AI transcript
0:00:02 [MUSIC PLAYING]
0:00:06 Pushkin.
0:00:13 Digging up metal from out of the ground
0:00:16 is a business that is literally thousands of years old.
0:00:19 But mining suddenly has new importance.
0:00:22 The energy transition, going from fossil fuels
0:00:24 to renewable energy, is going to take
0:00:29 a ridiculous amount of metal, metal like copper and lithium.
0:00:31 The need is so great and so urgent
0:00:33 that we’re going to have to come up with new ways
0:00:36 to find metal buried in the earth.
0:00:39 And as it happens, a new kind of mining company,
0:00:43 a mining company you might call an AI-driven startup,
0:00:47 just made the biggest copper discovery in over a decade.
0:00:49 It’s worth tens of billions of dollars.
0:00:52 [MUSIC PLAYING]
0:00:57 I’m Jacob Goldstein, and this is What’s Your Problem,
0:00:59 the show where I talk to people who are trying
0:01:01 to make technological progress.
0:01:05 My guest today is Kurt Hauss, the founder and CEO
0:01:07 of Cobald Metals.
0:01:09 Kurt’s problem is this.
0:01:13 How do you use AI, machine learning, data science,
0:01:17 to find the metals we need for the energy transition?
0:01:18 As you’ll hear, my conversation with Kurt
0:01:23 goes beyond mining and AI to cover Kurt’s really compelling way
0:01:27 of just thinking about making decisions in an uncertain world.
0:01:30 We started, though, by talking about how he came up
0:01:31 with the idea for his company.
0:01:39 So if you go back about eight years ago,
0:01:43 my co-founders and I were looking at the trends
0:01:47 in the energy transition, seeing the electric vehicle
0:01:49 and renewable energy sort of revolutions coming.
0:01:55 And it’s quite easy to convince yourself
0:01:59 that the material requirements for the energy transition
0:02:03 will be tremendous, the amount of very specific materials
0:02:04 that the world needs–
0:02:09 copper, lithium, cobalt, nickel, graphite, others.
0:02:13 This is basically stuff to build batteries and wires,
0:02:14 essentially.
0:02:17 This is just batteries and electrification, right?
0:02:19 Just an electric motor is a bundle
0:02:21 of copper wire surrounded by, you know,
0:02:23 surrounding a permanent magnet.
0:02:26 Every battery require– every mobile battery
0:02:30 requires lithium and nickel and cobalt.
0:02:31 These are all the–
0:02:34 these are the key materials for which, in some cases,
0:02:35 like the humanity has been using lots of copper
0:02:38 for a long time– there’s a big copper market–
0:02:42 but it needs to at least double from a very large base.
0:02:46 Lithium, humanity has not been using much lithium for very long.
0:02:50 And now the lithium market needs to grow by well more
0:02:57 than a factor of 10 to fully electrify just the transportation
0:02:59 sector.
0:03:02 So the sort of macro needs were very obvious.
0:03:04 So that’s observation one.
0:03:08 Observation two– say, OK, well, maybe the incumbents
0:03:10 are really good at finding new materials.
0:03:13 And as prices rise a little bit, they’ll find new materials,
0:03:14 and the market will just be well supplied.
0:03:17 And that turns out to be definitely wrong.
0:03:20 And it’s actually really easy to verify that it’s wrong,
0:03:23 because the large, well-resourced mining companies
0:03:26 basically don’t even do exploration, actually.
0:03:31 The big mining companies, they spend $65 billion a year
0:03:34 on dividends and share buybacks and less than half a billion
0:03:35 on exploration activities.
0:03:38 But that half a billion, that’s the deployment
0:03:42 of conventional exploration technologies, right?
0:03:46 Things that would be natural to most geologists
0:03:49 from the 1960s or ’70s, right?
0:03:51 So you can round to zero how much money they’re
0:03:53 spending on research and development
0:03:55 for new techniques and new technologies
0:03:57 to improve the exploration process.
0:04:00 So it was basically those sets of observations,
0:04:02 those two sets of observations.
0:04:04 We need metals, and nobody’s really looking for them.
0:04:05 They’re not looking for them, and they’re certainly not
0:04:06 getting better at it.
0:04:07 They’re getting worse at it.
0:04:13 We call that trend and the increasing cost of discovery.
0:04:17 We call that e-room’s law of mining.
0:04:18 Moore’s law backwards?
0:04:18 Very good.
0:04:19 Yeah, I’m impressed.
0:04:20 Yeah, exactly.
0:04:22 They talk about that in biotech as well.
0:04:24 We borrowed it from biotech.
0:04:29 Meaning, whereas microchips get cheaper and better every year,
0:04:32 mining gets worse and slower and more expensive every year.
0:04:36 It’s specifically exploration and discovery, specifically.
0:04:38 So those were the major needs.
0:04:40 So then you say, OK, what can we do?
0:04:41 How can we do?
0:04:43 What can we do differently?
0:04:44 How can we help?
0:04:50 And the answer is that exploration is fundamentally
0:04:52 an information problem.
0:04:53 Fundamentally, right?
0:04:55 We know for deep physical reasons,
0:04:59 which I can explain in a minute, we know there’s gobs
0:05:03 and gobs of undiscovered rich deposits out there.
