Data Types and Challenges in Mining Exploration

To create a clean-energy economy, the US badly needs an advanced mining industry that can provide huge amounts of key minerals for batteries and other technologies — and it’s nowhere close to where it needs to be. In this episode, KoBold Metals CEO Kurt House describes the current state of mineral exploration, the significant changes it needs to make, and how machine learning and artificial intelligence can help it get there.

(PDF transcript)

(Active transcript)

Text transcript:

David Roberts

Building the machines and batteries needed to decarbonize the economy will require enormous amounts of a few key minerals. The proven reserves of those minerals, sitting in mines now operating, are nowhere close to enough to satisfy what is expected to be skyrocketing demand.

Without the minerals, we can’t make the clean-energy economy. And we don't know where the minerals are going to come from.

What's worse, exploring for new mineral deposits has been getting less and less efficient over the last several decades, as the amount of investment needed per successful discovery has risen. We seem to be getting worse at finding this stuff right when we badly need to be getting better.

That state of affairs has drawn in several new startups that endeavor to use machine learning and artificial intelligence to improve mining’s hit rate. The most talked-about is KoBold Metals. With financial backing from Bill Gates, Jeff Bezos, and other big-name investors, KoBold is now exploring for minerals on four continents.

To get a better handle on mining and how we can improve at it, I contacted KoBold CEO Kurt House. We talked about the projected gap between supply and demand, the somewhat primitive way current exploration works, the massive data-gathering and coordination project the company has undertaken, and the role of justice and equity in this AI-accelerated future of mining.

Kurt House, CEO of KoBold Metals, welcome to Volts. Thank you so much for coming.

Kurt House

I'm so pleased to be here. I'm a huge fan. I listen to the podcast all the time, so it's fun to talk to you live.

David Roberts

We're going to talk about something that is of great interest these days, which is finding the stuff that we need to build the clean energy economy. This is something I did a series of articles on a couple of years ago, and it's come up repeatedly over the years. People talk about possible shortages of materials as one of the bottlenecks that might slow the clean energy transition. So maybe let's just start there with setting some context, talk a little bit about the big four minerals that you focus on and sort of what we know about how much we have access to and how much we project we're going to need.

Kurt House

Perfect setup question. So, the energy transition is fundamentally about getting off fossil fuels. It's fundamentally about electrifying the economy to the greatest extent possible. So we electrify transport, all electric generation becomes renewable, et cetera, et cetera. That requires a lot of very specific materials and very specific materials because different elements have different physical properties, obviously, and they do different things better and worse than others. And some of those elements are really difficult to substitute for, for very, very deep physical reasons. So KoBold is focused on what we call "the materials of the future," and those are lithium, cobalt, copper and nickel.

That's not at all to say that there aren't other important materials for the energy transition.

David Roberts

Are those four the most important? By just mass, just, we need most of those —

Kurt House

No, by total mass, it'd probably be aluminum and steel, iron for steel. The reason these are so important, there's two orthogonal reasons that we focus on these. One is how difficult they are to substitute for in specific applications. And I'll talk about that. And then the orthogonal element to it is that they are exploration problems. So aluminum is really useful in a whole bunch of reasons, but it's not an exploration problem. There's just gobs of bauxite, aluminum silicon oxide on the planet, and we know where it is. It's just a matter of processing it in more efficient and less carbon intensive ways.

So, it's a metallurgical challenge. It's not an exploration challenge. In the case of lithium, cobalt, copper and nickel, you could take any forecast you want, but basically the end state is something like 2 billion electric vehicles on the planet, plus a whole bunch of renewable energy build out. And any way you slice it, those are just gigantic numbers, and they require gigantic amounts of lithium, cobalt, copper and nickel. And then you can say, "Okay, that's how much we need at, say, 2050 to be mostly off fossil fuels by 2050, how much exists in the reserves of current mines?"

So if we take all the mines that are producing today, and they're going to produce out for the next several decades, that's another number. That's another quantity. And that you also have to add in all of the other uses for these minerals. Right. If the economy just goes on, and there's lots and lots of uses for nickel and copper, in particular in stainless steel and all manner of electrical applications for copper that just happen anyway. So you have to add up those, plus the energy transition metals and compare them to existing mine supply and existing mine reserves, and you get a gap.

And then, if you multiply by current commodity prices, that gap is about $15 trillion.

David Roberts

Good Lord.

Kurt House

We call that — exactly — the "missing metals gap". And so it's not the total amount of metal we need. We actually need a lot more. But that's the value of the metal that we need to find, right? We need to find and then develop into mines.

David Roberts

That represents metal we need, but we don't yet know where it is or where it's going to come from?

Kurt House

Exactly. Because I've subtracted out existing mine supply. Right. And not just existing mine supply, but existing mine supply, plus mines that are in late stages of development. We know they're going to be mines. We've just included that in existing mine supply. So it's really the things that we need to find new deposits that no one in the world knows where they are right now. And then we need to develop them into operating mines — to build new mines. And then those mines need to go into production, and they need to operate for many, many years to produce the necessary amount of metal.

And the value of that metal, roughly speaking, is $15 trillion.

David Roberts

And that all needs to happen — if you look, like, 2050 used to be a lot farther away than it is these days. And now if you look at those lines, they're going up and to the right pretty steeply. So all of that stuff needs to happen much more quickly than it has in the past.

Kurt House

Exactly right, David. And that is what makes this so difficult. So, in round numbers, in very round numbers, it's about 1000 new deposits need to be found, and then 1000 new mines need to be developed.

David Roberts

Wow.

Kurt House

And that, obviously, that's a function of I'm using sort of median mine production. It could be 700 if they're bigger or whatever, but it's order of magnitude 1000. It's a huge number. And then you can say, "okay, well, how fast are we building new mines today? Finding new discoveries and building new mines today?" And the answer is, "not nearly fast enough."

David Roberts

Well, this is something you told me about last time we talked, which has stuck in my head ever since. You called it "Eroom's law of mining," which is Moore's law backward. Explain what you mean by that.

