so there are a lot of metals. What we're looking for are those places in the Earth's crust where natural geological processes in Earth's history have gathered up a bunch of metals from a really large volume of rock, moved them, concentrated them, and then redeposited them in a much more concentrated form—more like 1% copper or 1% lithium, or even more than that. Then you can take it the rest of the way to 100% with industry.

That's what an ore deposit is. There are not only lots of lithium and lots of copper in the crust, but actually many, many places where those geological processes have happened, even though they're rare in the Earth as a whole.

The problem is, where are those special places where these natural processes happened, and how can we find them? We talk about exploration as an information problem because the scarce resource is not lithium or copper metal in the ground. It's actually information. The scarce resource is not the ore deposits; it's the information about where the ore deposits are located.

You have to first understand how an ore deposit is formed. You have to know the recipe, and you have to have some ideas about where those processes might have been occurring on the Earth and how they're going to be expressed in the data sets. Then you can marshal the data and start asking questions of it. You can make hypotheses, narrow down on some specific portion of the Earth, and then what you want to do is acquire the land.

I guess another overlay may be the geography relative to the governance of the country and its regulatory ability to actually mine things. My sense is, for example, the U.S. has a pretty diverse range of deposits.

We just don't want to mine certain locations anymore, or certain types of mining. We don't want to do certain types of mining, and so it's a bit more of a regulatory issue in some cases versus whether we can find stuff. Is that a correct understanding, or is it that these things are rare enough and scarce enough that you really have to scour the ends of the Earth to find them?

Regulatory constraints are really important, but at the same time, you can't be too narrow in your initial filter because these are rare enough that you want to put yourself in the place where you have the highest probability of success. You want to start with the best prior that you can, and that way your likelihood of success is going to be much higher. It isn't just a function of regulations.

We consider security of property rights. If we find something, we have to be able to develop it into a mine that is going to produce for decades, or we have to be able to sell it to someone who would do that. You have to be able to rely on the fact that you can continue to own the property for that period and that the tax rates and the royalty rates will be consistent over that period.

Development is challenging because you don't just have regulators. You have lots of different community interests, and these things are extremely local. The US is not monolithic. You have state regulators, and within a state you have many different communities and many different indigenous groups, and this is true the world over.

It's true in Zambia. There are 50 different chiefdoms, and so you have traditional leaders everywhere that you work. Technical success is not very helpful. Success is finding something that is really economic to develop, that either we can develop or we can sell to somebody who can develop it.

If we don't actually have the so-called social license to operate, if we haven't invested in the relationships with the community to be able to build, and we haven't started in a place where that's possible, then we're not going to be successful. These are hyperlocal problems, for sure.

Josh, can you give us a sense of just the scale of the operation for KoBold today—where you are looking, where you own land, where you're drilling, and what you've discovered?

3. KoBold's Global Exploration Portfolio

Absolutely. We operate exploration projects. Basically, the company does 2 things. We find places that are prospective for making discoveries, and then we test our hypotheses by collecting data, collecting rock samples, flying airborne surveys, drilling holes to get samples of rock from below the ground, and developing technology that we use for guiding our decision-making.

Our exploration portfolio is more than 60 projects, and they're on 4 continents: North America, Europe, Australia, and, critically, Africa. We're targeting copper, lithium, nickel, and cobalt, and likely other commodities to come. In all of these cases, we own the exploration rights either ourselves or in combination with a joint venture partner, and we are operating the exploration programs.

Almost all of these are pre-discovery opportunities. They're seeds we've planted. Any of them could become great ore deposits. What we have in Zambia is really an extraordinary deposit. It is the highest-grade large copper deposit that is not yet a mine.

The average concentration of copper in operating copper mines today is about 0.6%. So if you mine 1,000 kilograms of ore, not including the non-ore rocks all around it, there's 6 kilograms of copper in it that you can potentially extract. The Mingomba deposit in Zambia, the core of it, is over 5% copper, and it's very large. That's extraordinary.