0:05:05 We don’t know where they are.
0:05:08 So the gap is the knowledge of where they are.
0:05:13 If God gave you a perfect model of the Earth’s crust,
0:05:15 the location and form of every atom,
0:05:16 you’d be a perfect explorer.
0:05:20 You’d know where all the high-grade, high-concentration
0:05:21 anomalies were.
0:05:23 You’d also be a perfect miner.
0:05:26 The miner’s religious vision is the gift
0:05:28 from God of perfect information, yes.
0:05:30 Perfect, exactly.
0:05:31 But it’s not that.
0:05:32 So we don’t have that.
0:05:34 So we have a huge amount of uncertainty.
0:05:38 But the sort of managing the data that you have,
0:05:42 and then making probabilistic inferences on that data,
0:05:46 is fundamentally an information problem.
0:05:50 We look at it as this is kind of a perfect tailored
0:05:54 application for data science and modern scientific computing.
0:05:57 It’s a little different.
0:06:00 It has some sort of unique, really cool attributes to it.
0:06:02 But it is fundamentally an information problem
0:06:04 and fundamentally a search problem.
0:06:08 And so the thing that could be massively different
0:06:12 would be a company built from the ground up,
0:06:15 a Silicon Valley company built from the ground up
0:06:18 that combines the best existing knowledge
0:06:23 of geoscientists with world-class data scientists
0:06:27 and software engineers coming out of the major tech monopolies–
0:06:31 Google, Apple, Facebook, you name it–
0:06:33 who have never worked in the metals and mining business
0:06:34 before, right?
0:06:36 So it’s fundamentally sort of bringing
0:06:40 the tools of data science, machine learning, AI
0:06:43 to bear on geoscience.
0:06:43 Absolutely.
0:06:44 If I’m going to reduce the idea, it’s that.
0:06:46 Totally, yeah.
0:06:49 It’s amazing that nobody got to it before you did.
0:06:51 There are these giant billion-dollar mining companies.
0:06:54 And it was right there for them, but they didn’t do it.
0:06:57 I mean, why didn’t somebody do it before you?
0:07:00 What you will definitely hear is, oh, we use data science.
0:07:02 Like, we use scientific, right?
0:07:06 And it’s like not totally wrong, but what is definitely
0:07:10 unambiguously different, if not unique to Kobold,
0:07:13 is that we’re a full-stack explorer.
0:07:18 We were started and built on the concept
0:07:23 that applying vanguard scientific computing
0:07:27 techniques to these problems would improve the efficacy
0:07:28 and efficiency of exploration.
0:07:29 That is the goal.
0:07:34 We have– our technical staff is about 60% data scientists
0:07:37 or software engineers and about 40% geoscientists.
0:07:41 So we’re roughly equal numbers across the three disciplines.
0:07:44 And that’s completely– that’s unique.
0:07:47 Let’s talk about data, right?
0:07:51 I feel like discussions about AI, for me,
0:07:54 tend to get more interesting when we get into data.
0:07:56 And it seems like that’s where a lot of the action is.
0:07:59 And from what I understand about the story
0:08:02 of your company, kind of building the data set
0:08:04 is a big part of the story and a big part of what
0:08:05 has differentiated you.
0:08:07 So you have all these data scientists.
0:08:09 What they need is data.
0:08:11 How do you go about building this data set
0:08:12 to find these metals?
0:08:14 Yeah, it’s an incredibly good question.
0:08:19 So most of the data we use was collected
0:08:21 by other people at other times.
0:08:24 Humans have been collecting information
0:08:26 about the physics and the chemistry of the Earth’s crust
0:08:29 for a very, very long time.
0:08:32 They’ve been doing it for, well, in some sense, for millennia.
0:08:34 But certainly over the last century,
0:08:38 they’ve been doing it in ever more sophisticated ways.
0:08:42 And for reasons I can explain, almost all of that data
0:08:45 is actually in the public domain.
0:08:49 The problem is it is a utter mess.
0:08:53 It is like an n-member hard, messy data problem.
0:08:58 Think of different humans in different decades
0:08:59 speaking different languages in different places
0:09:03 of the world, collecting different types of data–
0:09:05 and I’ll get into the types of data in a moment–
0:09:08 with different measurement techniques
0:09:11 based on the vintage of the era.
0:09:13 And then storing it in all manner of storage media.
0:09:18 Everything from literally handwritten geologic notes
0:09:20 or handwritten drilling notes all the way
0:09:26 to cloud data structures and everything in between.
0:09:28 And so it is this incredible mess of data.
0:09:30 Give me some specific examples.
0:09:31 What are specific–
0:09:34 like, did you find stuff in a drawer or something?
0:09:36 Like, give me some specific examples.
0:09:39 So I’ll give you examples of there’s
0:09:47 geologic libraries, archives, with carefully constructed
0:09:52 geologic maps that might be 100 years old.
0:09:55 And they were a smart, skilled geologist
0:09:59 to make a doing field mapping, which basically means
0:10:03 observing and recording the observations of outcrops
0:10:06 and describing the rock types and those outcrops
0:10:09 and locating them in space.
0:10:11 And the Earth’s crust changes very slowly.