Kurt House

Yeah, precisely. So, Moore's law. The audience will be very familiar with Moore's law. Right. One of the most remarkable demonstrations of human ingenuity of all time, which is that the density of transistors on chips has doubled every 18 to 24 months for 55 years now. And the result is a ten to the 10th order of magnitude increase in computational speed, computational power. So we get better and better and better computation, and that's why you and I can talk remotely in real time from far — everything else that people know. Okay, that's Moore's law. So go back to 1990 and look at how much money the industry was spending in aggregate on exploration, and then divide that number by the number of good new discoveries they were making per year.

David Roberts

Right? Dollars per discovery.

Kurt House

Dollars per discovery. And by good discovery, I just mean a discovery that definitely becomes a mine. It's a good tier one, tier two discoveries, we'd say, in the industry, but it becomes a mine. That number was about $300 million in today's dollars. In 2023 dollars, that was about $300 million per good discovery. Today, that number is about $3 billion. It's gotten an order of magnitude more expensive to find the next deposit. We're getting worse. So we call this Eroom's law, because over the last 40 years, we've gotten ten x worse at exploration, we have to put in ten times the amount of resources to find the same amount of stuff.

Or put it another way, we'll find one 10th the stuff if we invest the same amount in exploration. And exploration expenditure is basically flat, roughly speaking. So we are way, way, way behind and we're spending roughly half a percent a year of what would be needed based on the current exploration effectiveness, dollars per discovery. So on the current rates and current expenditure investment, it will take about 200 years for humanity to find enough deposits.

David Roberts

$3 billion. If you think about 1000 new mines needed —

Kurt House

There you go.

David Roberts

1000 times $3 billion adds up to some large —

Kurt House

$3 trillion. And that's just exploration expenditure. That's not including the cost to actually build the mines, which is a lot more.

David Roberts

So let's break this down a little bit or unpack this a little bit, because with Moore's law, I think people get, on the one hand, getting more computing power out of tinier and tinier spaces gets harder and harder. Like the job gets harder and harder because you're working with just less space and tighter materials and et cetera. But our improvement at doing it is growing faster than the difficulty, basically. Like, we're getting better faster than it's getting harder, I guess, is the way you would put it. If we just remained the same good at doing that, productivity would be declining because it would be getting harder and harder and we wouldn't be getting better.

That seems to be what's happening in mining. It's not that we're getting dumber or worse at mining, it's just that finding the stuff is harder because we've already found the easiest stuff. So finding stuff gets harder and harder, but we're just not getting better at it.

Kurt House

You explained it perfectly. That's exactly right. Another example that I've used, that's even a better example because you just took Moore's law a step further. But another example I use is like fastballs in Major League Baseball were like 85 miles an hour in 1970, and then today they're like close to 100 mph. It's objectively harder to hit that fastball, but batting averages are about the same. Right? So pitchers got better, batters got better. Right. And your point is exactly right is that — are we actually getting dumber? I don't think so. We're not actually getting dumber. It's that the search space is getting way harder because, as you say correctly, the easy things have been found.

And what is an easy thing? It's actually really easy to understand. An easy thing is a deposit, an ore body that's going to be mined, that's sticking out of the ground that a skilled field geologist walks up to the outcrop, looks at the outcrop, identifies ore minerals in the outcrop, and says, "These are ore minerals right here. We should explore this because there might be enough of them to constitute a mine here."

David Roberts

Yeah, you told me that when we talked before, and it blew my mind a little bit because one of the things I'm finding out about this as I do this job more and more is just like normal american consumers are so used to everything going digital and everything being sort of like fancy and computerized now that when I go ask about other areas, I'm often struck by how analog they remain, sort of how kind of primitive they remain. And something you told me is that almost all, like literally almost all of the discoveries we've had and the mines we now have come from someone just seeing something on the surface, like literally the same way they found stuff to mine in 1800.

Kurt House

Yeah, it's absolutely right. If you were to build a time machine and bring the best exploration geologists from 1960 to today, they would be very comfortable working in the industry. Very comfortable. You have to teach them email. Right. You have to teach them a few things —

David Roberts

Zoom meetings.

Kurt House

Exactly. Zoom meetings. But in terms of the field work and the techniques, they would be very similar.

David Roberts

And so should we envision groups of people out walking around looking, or how do we search the surface today?

Kurt House

Yeah, I mean, field geology is a skill, and it's a hard skill to learn. It takes a lot of practice. And there are very skilled field geologists that KoBold employs, and they're absolutely essential to our business. They're fantastic. And just because a technology or technique is old doesn't mean it's bad. There's a lot tried and true methods just sort of continue for some good reason. Right. And so field mapping, for instance, which is basically, if you walk along the ground, think about, you're either walking along bedrock that's an exposed outcrop, or you're walking along soils or something else that's kind of covering the bedrock.

And so field mapping exercises are, there's actually a lot of kind of tricky geometry to it, and it's about identifying the outcrops. It's about making measurements about where the outcrops go underground and then extrapolating in a kind of heuristic way what those rock bodies would look like underneath the cover. Those are sort of useful techniques, and that's what historic field geology is all about.

David Roberts

And so, historically, when someone finds something sticking out of the surface and they say, "hey, this looks like an ore concentration. This looks like a concentration of some ore that we would like to mine" at that point, then what, the mining company just goes in and just starts poking holes down, digging down and looking?

Kurt House

Yeah. Once you find an occurrence, you might call that an occurrence, mineral occurrence. Then you go and explore to see if it's large enough and sufficiently high concentration to be an economic deposit. It's a good lead. Not every occurrence turns into a mine, but every mine at one point was an occurrence. Right.

David Roberts

Right. That's what I'm trying to get my head around is, what does that look like? What do those holes look like? Are they narrow, little pokey holes, or is this, like, a big operation when you're digging down, exploring?