The economics are much better because if you compare a high-grade and a low-grade deposit—a 5% and a 0.5% deposit—if they're producing the same amount of copper, they have the same revenue. But the high-grade deposit, if you have 10 times the grade, means you are producing 10 times less rock, at least. You have 10 times less stuff to haul out of the ground, 10 times less waste, and a 10-times-smaller plant.

That means the economics are far better, the capital intensity is lower, the operating costs are lower, and the environmental footprint is smaller. Those are the things that we are looking for. We're looking in a commodity business where everybody sells copper for the same price. It's a global commodity market, and our ability to make money depends on what our margin is.

That means we need to be a low-cost producer, and we want low-capital-intensity assets. So that is the definition of the exploration problem: finding the highest-quality assets. In Zambia, so far, we have a quite extraordinary and really world-class copper deposit.

Can you tell us a little bit more about the technology that you're using? Obviously, you mentioned you're mixing older-school data, modern image-based data, et cetera, and then you have to data-mine it or extrapolate where these potential deposits are. What sorts of models are you using? What approaches are you using? How do you think about, overall, what you're building from an AI and data perspective?

For sure. KoBold's technology is a full-stack system for guiding exploration decision-making. There are dozens of different products that work together, and they fit into 3 themes. The first one is sensors: hardware that we have developed that collects new kinds of data about the Earth.

The second is the data system for taking all the data we're collecting, all the historic data—structured data from many different kinds—and a huge corpus of unstructured data, and getting it all into 1 system so that we can interact with it systematically. Rather than hunting and pecking through this, we can interact with the whole corpus of data at the same time. LLMs and other technologies are very powerful for being able to interact with all of these different types of information.

The third theme is models: dozens of different models for making better predictions about where and how to look. These models operate at many different length scales. There are models trained on satellite imagery or our proprietary hyperspectral airborne imagery, and you've got some rock samples on the ground. We can predict from the imagery what types of rocks we're going to find at the surface and what the properties of those rocks are going to be.

What's really exciting is that it's not just that we have a model for lithium pegmatites or a model for mafic-to-ultramafic rocks that might host nickel deposits. We make a prediction and develop an initial set of hypotheses based on that. Then, when our team gets on the ground, every day that they're in the field, they are collecting new training data.

They're not just going to places where we have high confidence in what the rocks are, because we're not going to learn anything. We're going to places where the models are highly uncertain, and the new training data—a small amount of additional ground truth—can dramatically improve the predictive power of our models.

What happens is that you have geoscientists in the field making observations, and using those observations, we are retraining those models every day and serving new predictions out to the team. You've got this duet of technologists and geologists working together on the same problem.

Is it a process of forming a hypothesis and then validating or invalidating it? I'm imagining, “Okay, at this set of spots in Zambia, I'm going to go 20 feet below the surface, or whatever it is, and I'm going to find this concentration of something.”

Absolutely. Let me give you a whole bunch of examples. One example is: I'm going to go to this location, and I'm going to—pegmatites are the container rocks for lithium deposits. It's just a name for a rock, like a granite or something like that.

We're going to predict that we have these special rocks, pegmatites, that might contain lithium. We're going to predict that there's 1 in this location, and we're going to land on it, sample those rocks, and look at them. That's a prediction we're making at the surface.

We're making predictions in 3D, and we're saying, “Okay, here, I think there is a layer of conductive rocks, and I think those conductive rocks are prospective for hosting nickel, copper, and cobalt. I think this rock layer is going to be intersected between 200 and 300 meters below the surface, and it's going to be highly conductive. It's going to have a distribution for how much sulfur and how much nickel and copper are in it.”

More than that, we're going to say, “The best place to test this set of hypotheses is by putting a hole at this location and drilling it in this direction.” Other times, there's a known layer of rock, and we're saying, “Okay, we think this layer continues out in this direction. Here is a surface where we're predicting this layer is going to be at this depth, it's going to be this thick, and it's going to have this much copper in it.”

You're going to get a probability distribution for all of these at any given point. Those are the kinds of predictions we're making: We collect a piece of information, condition the model on the new data, and serve out a new prediction.