0:10:17 So provided that was well done 100 years ago,
0:10:18 it’s still valid.
0:10:21 It’s just that it’s literally in drawers,
0:10:27 piled on top of each other, and basically not used.
0:10:29 It would only be used by a very industrious human
0:10:34 being who spent countless hours sort of searching
0:10:35 through the old archives.
0:10:39 So we go to various archives, and we
0:10:43 make an arrangement to digitize the information at our expense.
0:10:47 And we give the owners a full digital copy.
0:10:50 It’s almost always public domain data.
0:10:52 And so we have a right to use it,
0:10:55 or we negotiate a specific use right.
0:11:01 So digitizing a geologic map is the very, very beginning.
0:11:03 Then you need to extract the information
0:11:06 from the digital copy of the map.
0:11:09 And you have many different types of information there.
0:11:12 You could have, in the paper records,
0:11:14 you might have chemical assays, so measurements
0:11:19 of the concentrations of the elemental concentrations of samples
0:11:22 taken from different locations on the map.
0:11:24 And that could be a part of the record.
0:11:27 So that’s tabular data, because it’ll say, well, this sample,
0:11:31 sample whatever, had x% calcium, and y% magnesium,
0:11:34 and z% silica, and et cetera, et cetera, et cetera.
0:11:36 That’s all valuable information.
0:11:39 So that’s tabular information that then gets extracted
0:11:43 by our systems and populated into what
0:11:46 we call our universal schema, which just
0:11:50 means that every data type is stored in a consistent format.
0:11:54 You’re standardizing this wildly messy heterogeneous data.
0:11:56 You’re trying to make it more homogenous.
0:11:57 That’s exactly right.
0:11:59 And we should talk about more about what the data is,
0:12:00 because it’s really fascinating.
0:12:02 So I gave you two examples.
0:12:05 I gave you the sort of qualitative, almost like drawn,
0:12:07 geologic map, which is incredibly useful information.
0:12:10 But qualitative and continuous in nature.
0:12:12 Then there’s the sort of tabular data
0:12:14 that would be any kind of assay data,
0:12:16 measurements of composition.
0:12:18 But then you have a whole different classes of data,
0:12:20 like geophysical data, which tells you something
0:12:22 about the physics of the Earth’s crust.
0:12:25 So for example, you probably know
0:12:28 that the Earth’s gravitational field changes from place
0:12:29 to place as you move around.
0:12:32 It changes because you can go up or down an elevation.
0:12:35 Well, that’s easy to adjust for, because you know the elevation.
0:12:37 It also changes because the density of the rocks
0:12:39 below you change.
0:12:43 And so if you’re standing over an ore body that
0:12:47 has twice the density of whose rocks are twice as dense
0:12:51 as the surrounding rocks, that’ll pull on you slightly more.
0:12:52 And you can measure that.
0:12:54 That I did not know.
0:12:56 And let’s go down this rabbit hole, actually,
0:12:57 because it’s super interesting.
0:12:58 OK.
0:12:58 I’m in.
0:13:00 Because imagine you make this measurement.
0:13:02 What are you actually measuring?
0:13:03 You’re measuring the force of gravity
0:13:05 in a particular location.
0:13:07 And you can measure, OK, I’ve adjusted
0:13:08 the elevation and the force of gravity
0:13:10 is a little bit higher here.
0:13:13 OK, that’s all you actually know at this moment.
0:13:14 So what is that telling you?
0:13:17 Is it telling you you have a modestly more dense object,
0:13:19 like just below the surface?
0:13:22 Or is it telling you you have a massively more dense object
0:13:23 deeper?
0:13:26 Well, it turns out that this is a fundamentally degenerate
0:13:28 problem, or non-unique problem.
0:13:30 It would be the way you describe it mathematically.
0:13:32 There are many ways to solve that problem.
0:13:34 There are many different configurations.
0:13:35 You don’t know the answer.
0:13:36 You don’t know the answer.
0:13:41 What you do know is that there’s a very large class
0:13:44 of invalid solutions.
0:13:48 And then there’s a smaller, but still very large class,
0:13:49 of valid solutions.
0:13:51 There’s a lot of things that it is not,
0:13:53 and there’s some things that it could be.
0:13:54 Could– exactly.
0:13:55 Incredibly well said.
0:13:57 That’s exactly right.
0:13:59 So here’s a really cool application
0:14:01 of our technology and our approach.
0:14:05 So the industry standard approach to this is basically,
0:14:07 what does a normal conventional company
0:14:09 do with this gravitational anomaly?
0:14:10 Well, they do one of two things.
0:14:13 Most of the time, maybe 90% of the time,
0:14:15 they’ll just look at it and say, OK,
0:14:16 here is a gravitational anomaly.
0:14:18 This is higher here than here.
0:14:19 That’s interesting.
0:14:22 And they just see it on a 2D map.
0:14:24 So that doesn’t do anything for the non-uniqueness.
0:14:27 It just tells you what the measurement is.
0:14:28 So they would just use it.
0:14:29 They would just say that.
0:14:31 Well, it’s interesting because it’s higher.
0:14:34 OK, so what Kobold does is very, very different.