Kurt House

So, for the exploration component, they're very narrow holes. Think a few inches in diameter that you drill. And what you're trying, you're extracting core samples. Right. You're extracting to characterize the full geologic body and to characterize the deposit. Right. And understand what the composition is, what the grade is, how extensive it is. Does it keep going at 100 meters deep, at 500 meters deep, or does it stop at 20 meters deep? Those are all the questions that you try to answer.

David Roberts

Of the occurrences that geologists find and say, "Hey, mining company, this looks promising." What percentage of those pan out into a mine?

Kurt House

Very low. Yeah, very low.

David Roberts

Oh, really?

Kurt House

Yeah. Well, less than 1%.

David Roberts

No shoot. No kidding.

Kurt House

No shoot. No kidding.

David Roberts

1%! I guess that explains the $3 billion.

Kurt House

Yeah, that's right. I'm glad you came full circle back to that, because the reason we've gotten on an underlying cause for Eroom's law, but a more directly observable cause, is that the success rate has dropped. And so you're spending — the individual success cases are sort of just as economic as ever, and they're wildly economic. The difference is your false positive rate has increased or your success rate has decreased. And so you spend more small investments on things that don't pan out. To make the numbers simple, imagine each little small investment characterization effort is $10 million.

David Roberts

So that's to go poke all the little holes and look, that's $10 million.

Kurt House

Exactly. That's to go see if that occurrence is actually an economic deposit. Right. And we're using round numbers here. 40 years ago, you might have gotten 1 out of 30, and today you get 1 out of 300.

David Roberts

Yeah. So $10 million. If you're having to do it 100 times to find metals —

Kurt House

That's the problem.

David Roberts

This isn't in danger of making mining overall uneconomic. Right. If you find the minerals, you're going to —

Kurt House

Right. So the individual success cases are still as good as ever. Actually better than ever, really, because you still spend, say, $10 million, roughly, and then you fail, fail, fail. And then the person that succeeds — and these are often different groups, right — the one that succeeds spends $10 million and then has something that's worth billions.

David Roberts

Right.

Kurt House

And all of the successes, sort of, by definition, have to pay for the failures or the industry just won't attract any capital at all in the aggregate. So you can kind of assume that the successes are worth something like $3 billion, and that's about right. And so the success cases do incredibly well.

David Roberts

A bit of a lottery vibe to it.

Kurt House

For sure. I mean, this is where the word eureka comes from. Literally. The unit economics of exploration are just superb. They're fantastic. They've always been fantastic.

David Roberts

Right.

Kurt House

And so if you can imagine, okay, if you could build a technology or a set of technologies to increase your success rate just marginally, just a little bit better, maybe two or three times better than the industry, then you'd have an incredibly valuable set of technology and company because you could turn exploration into a science. Yes, you're going to fail a lot, but if you fail nine out of ten times, as opposed to 99 out of 100 times, then you make superb money. Because every success —

David Roberts

If you think of it as a lottery, I mean, winning the lottery twice out of 100 times is a lot more than winning it 1 out of 100 times, right?

Kurt House

Exactly.

David Roberts

Because the winnings are very large. And I just want to get this on the record before we move past. As we're talking about exploration, just by way of background-background. None of the things we're discussing in terms of the difficulty of finding and extracting minerals are about absolute scarcity.

Kurt House

Correct.

David Roberts

All four of the minerals you cite —

Kurt House

yes.

David Roberts

are, on an absolute basis, far more common than we could ever use, right?

Kurt House

Absolutely. There's roughly in the top kilometer of the earth's crust, there's enough nickel to give every man, woman, and child on the planet about a million electric vehicles. So it has nothing to do with the number of atoms in the earth's crust. That's not a problem at all. In fact, right below your feet right now, below your house, or wherever you are, I would bet long odds that the concentration of nickel, say, in the ground below you is about 100 parts per million. That's about what it is in the background concentration of the earth. And so that's there.

And we could mine it. We could do it. We could go and mine 100 ppm. In fact, that would be a spectacularly good platinum mine if you had it in that concentration. But it would cost us like $100,000 a kilogram to extract it. The current nickel price is $16 a kilogram.

David Roberts

Right. So what you need to find is concentrations of these things. And the reason that the vast, vast, vast majority of the concentrations we found of these minerals are close to the surface is just that the surface is where we're looking. So what we find tends to be adjacent to the surface.

Kurt House

Exactly right. And it is the mother of all selection biases. Right. Understandably so. You could pose an alternative hypothesis: You could say, "Okay, yes, we found all these things at zero meter elevation, but how do we know? Maybe that's just where they form. Maybe they form when the right minerals contact the atmosphere or something." You can make up a scientific hypothesis.

David Roberts

Maybe they drift up toward the surface for some reason.

Kurt House

Yeah, exactly. That would be a reasonable scientific hypothesis. We know for certain it's not true for these four minerals. We know for certain because we know the pressures, temperatures, and oxidation conditions in which these minerals form, and they form deep in the subsurface. So actually, that's really good news, because that tells us that as we descend into the subsurface, take 100 meters depth slice as opposed to a zero meter depth slice. Your strong expectation should be that the aerial density of deposits increases because since they form deep in the subsurface, a kilometer deeper, it's only the rare few that have been moved to the surface through tectonics and erosion and then been exposed.

Right. And so we really have high confidence they're there in high concentration form. They're just way, way harder to find.

David Roberts

Right. And talk a little bit before we get to how your company is making that easier. Talk a little bit about what we're looking for. We talked before; you mentioned composition, depth and grade; just quickly sort of go over those. What makes for a good deposit? A promising deposit.

Kurt House

Yeah, those are sort of the big three. There's lots of small details, but grade is king. That's a phrase you hear in the industry all the time. And the reason that grade is king is really obvious. Like concentration really matters. So imagine this. Imagine you're looking at two different deposits, let's call them copper deposits. Okay. One copper deposit — so the average grade coming out of copper mines around the world today is 0.6%.

David Roberts

Yeah, I've heard about it. And that's been declining, pretty sharply declining. Right. I've been reading about that. It's very alarming.

Kurt House

Yeah, it has —

David Roberts

Because we need a lot of copper.