On the third theme—sensors—we use everything that's available today that we can get from a service provider. But most mining companies are not as keen to use new data types or invest in new kinds of technologies. Sometimes we need to go build our own.

An example of this is our hyperspectral imaging technology. There were new imaging chips available that were not yet deployed in the service market, and the mining industry was adopting them too slowly. We built our own hyperspectral imaging system. In less than a year, we had it flying on a light aircraft, surveying areas that we're interested in.

We're using it to get data in 600 colors at dramatically lower acquisition cost and with a much, much faster time to deliver processed images. We're integrating that information with other types of data and using it to make decisions about where to go in the first place, and then how to change our exploration plans while we're in the field.

Was there any tool or dataset that was most crucial for that marquee discovery you made in Zambia? Was there a piece of data that others had overlooked? Was it just looking in that geography, or was it a specific tool?

There is no one piece of data that enabled that, and that's really a critical theme. Often, new technologies are invented in this industry where people think, "This is going to be the silver bullet. It's going to help us find all the ore deposits, or this dataset alone is going to let us do that."

Mm-hmm.

Actually, the data is very high-dimensional. When you can add dimensionality to the data, then you can have improved predictive power. That's the story there, as it is everywhere else.

It's a combination of new analytical methods, the ability to quantify uncertainty and understand the range of possibilities, and critical scientific insights about the way that these ore systems are formed. All of those things in combination are what make it possible. There is no way to isolate the AI from the HI. There's no way to isolate one piece of data that's uniquely powerful.

That's one of the reasons I think innovation has been limited as well. We think, "This new airborne gravity gradiometry invented in the 1990s is going to find all the ore deposits." It doesn't, but it's really powerful. We're really happy when we can get that data. We go collect it ourselves.

These are incremental improvements to predictive power, but it's only possible if you can work with all of these different datasets together in a unified way.

How does a project like this get valued? If you sell it to somebody else or develop it, it sounds like copper is whatever price it is, and you take some risk on that over time. Then there's the cost of operation, based on how concentrated the deposit is and how large it is. Those give you some sort of cash flow model for the business?

4. The Economics Of Discovery

That's exactly right. It's actually really easy to value a natural resource asset like this. They all trade on their present value of future production, which is very knowable. It is much easier to know what a mine is going to produce 20 years from now than it is to know what a SaaS company's sales volume is going to be 20 years from now and how it's going to be priced, right?

I feel attacked.

It is. No, but they're very different kinds of businesses. You think you build a mine that can move, say, 10 million tons of ore per year, and then what you're going to do is dig 10 million tons of ore per year. You're going to dig the highest-grade part first, then the next-highest grade. On average, it's going to produce whatever percentage of copper it's going to produce.

It's very simple, right? The revenue is whatever the commodity price is. The volume is based on the size of the mine you build and how you cost it. The cost is very knowable because you need to know how many trucks you need to move, how much water you need to pump, and what it costs to pump the water. That's all straightforward stuff.

Then you need the capital cost. You're going to build a plant, and these things are big vessels. You have a tank and a crusher, and the mill has some steel balls in it. These are knowable things, and they're typically built. You can figure out what the margin is going to be and see what the capital profile is going to be.

You have to assess what fraction of the copper you can recover. Those are your sensitivities. You might say, "I think we can get—"

Okay.

Ninety percent of the copper. If we can get 92%, the economics are juicier. If we only get 88%, it's a little dilutive.

Those are the uncertainties. Then you discount that according to the risk profile of the asset. What stage is it? How close are you to production? You might demand a higher rate of return if you're in a less stable jurisdiction.

It's quite straightforward. We know with high confidence what the sales volume will be from Mingomba 20 years from today, and that's amazing. There's potential upside if we find more and more resources. One of the things about these deposits is that once you get underground and start mining, you learn more and more about the geology, and you keep finding extensions.

Mines are often designed around the first 20 years. You underwrite an investment based on those first 20 years, but many of these mines operate for decades—often many decades longer: 50, 60, or 70 years—because the resource keeps going, and you can keep adding to it as you go.