0:14:40 And impossible to have done even 10 years ago, maybe even
0:14:41 five years ago.
0:14:44 What we do is we solve, conditioned
0:14:47 on a set of geologic hypotheses that we find interesting,
0:14:53 we solve for the full set of possible subsurface.
0:14:57 So we might actually test like a billion subsurface
0:15:06 and say 999,999,900,000 of those don’t match the data.
0:15:07 We tried, they don’t match it.
0:15:09 So they’ve been rejected.
0:15:11 But we still have 100,000 now that do.
0:15:13 So now we have all of the good–
0:15:15 we’ve rejected the bad possibilities
0:15:17 and we’ve narrowed on the good possibilities.
0:15:19 But it’s still an incredibly large subspace.
0:15:19 Yes.
0:15:21 Impractical still.
0:15:24 It’s got to get a lot smaller for you to do anything like it.
0:15:27 But there’s a lot of information in the uncertainty
0:15:28 that we’ve now quantified.
0:15:30 We’ve now quantified the uncertainty.
0:15:31 And so we apply something that we
0:15:34 call efficacy of information, which
0:15:36 is a phrase that we coined.
0:15:37 And you can read scientific papers about it.
0:15:40 It’s a very general and really cool concept.
0:15:42 And it’s also kind of obvious, actually,
0:15:43 although it’s hard to formalize.
0:15:45 But you look very happy right now.
0:15:46 I get really excited about this.
0:15:48 Because this is an audio medium.
0:15:50 Like, your face is full of delight right now.
0:15:51 I don’t want to interrupt you.
0:15:51 Keep going.
0:15:52 That’s good.
0:15:53 Yeah, I’m excited.
0:15:55 I love talking about this stuff, because it’s super cool.
0:15:58 So we say, OK, the basic idea behind EOI
0:16:02 is, you’re going to collect some information next.
0:16:02 OK.
0:16:05 And that’s really what every exploration process always is.
0:16:05 Right.
0:16:06 Is a sequential set of decisions.
0:16:08 You’re going to go try and get more information
0:16:10 to figure out what is going on in the rocks
0:16:11 under the surface.
0:16:15 So the question you obviously want to ask
0:16:18 is, what next piece of information
0:16:21 will tell me the most?
0:16:26 Given, sort of, per unit of dollar that I expend,
0:16:27 what am I going to learn the most from?
0:16:30 What has the highest return on investment?
0:16:33 Yeah, and from a knowledge and information perspective,
0:16:35 what am I going to learn the most from?
0:16:36 And so here’s a way to think about that.
0:16:38 It is the piece of information that
0:16:40 decreases your uncertainty the most.
0:16:46 And because we have the 100,000 possible answers,
0:16:49 the piece of information that will decrease our uncertainty
0:16:52 the most is, in fact, the piece of information
0:16:57 that tests the most number that falsifies the highest number
0:16:58 of those 100,000.
0:17:01 So for instance, say we’re going to drill
0:17:03 in different directions.
0:17:04 And so we’re going to drill, and we’re
0:17:08 going to intersect the various predictions of concentration.
0:17:12 If we can do one drill hole and it would test 100 of the 100,000.
0:17:13 Yeah.
0:17:18 That’s only– we’re only testing 0.1% of the possible answers.
0:17:20 Don’t drill there.
0:17:20 Don’t do that.
0:17:21 We’re going to learn very little.
0:17:23 We’re going to end up with basically
0:17:24 the same amount of uncertainty.
0:17:27 But if we could drill a different hole, a different core,
0:17:30 that tests 50,000, say, half of them,
0:17:32 we massively reduce our search space.
0:17:37 We falsify 50% of the possible answers.
0:17:40 And sometimes we can actually falsify like 80% and 90%
0:17:41 of the possible answers.
0:17:44 So we massively reduce our search space.
0:17:48 And fundamentally, it’s the most possible information
0:17:49 you can get per unit dollar.
0:17:52 So every time we go to collect any information,
0:17:58 we try to formally quantify the uncertainty
0:18:01 and then calculate this EOI content of term, which
0:18:03 is the piece of information that we have the greatest
0:18:07 expectation will reduce our uncertainty the most.
0:18:08 Compelling.
0:18:09 I’m glad you think so.
0:18:13 It would be nice to be able to do that in life more generally.
0:18:14 It’s super, super hard.
0:18:16 And it comes– the really hard part
0:18:19 is quantifying the uncertainty correctly.
0:18:21 Once you have that quantified correctly,
0:18:24 then calculating the EOI, if you know for sure,
0:18:28 you’ve correctly quantified the uncertainty,
0:18:31 calculating the EOI is kind of an engineering optimization.
0:18:33 It’s kind of straightforward.
0:18:35 In this instance, quantifying the uncertainty
0:18:39 is basically how many ways could the rock under the surface
0:18:39 be?
0:18:40 Exactly.
0:18:40 Yeah.
0:18:41 That’s exactly right.
0:18:43 So you’ve been talking about sort
0:18:45 of gathering this very old school data
0:18:49 and making it useful to you and then what you do with it.
0:18:53 There is another piece of your data gathering operation
0:18:57 or another set of pieces that are more high tech
0:18:59 and that involve going out into the world and getting new data
0:19:01 that doesn’t already exist.