Kurt House

We need a lot of copper. And we've been creaming the curve. We've been getting the lowest cost stuff first.

David Roberts

Right.

Kurt House

So 0.6%. So that means if you mine a ton of rock, you do all the work to get a ton of rock out of the ground. You get 1000 kilograms rock. Right. And you get 6 kilograms copper. 0.6%. So you get 6 kilograms of copper for every one ton of rock that you mine.

David Roberts

God, that just seems crazy.

Kurt House

It is crazy, seems crazy, but stay with me on this. So the ton of rock, all your costs go into mining the ore. The rock. Right. The costs scale with ore. Revenue scales with metal, because you don't sell the ore, you sell the metal. Right.

David Roberts

Right.

Kurt House

So now imagine a different deposit that is 6% copper instead of 0.6% that deposit. Now you spend the same amount of money to get that one ton of rock out of the ground. So their costs are the same. But now you get to sell 60 kilograms copper instead of 6. So you sort of, by definition, have a 90% plus profit margin because the costs — the other operator was there and maybe breaking even at 0.6%. And now you're selling 54 more kilograms of copper for no incremental cost.

David Roberts

So find distinctions in the concentration of the mineral in the ore make a huge difference.

Kurt House

An enormous difference.

David Roberts

That's great.

Kurt House

Yeah. So grade is king. The second most important variable is composition, which really gets down to the ore minerals themselves. Right. So a good example here is nickel. And so you actually can find relatively high grade nickel in silicate form. So olivine is a nickel silicate mineral that can get relatively high concentrations. It's not uncommon to get 1% nickel silicates, but it's very hard to process, very expensive to process that. Alternatively, you can get nickel sulfides at 1% or 2% that are very easy to process, relatively speaking, very easy. So a 1% nickel sulfide deposit is way better than 1% nickel and silicate deposit.

And then depth obviously matters a lot. It matters kind of less than you might think. Believe it or not, there's a big distinction between whether it's an open pit mine, i.e. just a big hole that you're digging in the ground, versus an underground mine where you're digging tunnels. There's a relatively large distinction there in terms of cost of operations. But once you go underground, depth doesn't actually matter that much. I mean, up to a point, you can't go 10 km below the earth, but the difference between 300 meters and 700 meters doesn't matter that much, actually.

David Roberts

So digging that additional 100 meters down is not a huge cost.

Kurt House

It doesn't matter a lot. Yeah. I'd much rather have higher grade deeper than lower grade, shallower.

David Roberts

So what would be like a really high grade, say, for copper? You say 0.6 is the average these days. That seems — I mean, what do I know about. I have nothing to compare it to, but it sounds low.

Kurt House

It is low.

David Roberts

What's a really good grade?

Kurt House

Over 3% is superb. Like superb, superb.

David Roberts

Got it.

Kurt House

Over 5% is absolutely world class. And there's almost nothing like it out there.

David Roberts

Interesting. Okay. What we've established is currently we're finding minerals based on some exploration geologist on the surface, squinting at rocks, finding good rocks. And then you poke a bunch of holes. If you find a decent looking rock, you poke a bunch of holes around it, find out if it's down there in the 1 out of 100 times that you find suitable deposits down there. You dig down there and open a mine. And this is why 98%, 99% of mines and deposits have been near the surface or close to the surface.

Kurt House

Correct.

David Roberts

So this is all background for current mining. And the productivity of that is declining, presumably just because the big obvious surface concentrations that we could find, we've mostly found. And so now we're like squinting, we're going after harder to find stuff, et cetera. And the whole productivity of the sector is declining, even as we desperately need it to 10x its output in the next 20 years. All in all, an alarming situation.

Kurt House

That was a perfect summary.

David Roberts

So then along comes KoBold. So just tell us, how is KoBold going to help that situation? What is it you do to help miners?

Kurt House

Yes. So KoBold's main objective is to improve the success rate of exploration, right? Exactly. This problem. The declining success rate of exploration is Eroom's law that's resulting in the higher cost, the more cumulative expense to make a discovery. And so our objective is to improve the overall success rate or really the exploration efficiency or effectiveness, which is dollars per discovery. That's our real goal. We want to have our own exploration success rate of closer to $100 million per discovery, as opposed to $3 billion per discovery.

David Roberts

And that's the money side. What about the percentage side? Like you said, 1% success rate. Now, we're skipping ahead a little bit here, but if KoBold continued to advance and miners really took it up and took it seriously, what would a really good success rate look like?

Kurt House

Yeah, well, let me actually correct you on one thing. We're not a service company or a SaaS company at all. We don't provide, we don't sell anything to mining companies. Yeah, we are a full stack exploration, development and mining company ourselves, started in Silicon Valley. We are Silicon Valley's mining company, so to speak. Right. Your confusion is totally understandable, and it's common.

David Roberts

I thought you were striking deals with big mining companies.

Kurt House

We do, but only on co investment into particular projects. We own — we don't get paid anything for services or anything for software. We make money by making discoveries, and we partner with mining companies on their properties as well as we have our own. So we have something like 60 projects worldwide, and about half of those we own 100% by ourselves, and about half of those we own in partnership with other companies. Partnerships are very important to us because we want to extend the reach of our technology. And lots of companies own a lot of ground that they're not exploring.

And so we can sort of leverage that opportunity and say, "Okay, this is interesting ground. It's actually worth more in our hands than it is in your hands, because we can deploy our technology, let's work it together, and we'll both benefit from any discovery."

David Roberts

So you're co-mining those, you're both involved in the mining?

Kurt House

Very common misconception and very understandable, because if you look at our team, it's about two thirds of the technical staff have never worked in the mining business before. They come from Google and Apple and Meta and Silicon Valley and all the tech monopolies that you know and love or hate, and they probably never will again. Right. They don't think of themselves necessarily as working for a mining company. I mean, of course they do. They understand the business, but they think of themselves as working for a tech company. And they are deploying their skills as data scientists, software engineering, software engineers for this purpose.