It's actually pretty straightforward to understand how these are valued. They're hard assets. There's a property interest, and the market values these accordingly. They all trade on their present value of future production.

How successful are exploration companies in general today? If I start sampling 100 sites and have 100 theses, do I find 1? Do I find 0? Do I find 10? How much better do you think KoBold can be?

This is a key question: What is the success rate in the industry, and how much better do we hope we can do? In the industry, it's gotten 10× worse in the last 30 years because the problem has gotten harder and the industry is slow to innovate.

The way to think about it is not the number of successes. There are studies that will say a 0.5% success rate or something like that, but what actually counts as an attempt is ambiguous. The key resource input is that you have to invest some capital to run an exploration program. You have to put a geologist on a helicopter, go out and take samples, and drill holes.

If you take a portfolio of exploration projects that costs some money—say, $1 billion industry-wide—how many successes will you have? Industry-wide, $1 billion in exploration spend would have produced hundreds of failures but 8 discoveries 30 years ago. Today, it produces fewer than 1 high-quality economic deposit.

That's why exploration in the aggregate is not a great business. At KoBold, we target $50 million to $100 million per discovery. That's how well we want to do. So far, we now have an extraordinary copper deposit, and we have succeeded. Now we need to do it again and again and again.

One thing I've heard on the capital side, which may or may not be true, so it would be great to get your sense of this, is that a lot of the people who used to buy and run some of these assets—mining assets, or things like that—at least in the Western world, have run into more and more capital constraints because the funders have dried up, in part due to ESG or other programs.

Has that at all been the case, or is it something that's impacted your perception of the sorts of players that are in this business these days? Or do you think that really doesn't matter, and there's plenty of capital availability and it's just hard to find these deposits?

I think the real scarcity is good-quality ore deposits. Great projects don't have problems getting funded, whoever owns them. Great projects have lots of suitors—people who want to buy them. The problem is there just aren't very many great projects. That's what we need to do. We need to go find more really high-quality deposits.

Are there parts of the world that you feel are dramatically underexplored relative to that?

It varies a lot by commodity.

Copper has been an exploration target for a long time, and people have been looking for copper in South America and Central Africa, yet there are still parts of these places that are quite underexplored. We’re very active in Zambia, where, of course, Mingomba is, along with a number of other exploration projects. There are parts of Zambia—like where Mingomba lies—that are deeper underground, where you don’t have surface expression. The deeper parts of the basins in Zambia that host copper deposits are quite underexplored.

Mm-hmm.

You have a jurisdiction like Congo that has had a number of challenges. The exploration potential remains great across many commodities. There has been a lot of activity, but there could be dramatically more activity.

Lithium—much of the world is underexplored for lithium. Lithium hasn’t been a primary exploration target until very recently, until the growth of lithium-ion batteries for big devices like EVs and drones and whatnot, not just personal devices. The big lithium deposits in production today, at least the hard-rock lithium deposits, were found by people looking for tantalum for capacitors for the electronics industry in the 1980s. The science of how lithium ore deposits form is incipient. That’s really exciting because a little bit of increased scientific understanding can be a really potent differentiator. So there’s potential for big breakthroughs.

Are there any commodities that you think are overstated in terms of their scarcity? An example that I’ve heard is that rare earth minerals may not be as rare as people say, and there are deposits more broadly than just in China, where it’s often spoken about. What are the things that you think are actually not that scarce that people talk about as scarce?

That’s at the top of the list. Rare earths—a lot of the noise about rare earths is because they have the word “rare” in their name.

Not that rare. Also, lithium, copper, nickel, and cobalt are not rare earth elements. Rare earth elements are a well-defined term. They’re not just things that are rare, but include neodymium and dysprosium, which are important for permanent magnets, which are important for electric motors and so on. They are important.

The reason that rare earths get so much attention, besides the name “rare,” is the concentration of downstream processing capacity in China. Spurred by Chinese incentives, there’s been a lot of processing—not just for rare earth processing, but also for lithium and now copper smelters as well.