0:19:03 And some of those are kind of fun.
0:19:06 And so I want to talk about that a little bit.
0:19:06 Sure.
0:19:08 Tell me about detecting muons.
0:19:14 Yeah, so this is kind of one of our frontier R&D projects
0:19:16 within the company, something we would–
0:19:19 The opposite of a 100-year-old map in a drawer somewhere.
0:19:19 Yeah, totally.
0:19:20 Exactly.
0:19:21 Exactly.
0:19:23 And we have a lot of physicists at the company,
0:19:24 so they love this stuff.
0:19:25 So what is a muon?
0:19:27 Let’s start with that.
0:19:29 And then I’ll get to why they can be useful.
0:19:34 So cosmic rays are hitting air molecules
0:19:37 in the upper atmosphere all the time.
0:19:39 And when they collide, sometimes
0:19:42 they produce muons in the reaction.
0:19:47 So it’s a subatomic particle, a muon.
0:19:50 And it travels very, very, very fast.
0:19:52 It’s a sort of relativistic particle.
0:19:54 It travels near the speed of light.
0:19:57 So right now, muons are showering through us.
0:20:00 You and you in your studio and me in my home,
0:20:01 muons are coming through us.
0:20:04 And I think about– if you put your hand flat,
0:20:06 you can expect about one muon per second
0:20:08 to be going through your hand.
0:20:10 You don’t notice they mostly go right through you.
0:20:13 It doesn’t cause you any harm.
0:20:15 They do interact with electrons.
0:20:18 And it turns out that every time they
0:20:22 interact with electron, they slow down a little bit.
0:20:25 And when they slow down, eventually they slow down
0:20:27 enough that they decay into other things.
0:20:29 So they disappear.
0:20:32 So let’s say you have a muon detector
0:20:34 and it’s sitting at the surface.
0:20:37 And you’re listening– you listen to its detection.
0:20:39 So it’s like click, click, click, click.
0:20:41 That’s just telling you, OK, muon’s going through,
0:20:43 muon’s going through, muon’s going through, right?
0:20:46 Now, I drill a borehole.
0:20:50 And I start lowering the muon detector into the borehole.
0:20:51 And you know the rate it was at the surface.
0:20:55 Then as I get to say 100 meters, now it’ll be like click,
0:20:57 click, click.
0:20:59 And then as I go to like 500 meters,
0:21:04 it’ll be like click, click, OK.
0:21:06 So what’s happening is getting fewer and fewer muons
0:21:07 are hitting that location.
0:21:10 And the reason is because you’ve got so many more atoms
0:21:13 between you and the top of the atmosphere
0:21:16 that the muon just aren’t surviving.
0:21:17 They’re hitting the rocks.
0:21:18 They’re hitting the rocks, OK?
0:21:22 And so now you think of the journey of a specific muon
0:21:24 as it’s going through the rock.
0:21:29 The muon that interacts with the fewest electrons
0:21:31 is the most likely to hit you.
0:21:35 And the muon that interacts with the most atoms, we’ll say,
0:21:38 is less likely to hit you, because it’s likely to decay,
0:21:41 because it’s likely to basically lose its energy.
0:21:46 Think of a ball bouncing, hitting a bunch of other balls.
0:21:49 So now if you’re sitting at the muon detector that’s
0:21:52 been lowered into a location underground,
0:21:54 and you’re looking up in kind of a cone,
0:21:58 and you can look in three dimensions,
0:22:02 you’re seeing muons say they’re coming from the right a lot,
0:22:04 but they’re not coming from the left very much.
0:22:05 Yes, yes.
0:22:08 That’s telling you something about the relative number
0:22:11 of atoms to the left, which tells you about the density.
0:22:14 That tells you the rocks above you
0:22:18 and to the left of you are denser than the rocks
0:22:19 up and to the right of you.
0:22:22 And denser might mean an ore body.
0:22:25 It might mean a high concentration set of metal
0:22:26 in the rock.
0:22:31 And so it allows us to probe, in really sophisticated ways,
0:22:33 the density of the earth.
0:22:36 In the same way we were talking about the gravitational force
0:22:37 changes around the earth.
0:22:39 It’s the same thing, but it’s a much higher precision
0:22:40 measurement.
0:22:43 So we’ve designed our own novel muon detector.
0:22:46 We did it in collaboration with Occidental College.
0:22:50 And it’s working, and it’s in a pilot hole collecting muons.
0:22:52 And then we have a bunch of new ideas
0:22:57 about how to use this new kind of new data type.
0:23:00 There’s a few other companies doing this,
0:23:01 but it’s very new, very new concept.
0:23:08 After the break, from theory to practice,
0:23:13 Kurt talks about Kobold’s huge copper discovery in Zambia.
0:23:16 [MUSIC PLAYING]
0:23:23 Let’s talk about Zambia.
0:23:25 Sure.
0:23:28 I want to talk about Zambia because it suggests
0:23:34 that your hypothesis for the company is worked, right?
0:23:36 Yeah, to a large degree, yes.
0:23:40 Let me first say, why are we in Zambia in the first place?
0:23:43 So we look around the world, and we
0:23:49 evaluate jurisdictions on four dimensions, OK?