David Roberts

Right. So you own big machines that dig up the earth.

Kurt House

Principal asset are licenses to deposits. Right. Our principal assets are the deposit itself or the deposit itself. And then, of course, we use all kinds of capital equipment to explore and ultimately develop the deposits.

David Roberts

Okay, so if you own these yourself, then you don't have to speculate what better discovery rate would look like. Presumably, you have established one. 1 out of 100 is the baseline here. What does your discovery rate look like?

Kurt House

Yeah. So I prefer dollars per discovery. And the reason I prefer — I'm going to answer your question, but it's important to think about the different metrics. The reason I prefer dollars prefer discovery is we encourage lots of little projects that falsify their hypotheses fast and for low money. Right. So, there was a project we had recently where one of our scientists had a hypothesis about a particular deposit. They came up with a clear falsification criteria. We staked the property for probably $1,000. We went out to the field. They made several measurements, geochemical measurements. It falsified the hypothesis.

We moved on from the project at something like $5,000. Right. It's an extremely efficient condemnation. And that's really important because there is an enormous amount of uncertainty in this business. That is the nature of the data science problem is it's a sparse data problem, and we're making inference on very select data about the physics and chemistry of the Earth crust, trying to make predictions with quantified uncertainty and then seeing how accurate those predictions are and then updating our models accordingly. So we want lots and lots and lots and lots and lots of shots on goal. So I don't actually track total number of attempts because actually the numerator is high.

But dollars per discovery, I very much do track. And we are on track for we have made one major discovery, which is public. It's in northern Zambia. Cumulative exploration expenditure up to the point of that discovery was actually less than 50 million.

David Roberts

That's quite a bit better than 3 billion.

Kurt House

Yeah, exactly. So we're very much on track in that sense, except that we only have one that's totally unambiguous, and then we have a lot more that we're working toward. So we'll see. It'll take time, we'll see over the next five or ten years. But I'm very encouraged that we are on track at this point.

David Roberts

Okay, well, so let's back up. What the company does is gather data, use AI, machine learning to analyze the data in order to predict where you're going to find concentrations, basically.

Kurt House

Correct.

David Roberts

That's the nutshell. So let's talk about the data. I think when people hear this, their first thought is like, "Well, if the data was there, why weren't other people using it to find these things?" Do you know what I mean? If the data was publicly available, what's your magic sauce? So, first of all, let's just talk about, and this struck me, too, the last time we talked, is just the wild range of data that you guys are gathering. Talk a little bit about your data gathering.

Kurt House

For sure. So that is the right question for any kind of AI/ML technology ever, any application is, what is the data? Right.

David Roberts

What are you learning from?

Kurt House

Right. The algorithms, with all due respect, the algorithms are pretty straightforward, actually. They're brilliant, but they're easy to replicate and they're kind of a commodity at this point. What the real secret sauce of any AI company is, do they have superior data source?

David Roberts

Yeah, so I should mention, I was going to mention later, but I'll just mention now, obviously, KoBold is not alone in this. There are several companies now trying to use basically data gathering and AI to improve success rates in mining exploration. There's a bunch in Australia, I think there's one in Europe. So insofar as you're competing, I mean, it's probably a giant market, and there's room for plenty more entrants, I would imagine. But insofar as you're competing, that's going to be the arena in which you're competing is who can find the cleverest new sources of data to improve their results.

Kurt House

Yeah. There's a lot of truth to what you just said. I will mention on competition — this is useful — I am really rooting for those other companies. And the reason is I literally never think about competitors. And that sounds cocky, but it's not really, because let's say we need 1000 new mines for the energy transition. If KoBold is responsible for 50 of those new mines, we are going to be wildly successful beyond my possible imagination. And that's 5% of the problem.

David Roberts

Right.

Kurt House

And so we still need other people to find and develop the other 900-plus mines. And so what is it that's going to stop KoBold from getting from 1 to 50? It is not that someone finds that one first because there's another 950 we need to find. Right. What it is, is that we just fail to find it. Right. So we are fundamentally just kind of competing with ourselves, in a sense.

David Roberts

Right. More of a pass-fail than a grade relative to competitors.

Kurt House

Yeah, totally. It's just like, if you're like what I tell my son in track, you're just focusing on your own pr. Just get your own time down and the rest will work out. Right.

David Roberts

All right, so let's talk about data.

Kurt House

Yeah, so, data. So this is really interesting. Right? So humans have been collecting information about the physics and the chemistry of the earth's crust for a very long time. Yeah.

David Roberts

Mining is real old.

Kurt House

Really, really old. Right. In some ways, they've been doing this for. I mean, we've been doing this for millennia. In some ways, certainly in the last century, been collecting information in more and more sophisticated ways. And virtually all of that information is in the public domain, actually. And the reason it's in the public domain is either it was actually collected at the public's expense in the form of various state geologic surveys around the world and then made public. That's thing one. Thing two: It was collected by academics and then made public.

David Roberts

Right.

Kurt House

That's thing two. Thing three is created by private companies, but almost always, with some exceptions, the rules. The laws in every jurisdiction out there are that when you have an exploration license or a mining license, you are required to submit the data that you collect.

David Roberts

Oh, interesting. That's by law. Everywhere by law.

Kurt House

That's right. And these are really good rules, actually, because what you don't want, if you want efficient exploration of resources, you don't want everyone collecting the same data, proving the false positive again and again and again. Right. That's bad. So usually the way the rules work is the company gets a couple of years to husband the data by themselves, and then they have to make it public. Right. And that's true in basically every jurisdiction we operate in. So there's a huge amount of information. We'll get to what the information is in a moment. But the challenge is that this is the messy data of all messy data problems.

Right? So it has been collected. You're talking about everything from modern worldview three, high resolution, high spatial, high band resolution spectral imagery all the way to 1920s handwritten drilling reports. Right? That's all the information. And there's everything in between. And so it's every media, every storage media, from paper to cloud storage, every storage media you can imagine.

David Roberts

Microfiche?