You extract the minerals from the ground, and then you have to refine them into a metal that you can put into a product. There’s been a huge build-out of that. That does a couple of things. One is that it means it’s really hard for somebody else to go build a processing facility because you’re competing for feedstock. You want to take copper concentrate from somewhere and smelt it into copper metal. Well, you have to go buy your copper concentrate. If a Chinese party is willing to buy it for more than you because they will accept less margin, that makes it much harder. It’s much harder to underwrite a project like that.

That has had a deterrent effect on private commercial actors willing to put capital to work and invest in processing capacity. It makes it hard for another private actor to do the same without guarantees or subsidies or something, which we don’t have as a business. That’s a strength of Cobalt: We just have great assets rather than a subsidy.

The second is that, because there’s so much downstream processing capacity in China, you have the raw materials going to China, and then you have a concentration of the downstream supply chain from there, and then you make products from that. It’s a big strength for Chinese manufacturing capacity: You have all of these materials landed there already. If you think about that on an integrated economic basis, it can be very powerful. That’s one of the reasons that rare earths are in the news a lot, too.

Is there anything that’s the other way around, where you actually worry about some commodity or material not being able to meet demand for something that’s industrially important for us?

The ones that I listed for us are the ones where we think there’s a lot out there to find and the demand tailwinds are really strong. There’s going to be some depth to those commodity markets, so you don’t have to have a really well-dialed view on commodity prices, which we don’t. Again, our goal is that we want to be the low-cost producers, and so surprises in that market are not great. We are looking at other commodities that could be those unusual ones, but there isn’t one today that stands out that we’re tackling.

Mm.

So it’s not that important that we buy Greenland?

Not going to go there.

My joke version of this is to take over Baja because it’s already called California. It’s nice and beachy and sandy. That seems like a really great place to annex if you were to annex somewhere.

I’m happy to go to these places regardless of which flag.

Fair enough. Yeah, me too. It actually sounds nice.

Only if the algorithms and the initial rock samples tell you that it’s going to be efficient to get the lithium out, I suppose.

Yeah, we need more lithium out of Baja. So let me know if you go down there. Yeah.

Josh, when we last saw each other, we had a really interesting discussion about how important you felt philosophy was to the business and the investments you’d made about how the company operates. Can you talk about this a little bit?

5. The Epistemology Of Exploration

Yeah. Cobalt is kind of an epistemic project, really. Our business is about making better predictions. That’s what we’re doing, right? The thing we lack is information about where the ore deposits are, and the actual business activity is that we make a prediction, make a hypothesis. We go out and deploy capital, and we spend time testing our hypotheses. So we are successful as a business depending upon how good our predictions are.

That’s what the models are meant to do. We’re making predictions about what the rocks are at the surface and below the surface, and what their properties are, like their density and how much copper, nickel, and other things they contain. So how good are we at doing that? Well, we have to think hard about what basis we are using to make those predictions. What things do we know about the world? One of the critical elements of this is dealing with uncertainty.

When you have sparse data, then you make a prediction about everything in between your data points. There are many possible geologies that are consistent with the data. When you make a prediction based on data that you have from the surface or from an aircraft, and you’re making a prediction about what the properties of the rocks are underground, there are many possible geologies that are consistent with the data.

Standard practice in the industry is to choose just 1 and make your 1 best model because, well, what else are you going to do? It’s hard. You can’t work with 10,000 different models. It’s very difficult to keep multiple inconsistent hypotheses in your mind at the same time, but it’s what we have to do. That is how we become better: by embracing that uncertainty and recognizing that our job is to judiciously reduce that uncertainty. That’s what we do when we go out and collect data, and the data is useful insofar as it reduces uncertainty.

The way that we think about this informs our practice for how we actually explore: What is it that our teams are doing every day? Scientific culture is one of the critical aspects of the business. We have some unusual things. We have a document in the company called Cobalt's Epistemology of Exploration, and it has only a small number of core ideas.