0:23:51 Our physical prospectivity, or how
0:23:53 we perceive the physical prospect,
0:23:54 the likelihood that there’s going to be something new
0:23:58 to discover in a particular location.
0:23:58 That’s thing one.
0:24:01 Thing two is the rule of law.
0:24:04 If we make a discovery, we have a property, right?
0:24:06 How robust is that property, right?
0:24:07 That’s thing two.
0:24:10 Thing three is access to markets, infrastructure, right?
0:24:13 If you find something in the middle of Antarctica,
0:24:15 you’re not going to be able to get it to the market,
0:24:16 no matter how great it is.
0:24:21 And then thing four is, how much resistance or/support
0:24:23 will there be to building a new industrial project
0:24:25 in that location, right?
0:24:26 If you find something in Palo Alto,
0:24:28 no one’s going to ever allow you to build it, right?
0:24:29 So it just doesn’t matter.
0:24:31 You can’t even build an apartment building there,
0:24:32 much less a lithium mine.
0:24:33 Exactly.
0:24:34 Exactly.
0:24:36 So those are the four dimensions we look at.
0:24:38 And we looked around the world early on.
0:24:40 Zambia rose to the very top on all four of those.
0:24:43 It’s a fantastic jurisdiction.
0:24:46 It’s the most consistent and stable democracy
0:24:47 in Southern Africa.
0:24:49 The physical prospectivity is tremendous,
0:24:52 because there’s been mining there for 100 years.
0:24:55 And so you might look at that and say, well, sure,
0:24:58 it was a good place to look 100 years ago,
0:25:00 but isn’t it all picked over?
0:25:02 And the answer is definitively not.
0:25:07 This is easy to verify, because basically all the deposits
0:25:10 that were mined over the last 100 years
0:25:12 were actually sticking out of the surface.
0:25:15 They were known about a century ago.
0:25:17 And there’s been almost no exploration
0:25:20 into the deep parts of the basins,
0:25:22 what we’d call blind exploration, right?
0:25:24 This is not directly evident, like you
0:25:26 can see it at the surface.
0:25:27 There’s been almost none.
0:25:30 It was this kind of perfect location in that sense.
0:25:33 It’s right on the central African copper belt, which
0:25:36 provides a significant majority of the world’s copper.
0:25:37 So it’s easy to get it to market, right?
0:25:40 And it’s a legacy mining country that’s
0:25:42 very supportive of development, right?
0:25:43 And so it’s like perfect.
0:25:46 It rose to the top across the board, and we’d love it.
0:25:49 And we love our Zambian colleagues.
0:25:51 And we think it’s one of the best jurisdictions
0:25:54 in the world for us to operate.
0:25:56 So that’s why we were there in the first instance.
0:26:04 And we started exploring in 2020 there in very modest fashion
0:26:08 loosely.
0:26:10 But we were exploring in areas right
0:26:13 in the heart of active mining basins.
0:26:16 Because again, there were active mines.
0:26:19 But if you went out into the deeper parts of the basin,
0:26:21 deeper than 500, 600 meters, there
0:26:24 were areas that just had never been probed or explored.
0:26:28 Yeah, so briefly, what did you find and how did you find it?
0:26:31 Looking at all the data, some of our geoscientists
0:26:36 had really, really clever ideas about how the mineralogy was
0:26:40 changing and how we actually might have very, very distinct
0:26:41 mineralogy.
0:26:43 So we might have areas where it looks
0:26:47 like it’s all this kind of one distribution, one sort
0:26:48 of set of statistics.
0:26:50 But actually, as you cross this boundary,
0:26:53 it’s a totally different geochemical system.
0:26:55 Different sets of geochemical reactions occurred.
0:26:58 So if you’re able to draw that boundary and then only go
0:27:00 and explore within that complicated three-dimensional
0:27:02 boundary, then you’d consistently
0:27:04 have high-grade and thick.
0:27:06 That was the assertion.
0:27:11 What you have in the reservoir, in the legacy data,
0:27:17 95% of the data is this low, modest-grade thin stuff.
0:27:20 And then there’s 5% of the data is this higher-grade stuff.
0:27:23 And if we drill there and within that boundary
0:27:26 and only drill there, it’s going to be good hole, good hole,
0:27:29 good hole, good hole, good hole.
0:27:30 So that’s the hypothesis?
0:27:31 Yes.
0:27:33 You test the hypothesis, and does it happen?
0:27:33 And we’re right.
0:27:34 Yeah, exactly.
0:27:35 So you found it.
0:27:35 Yes.
0:27:37 And we proved it, and it’s there.
0:27:41 And it’s gone from marginally economic to very economic.
0:27:44 For the same unit of rock that we move,
0:27:49 we sell 10 times as much copper as the average copper mine
0:27:49 today.
0:27:50 Yes.
0:27:50 OK.
0:27:51 So you found it.
0:27:51 Correct.
0:27:53 The hypothesis was true.
0:27:55 The system worked.
0:27:55 You found it.
0:27:56 Correct.
0:28:01 Now you are going to become a wildly different company,
0:28:02 it seems to me.
0:28:04 Like you are in Silicon Valley.