Kurt House

Microfiche, you know, magnetic tape, you name it — it's all out there. And then it's every jurisdiction; every sub-jurisiction has different formatting requirements over time.

David Roberts

To what extent has there been a kind of shared format for this? Or is it all completely disparate?

Kurt House

Basically none. I mean, people could quibble with me and they could point out some standardization efforts in the past, which is not wrong. But basically, to first order, if I show you 100 data sets that we found in the public domain, they're going to be in 100 different formats.

David Roberts

Oh, good grief.

Kurt House

Yeah. So the first thing that we do, and this is a major effort, we've been doing this for five years, and we're going to continue to do it for many more years, is we identify these data stores, and then we ingest the data, bring it into our system. Which could mean, it could mean, digitizing paper records, right? And we have those operations around the world, including a very extensive one in Zambia, which is really cool. We digitize the paper records, we do various optical character recognition techniques on those records, and then we use various extraction, NLP and other extraction technologies to extract the data, transform it into what we call our universal data schemas.

So we are the major effort to create a standardized format for all of this data, which is our own proprietary format.

David Roberts

When I'm thinking about this process, how much of that is done by AI and machine learning versus some poor schmuck squinting at two columns of numbers and typing things in?

Kurt House

Great question. So I like to say that KoBold runs on AI, HI and HS. So that's artificial intelligence, human intelligence, and human sweat. And the last one may be the most important. It's hugely important. And we're automating things all the time. There's all kinds of methods of automation, but these are really hard problems, because not just is the data in all these myriad formats, but what the data is requires real scientific judgment. Right? Different. A magnetic survey collected in 1965 is just hugely different than a magnetic survey collected in 1990.

David Roberts

So this is not something like, you can't hire like a teenager to do data entry kind of thing. You need an expert to be looking at.

Kurt House

Yeah. The things that you could easily outsource to a low skilled person that's automated, we just automate those things. But then there's lots and lots and lots of areas where human judgment is really needed. And the technology basically makes the humans much more productive. Right. Because it enables them to see, like, kinds of data fast, to see what the units were, to see what the data collection method was. It highlights the relevant information for humans, but it is definitely a human in the loop process, and it's, for all intent and purposes, will stay that way for a very long time. It requires a lot of investment.

David Roberts

Right. So this sounds like the bulk of your value add then, mainly like this precisely. Gathering and standardizing information.

Kurt House

I would say it slightly differently. But I think it's a reasonable assertion. I would say it's the least ambiguous value add. Right. Like, it's clearly really valuable to get all the information and make it all usable. That is like, nobody will disagree with that. Right. And so that's definitely the case, then, once we have the information, I should talk about what the data types are in a second. Once we have the information in accessible form, then there's just a huge number of scientific computing and AI algorithms that interrogate the data and do all manner of things to predict where we're likely to find compositional anomalies and concentration anomalies.

David Roberts

So what are some of the kinds of data like? Presumably people have dug down and pulled up cores.

Kurt House

Yeah. So we'll start with geophysics. So the earth's gravitational field changes from place to place on the planet because the density of the rocks beneath you change from place to place. Yeah, it's really cool. And you can actually measure these changes with gravitometers that existed in terms of sufficient precision for 50 years. People have been doing gravitational surveys for many, many decades in different places and different evolving technology platforms. But that's one type of geophysical data. The earth's magnetic field changes from place to place as you go around the earth because the background field gets distorted by the magnetic properties of the rocks in the near surface.

There's various types of electromagnetic data that tell you something about the distribution of conductivity of the rocks in the subsurface. Right? And there's electromagnetic data that's based on active surveys as well as it's based on passive surveys. There is all manner of imagery. Right? So aircraft and satellite imagery in the visual and in the wider non visual bands. And that can tell you lots about both outcropping rocks as well as the materials overlying outlying rocks. There's all manner of chemistry. Right? So you referenced this just now. There's groundwater chemistry, there's soil chemistry, there's sediment chemistry and streams.

And then there's rock chemistry, both from outcrops and from deep drill holes. Then there's mineralogic data. So chemistry tends. We use that as a shorthand for the elemental concentrations in a rock sample. So the percentage of copper, the percentage of nickel, the percentage of cobalt, et cetera, and the percentage of weird stuff, too. And those trace elements are really important. The percentage of caesium, the percentage of tantalum. Right. This stuff tells you a lot. So you end up getting, if you have 50 or 60 element concentrations, you can do some very sophisticated high dimensionality sort of machine learning to predict how the rocks are changing in the subsurface.

And then there's mineralogic data, which tells you about not just the elemental concentrations, but what minerals they're in, the forms of the molecules themselves. Right. Is it nickel in silicate, which would be less interesting than nickel in sulfur? Right. And you need to actually understand the molecular form, not just the elemental concentration.

David Roberts

So how close — and maybe this is like too sprawling of a question to answer, but how close are you? Well, a) are you still finding new sources of information? And b) how close are you to ingesting and systematizing and standardizing all the data that you do have?

Kurt House

The last question, I would say we're way ahead of anyone else in terms of aggregating all of the information about the earth's crust. And we got a long way to go.

David Roberts

Oh, really? It's still early days with the information that's available.

Kurt House

That's available. Yeah. I would be shocked if we've already fully aggregated a few percentage points.

David Roberts

Oh, interesting.

Kurt House

There's a huge amount, yeah.

David Roberts

So that's good reason then to think that there's lots of Runway for improvement here as more and more information comes into the system.

Kurt House

For sure. Maybe the neatest innovation we have, data science innovation, kind of foundational innovation, is something we call efficacy of information. And what this is is we ingest all of this legacy information and we make these predictions. We have, in most cases, very sparse data, we have little hints of data here and there. So we have a huge amount of uncertainty. And the game now is to make a set of decisions that will decrease our uncertainty. That will maximally decrease our uncertainty. Right. And so that's actually the way we think of exploration. We think of it as an information problem.