Epistemology is important for the reasons that I talked about. We have to make really definite predictions, and that means they have to be falsifiable. You have to go on record before you collect the data about what you could observe that would cause you to abandon this hypothesis. This is how we avoid confirmation bias, which we are very, very susceptible to in this business. You come up with an idea, and then you collect some data, and you figure out how to modify your hypothesis to accommodate it. Then you justify going out and spending more time and more money.

The third idea is that you have to work with multiple alternative hypotheses. Not just 1 hypothesis, but what are the other possibilities? The point of data collection is to distinguish between them. At least 1 of those hypotheses has to be economically relevant. We are a business, not a science project, right? But careful thinking about what you’re doing is really important.

So the epistemology of exploration—there’s a lot of vocabulary around this that feels like philosophical vocabulary, but it’s really important. Oddly, we have a chief philosopher who is an epistemologist. This is Michael Shrevens. He wrote a wonderful book called The Knowledge Machine about what science is and how it is different from other ways of knowing.

This really guides exploration practice and technology development. A lot of the technologies are designed to quantify uncertainty. Then, given a set of possibilities, we determine what information we can collect that will most effectively reduce that uncertainty.

For those of us who don’t have an in-house philosopher, epistemology is the study of what we know, what knowledge is, how we know it is knowledge, and what constitutes justifiable understanding, right? But maybe one last thing on this. You’re a math and physics guy originally, right?

Mm-hmm.

And you went and did consulting, and you worked in oil and gas. You worked in private equity around it, and so that feels more relevant. But this is such a cool discovery of an interesting problem that you might go apply decision-making science and data to. How did you decide that you wanted to go work on mining and better exploration?

Yes.

I’ve always been interested in the intersection of energy and technology. I studied physics because I like grappling with hard questions, so I did a PhD in quantum computing. I’ve been interested in physics because I like working on hard problems. I like learning things. But I wanted to apply that to the most relevant things in our society today, which relate to our energy systems.

I went and worked with energy companies as a management consultant—with power companies, oil and gas companies, and industrial companies that make power equipment and oil field equipment. My co-founder, Kurt House, and I were doing private equity investment in oil and gas together, in the private equity firm whose leaders had sponsored his previous startup company. We had become friends as graduate students at Harvard. Kurt studied physics and philosophy as an undergraduate, and then applied math and earth sciences as a graduate student.

We would read papers on energy topics and go visit power plants and coal mines and things like that. We were already working in the energy system and quite interested in how the raw materials relate to the global economy, and we decided we didn’t want to work on fossil fuels anymore. This was 2018. We thought from first principles about what raw materials the future economy would need and where those were going to come from.

Think about all the raw materials. Look around you at your desk, in your house, and everything. Every one of these products ultimately originated from agriculture—we grew it—or from rocks—we mined it. What materials are we going to need? Think about the big trends in the global economy: batteries and AI.

Batteries, whether it’s cars and trucks or drones, aircraft, and robots—to make a vehicle that has a long range and is durable, the battery needs lithium. That’s different from fuel-burning vehicles, which have no lithium in them at all. AI is a trend I don’t have to explain to anyone who’s listening to this: there’s a huge build-out of data centers, and the electricity to power those data centers requires an enormous amount of copper. The scale we’re talking about here is gigantic, right?

To build a future that is powered by batteries and AI, by mid-century we will need to mine, over the next 25 years, more copper than has been mined so far in all of human history. To get to a high penetration of battery-powered devices, we need a 10-fold increase in lithium production relative to today. So where are these materials going to come from? We have to go find more of them.

Recognizing that the problem is getting much harder because innovation has been slowing down, this is a perfect application where we can use technology to create a differentiated business that does something really important for our society. That’s really personally motivating to me and to other people who join KoBold.

It’s very exciting. It’s so cool what you’re doing.

No, congrats. I hope you find others. Okay, we’ll keep you posted. Thanks, Josh. Find us on Twitter at nopriorspod. Subscribe to our YouTube channel if you wanna see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.