0:28:06 You are working with data scientists.
0:28:07 You have a technical background.
0:28:11 You have been running, essentially, a high-tech startup.
0:28:13 You are about to be running a mining company,
0:28:16 where your problems are not just being very clever
0:28:17 and hiring the right AI people.
0:28:21 They’re like getting the US government
0:28:25 to finance a railroad in Zambia and making sure
0:28:27 that the Zambian’s like you.
0:28:30 Like that seems entirely different than what
0:28:33 you have been doing so far.
0:28:37 Partially true.
0:28:39 It’s a little more continuous than you might have implied.
0:28:41 Yes, I’m just trying to make it a good story.
0:28:45 But look, it’s not a regular startup anymore.
0:28:46 Correct.
0:28:48 Like maybe it never was.
0:28:50 But it’s a super different job.
0:28:52 It’s a super different skill set.
0:28:54 I mean, I’m sure to some extent you’ve been dealing with this.
0:28:56 But it’s really different than what
0:28:57 we have been talking about.
0:29:00 It’s a whole different universe of problems and hard things
0:29:02 to deal with and a very different domain.
0:29:03 Sure.
0:29:06 No, these are excellent questions.
0:29:08 So you could think about the company now
0:29:11 as there is the discovery machine.
0:29:13 The discovery machine is everything
0:29:14 we’ve been talking about.
0:29:17 And the discovery, the unit economics of exploration
0:29:22 are great because you can make 100 times your money
0:29:25 or even more on proving the deposit.
0:29:28 Because it’s worth so much once it’s clearly economic.
0:29:28 Right.
0:29:30 So you could just sell the rights in some fashion, right?
0:29:32 You don’t have to be a mining company.
0:29:34 You could be a discovery company, right?
0:29:35 And it’s correct.
0:29:38 So the most important part of the company, the heart and soul
0:29:41 of the company, is the discovery machine.
0:29:44 Now we have at least one deposit that
0:29:47 is sort of unambiguously going to be a mine.
0:29:51 And the question is what happens from here with that mine?
0:29:53 And the odds are very high that we’re
0:29:57 going to bring in a partner with complementary capabilities
0:29:59 to sort of help bring it to production
0:30:01 and bring in a partner.
0:30:03 It’s a simple way to think of this
0:30:06 as we own 80% of it now.
0:30:10 The Zambian Peristatal Mining Company owns 20% of it.
0:30:11 It’s worth a certain amount.
0:30:15 You could imagine what someone would pay to own it entirely.
0:30:18 And that order of magnitude is billions of dollars?
0:30:18 Correct.
0:30:19 Yeah.
0:30:21 Yeah.
0:30:24 And so someone could come in and pay and contribute
0:30:27 to capital and capabilities to bring it into a mine.
0:30:30 And we could, in principle, spend no more money.
0:30:34 And we still have a large share of all the future cash flows.
0:30:37 It’s worth a lot to have the stake you have in this thing.
0:30:39 You have a set of choices about what to do with it
0:30:42 and how much sort of money to take
0:30:43 and how much of your interest to sell.
0:30:44 Correct.
0:30:45 That’s exactly right.
0:30:47 So is the answer you’re trying to figure out
0:30:49 how much you’re going to be a mining company
0:30:51 versus a discovery company?
0:30:52 I mean, is that–
0:30:52 Yeah.
0:30:55 And we know for sure that the most important thing is
0:30:59 to not weaken the discovery engine.
0:31:00 That’s the most–
0:31:02 and there’s a lot of culture around the discovery engine.
0:31:03 There’s a lot of attention.
0:31:04 That’s the most important thing.
0:31:09 We also know that it’s very, very important
0:31:14 that the value potential gets realized in the Zambian
0:31:15 deposit.
0:31:17 Is there about to be some deal?
0:31:19 If this show comes out in two weeks,
0:31:21 will there be some news between now and then?
0:31:23 No, and I can confirm that we’re not actually
0:31:26 looking for a partner for a couple of years, actually.
0:31:28 We’re not– we won’t formally partner with anyone
0:31:29 for a couple of years.
0:31:30 And the reason they’re actually really obvious,
0:31:33 it’s just that what we have discovered
0:31:36 sits on about four square kilometers.
0:31:38 And there’s another 150 square kilometers
0:31:41 on the license that we own that are totally untouched,
0:31:44 totally unexplored, completely unexplored.
0:31:47 So we are going to fully explore that whole area.
0:31:49 And totally know what we have before we formally
0:31:51 do any partnership.
0:31:53 But we’re also building the capabilities
0:31:57 to take the project as far as we want to.
0:31:59 So we recently hired–
0:32:01 we’ve stood up an amazing Zambian leadership
0:32:08 team of project developers and engineers,
0:32:10 metallurgists, et cetera, hydrologists
0:32:13 to continue to do the engineering optimization of what
0:32:16 a mine will look like in this location.
0:32:18 And so we’re doing all of that because that
0:32:19 has to be done anyway.
0:32:22 It adds a ton of value to figure out exactly how you’ll
0:32:25 optimize the operations.
0:32:28 And it just moves the project forward.
0:32:32 And our goal, our stated intention,
0:32:36 is to start construction on the mine within two years
0:32:39 and to have it in production in the early part
0:32:40 of the next decade.