That's about the maximum reduction of uncertainty. No other mining company talks that way. Right? Talks about sort of information theory and maximally reducing uncertainty. But it's fundamentally the way we think of exploration, of the exploitation process. So efficacy of information technology, what it does is it actually indicates what information we should collect that will reduce our uncertainty the most. Right. Which is different than what information you should collect to find a deposit. Like, it's a different question. It's no, how do we decrease our uncertainty the most? I'll give you a fun example, a very tangible example of this.

So we have a lithium prospect in Quebec. And the way we got on to this prospect was by initially searching old records for the lithium mineral that we're looking for is called spodamine. So that's an aluminum silicate lithium mineral. And so you can imagine that you'd go look for spodamine, and there's the word in old geologic records, but it turns out that that's not very fruitful. And the reason that's not very fruitful is because unlike copper and nickel, people have not been looking for lithium for all of human history. In fact, they weren't even really looking for lithium ten years ago.

It is an extremely new thing, and that makes it really interesting. So most geologists weren't even thinking about it. They weren't thinking about spodamine 40 or 50 years ago. So if they weren't thinking about it, they were less likely to see it also. And this is non obvious at all, but it turns out that the spodamine mineral is really, really hard to detect in rocks, which some minerals are easy to see for a skilled geologist, and some are really hard, and spodamine just turned out to be really hard.

David Roberts

So you're looking for proxies, then it's what you end up doing.

Kurt House

Exactly. So we're looking for proxies, things that we have figured out through basic science and through data science that correlate with the presence of spodamine. So we found some old reports that were just littered with those proxies. So we send people out, but we didn't send them. We actually staked a big claim, 300 square kilometers of area, and this is an area that has snow on the ground most of the year. So you have a narrow window to send people out and collect new information. And it's expensive. Right. This is another cool element of the data science challenge of exploration, is that the marginal cost of data is very high.

Right. And that's really different than like a SaaS company or a social media company that's getting swamped with data, and they're trying to find subtle signals in the noise. We have this high marginal cost of data, so the whole game is to collect the most useful next piece of information. So we have this big area, we have maybe two weeks in the field, helicopter supported. So this is not cheap. Ten highly skilled people supported by a helicopter for ten days. And so we use machine learning algorithms to tell them where they're likely to find the right rock sticking out of the ground.

Okay? And so the algorithms were trained on data that we had a priori, and the team goes out there and they find a couple of good examples, and they say, "Yeah, this is good. This is a pegmatite. This is the sort of rock type we're looking for." And it had this white moss growing on it, it's a white like rock and has this white moss growing on it. Most of the places that the ML sent them, most of the places were not the right rocks with that white moss growing on them. They were just fields of white moss.

It was bad, it was false positives all over the place. So we sent them to all. So it was really bad, right? So they're in the field and they're on satellite, and they're like, "You guys, data scientists, this is not right, this is moss, these aren't pegmatites." But that's okay, that's the way this iterative process works. So the team goes, "Okay, we have a few true positives in the pegmatites you did find, and we have lots of true negatives now, right? Those false positives are the same thing as true negatives, right? We know those are not the right things."

So that's really, really good training data.

David Roberts

Somebody's going to go write a moss algorithm now.

Kurt House

Exactly right. So we include that into the model as not the thing we're looking for, as well as some positive cases of the thing we're looking for. The model gets way better, way more predictive in just a couple of days, inside the window of the field campaign. So then the new model, while the people are still out there, the new model says, "Okay, here's a much more precise model. Many fewer predictions, some novel new predictions, many fewer predictions. Go check out these that are like 50 km away." And they were spot on. They were exactly right.

And now, the model was really predictive. Just a little bit of high quality training data made the model way, way better and got people to not just the right pegmatype deposits, but pegmatite deposits that were rich in spodamine, which is the lithium bearing mineral. So that could be a new lithium discovery, and we're able to make it way faster and for way less expense because we directed the team so efficiently in the field.

David Roberts

Right. This kind of gets up. My next question, which is most of what you're doing, is assessing and analyzing existing data sources that you've gathered. And the next thing I was going to ask is, are you yourself producing another data set? Are you doing any sort of large scale scanning or, I don't know, whatever it would be to produce another data set. But it sounds like you're doing something much more targeted than that.

Kurt House

Yes, both. So, yes and yes. So the targeted exploration is critical for efficient exploration, and then we are collecting new data, highly selective, high quality data, like I just described, like those moss covered fields good, strong, true negatives, okay? But here's what we also did, and it was really cool. We actually have a whole hardware program where we're developing new measurement techniques and new data types, and we have something we've invented called the KoBold hyperpod, which is a camera that we designed that is the highest spatial resolution and highest spectral resolution of any camera used in mineral exploration.

We mount it to the side of a Cessna aircraft or a helicopter. So after we found this remote, good outcrop, we then flew over it with the hyperpod. And the hyperpod, to make this really clear, so the satellite imagery has a pixel resolution of about four meters. So it's about a four meter spatial resolution, okay. And then it had a band resolution, spectral resolution of eight different bands. So eight different wavelengths of light. The intensity was being measured. So you can imagine for a four x four pixel, you would get eight numbers, okay? The hyperpod. The new hyperpod has, in a four x four area, 20 pixels, so it has 20 times the spatial resolution.

So you get 20 different pixels for every one pixel, and then every pixel has 80 times the number of bands. So it's 80 times the band resolution going all the way up to three microns, like really long wavelength. So we collect much more rich information, 1600 times as much information about that outcrop than you had from the satellite imagery. And then we fly a wide area, we fly that camera around a wide area, looking for more of the same. So in this way, the whole exploration machine becomes much more effective at every stage, because the software enabled the more effective exploration, which found the better training data.

And the novel hardware then collected really good quality data, which then gets used to find the next prospect after that.

David Roberts

So you're laddering up then, and ought to be getting better and better at this, basically, because.

Kurt House

Correct.

David Roberts

The more information you get, the better data you have. So we're getting close to the end. One or two other things I wanted to get before we're done. One is just about what you found so far. You said you're relatively new at this. You found one big demonstrated deposit of what is the —?

Kurt House

Copper and cobalt, mostly copper, a little bit of cobalt in Zambia.

David Roberts

And it's new that you're looking for lithium. That was a relatively recent announcement, I thought.

Kurt House

Yeah, we've been — so we started looking for nickel, copper, and cobalt. And for the first two years of the company, actually, for the first three years of the company, we exclusively looked for those three. We started our lithium program about a year and a half ago, and it's a very different search space in really interesting ways, but we've spun up a really exciting lithium program. Now we're exploring on four continents. We're exploring in Asia, Australia, North America, and Africa for lithium, and we have projects on all four.

David Roberts

Oh, wow. So you have one big demonstrated find on your record now, and you're exploring in other places. You're exploring in a bunch of other places.

Kurt House

Yeah. We have nickel, copper, and lithium projects in twelve jurisdictions, I think Greenland, Quebec, Ontario, Saskatchewan, Nevada, Alaska, Western Australia, Namibia, Botswana, Zambia, South Korea. So, yeah, we're exploring all over the world for those key metals.

David Roberts

Okay, and final question, and this is, I'll get yelled at by my audience if I don't ask you this, which is just, we've been discussing all this as a sort of purely mechanical kind of a technical challenge, but obviously, anytime you talk about mining, I'm sure you're very familiar, having been in the space for a while, first thing that leaps to people's mind is environmental degradation and social problems. The mining there are, quite famously, a lot of bad mines, a lot of bad conditions at a lot of mines. How do you think about the equity justice angle in your work?

Kurt House

Yeah, I'm so glad you asked. It's so, so important. Right. So the mining industry has a spotted history, for sure. For sure. There's some good stories and there's some very bad stories, for sure. These are things we take enormously seriously for a whole host of reasons. Sort of with the giant picture, the big picture, which, of course, we started with in the beginning, which is just humanity has a choice, right? We either get off fossil fuels or we fry the planet. And I don't think that's an acceptable choice. And so what does it mean to get off fossil fuels?

It means massive electrification, and that, just because of physics, requires a lot more of these key materials. If we're going to solve climate change, we need to find and develop these materials one way or another. My view is very strong, that if the mining industry doesn't act exceptionally well in development over the next ten years, then reasonable local stakeholder opposition to any projects will thwart that effort. Right. It'll thwart the effort to get the materials. So we have to be absolutely best in class, and we strive to lead in that way. So most companies will start a major community relations program once they're there to actually develop a deposit.

We start it earlier. We start when we're exploring. Right. Even though we know most of our projects won't actually come to fruition, we want to make sure the places we are exploring are places that we are welcomed and we want to be, we want to be welcomed by the locals and we want to work with the locals so that we remain welcomed by the locals. And that means a huge number of things. But it means fundamentally, it means a really serious investment, right. With an entire group inside the company that's dedicated exactly to that. And people that spend all day, every day talking to the local community about their concerns.

At this point, KoBold, the majority of KoBold employees are actually in Zambia. And so one of the things we are trying to lead on heavily is to mostly hire Zambians for our major project in Zambia. It is kind of standard operations that when there's a big project like this, western companies move in and bring in a lot of expats to do the high priced jobs. We have a strong commitment to not do that. Right. We have some expats, of course, for very specific technical expertise, but mostly we're hiring local and we're training local very aggressively. And that's in our long term interest.

We want to have the best zambian workforce that we can possibly have because we don't want to just develop one project in Zambia. We want to develop ten projects in Zambia. And we're investing in all the neighboring countries. And so the more capabilities we have in that country, the more effective we'll be in the neighboring countries.

David Roberts

Let me ask you this, since I discovered that you're an actual mining company. It's nice that you are making efforts to do this, right in terms of dealing well with the locals and labor standards and stuff like that. Is there demand side pull for better standards? Is there now a market force pushing for better standards, as one would hope?

Kurt House

Great question, and this is a great coda here for the conversation, because I think your audience can help here. I think demanding high environmental standards and high labor standards in your key critical materials in end use products is something that every listener can do. You guys should actually tell when you buy your next iPhone, you should ask, right? Like you should ask about these things.

David Roberts

Are there programs or standards or certificates or something? Like what can a consumer use to know?

Kurt House

Yes, there are programs they're starting. They're nascent. Right. And so they need just grassroots consumer support. Right. So right now, no products that you buy — you don't have two different prices. Like a price for a —

David Roberts

"Dirty phone."

Kurt House

Yeah, dirty phone or a clean phone. Maybe that's impossible because maybe no one would want to advertise their phone as a dirty phone. But you definitely want companies advertising their phones as clean phones, right? You definitely want that. And honestly, I would say the diamond industry, even though I think diamond exploration is kind of stupid and wasteful, they've actually done a pretty good job here with the so called Kimberley process and things like that, because blood diamonds were a horrendous issue.

Still are. But they did a pretty darn good job with that. Such that the provenance can be relatively tracked. So there are techniques, but it really requires sustained consumer engagement so that all the companies involved, so the entire industry takes it up. The providence of our metals will be exceptional. You'll be able to track every — once we're producing, our mines aren't producing yet, but once they're producing — you will be able to track every gram all the way back in the entire supply chain in an extremely transparent way. And we hope we can set an example that way.

David Roberts

Awesome. Well, that's something listeners can do, especially if listeners are associated with institutions. Try to get institution procurement or government procurement aimed in the right direction is a big piece of the puzzle, too. Well, Kurt, this has been absolutely fascinating. Mining is not something I thought I would ever have to get into or care about. But as you say, you pull this string and here's where you end up. So I'm glad somebody's out there working on it. Thanks for coming on and walking us through it.

Kurt House

It was a great conversation. Thanks so much for having me, David.

David Roberts

Thank you for listening to the Volts podcast. It is ad-free, powered entirely by listeners like you. If you value conversations like this, please consider becoming a paid Volts subscriber at volts.wtf. Yes, that's volts.wtf. So that I can continue doing this work. Thank you so much and I'll see you next time.

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.volts.wtf/subscribe