0:32:43 So I know that we need a lot of copper.
0:32:47 And it’s great that you just found more.
0:32:50 It is also the case that mines have often
0:32:52 been bad for the places where the mines were
0:32:55 and for the people who worked in the mines.
0:32:59 Like, how do you deal with that, harming people and the world?
0:33:02 What’s super exciting about this particular deposit,
0:33:04 and in general, the deposits we’re looking for,
0:33:06 is it’s super high grade, 10 times higher grade
0:33:09 than the average copper mine around the world.
0:33:13 That means 10 times less waste for the same amount of copper.
0:33:16 It’s also an underground mine as opposed to an open pit mine.
0:33:18 So when you consider the overburden
0:33:20 that open pit mines, big holes in the ground,
0:33:22 would otherwise have, it actually ends up being about 30
0:33:25 times less waste.
0:33:27 And we’re going to take almost all of that waste.
0:33:33 And as we excavate the locations underground,
0:33:35 we take the waste and we put it back in.
0:33:37 We backfill is what it’s called.
0:33:39 We stuff it back into the area.
0:33:42 So at any given time, there’s only a modest volume,
0:33:45 modest cavity that’s open.
0:33:50 And then in terms of the, we are passionate and obsessed
0:33:51 with skills transfer.
0:33:53 And it’s just the reason that we are really building
0:33:55 a Zambian mining company to develop this project.
0:34:04 Our CEO, Makai Makai, she’s the CEO of Kobold Africa.
0:34:08 We have 90% of the employees in country are Zambian.
0:34:10 Chief metallurgist, chief mine engineer, chief project
0:34:18 director, a site geologist, all of them are Zambian.
0:34:20 And they’re extraordinary.
0:34:23 And we’re investing tremendous amounts
0:34:28 in helping them be the best professionals they can be,
0:34:31 because we’re going to be there for 50 years, right?
0:34:33 And we want to be there for 50 years.
0:34:38 So what’s the next big discovery?
0:34:45 So I will predict nickel in Canada,
0:34:57 lithium in Australia, and another copper discovery in Zambia.
0:35:00 I appreciate the specificity of that.
0:35:03 I love falsifiable predictions.
0:35:04 God bless you.
0:35:08 Something I make wagers all the time.
0:35:09 Weird wagers.
0:35:12 But it’s not because I’m a gambling man.
0:35:14 I’ve never played a spin of roulette in my life.
0:35:16 But what I love to do is people make vague predictions
0:35:16 about the future.
0:35:19 And I try to pin them down on something
0:35:20 that is clearly testable.
0:35:24 And then we usually bet a nice bottle of wine or a dinner
0:35:24 that we enjoy together.
0:35:26 So it’s fun.
0:35:28 But the loser pays, obviously.
0:35:34 And yeah, so we have, in fact, a big part of our company
0:35:36 culture is what we call the culture of falsification.
0:35:38 So when you go out to test the hypothesis,
0:35:40 your job is not to collect information
0:35:42 to confirm that hypothesis.
0:35:43 Because you can always do that.
0:35:45 You can always paint a new story.
0:35:46 That’s inductive reasoning.
0:35:47 It’s invalid.
0:35:49 Your job is to go out, is to tell me
0:35:51 how you’re going to test it, how you’re going to prove it
0:35:53 wrong, and go falsify it.
0:35:54 And one of two things happens.
0:35:57 You either successfully falsify it, in which we move on,
0:35:58 and we celebrate that.
0:35:59 We celebrate falsification.
0:36:03 Or you fail to falsify it, which means, OK, so it’s not
0:36:05 dead yet.
0:36:06 Now how do we test it now?
0:36:07 How do we test it again, right?
0:36:10 Yes.
0:36:11 And what’s the most efficient way?
0:36:14 What’s the highest return we can get on the question?
0:36:15 What’s the EOI?
0:36:15 You got it.
0:36:21 We’ll be back in a minute with the lightning round.
0:36:35 OK, we’re going to finish with the lightning round.
0:36:40 And so I got to ask, what is one weird bet you made?
0:36:45 Since we just had the Olympics, I bet that Usain Bolt’s
0:36:49 9.58 seconds world record in the 100 meter
0:36:52 will still be the world record in the year 2036,
0:36:53 by the end of the year 2036.
0:36:54 That’s a long bet.
0:36:55 You’re playing the long game.
0:36:58 Yeah, that’s going to be a longstanding record.
0:36:59 He said it in 2009.
0:37:02 It’s crazy for a record to last that long.
0:37:05 Yeah, would you like to make that wager with me?
0:37:06 I don’t have enough information.
0:37:08 I know enough to know that I’m ignorant.
0:37:10 But what made you make the bet?
0:37:12 It’s a dramatic outlier time.
0:37:14 It’s a total outlier.
0:37:16 Yeah, and he has the four fastest times.
0:37:18 So nobody– the next fastest human
0:37:20 is going to be the fifth fastest time in the world.
0:37:24 It’s like he is an outlier, and that time is an outlier for him.
0:37:25 Correct.
0:37:29 Correct.
0:37:30 What’s a cobalt?
0:37:32 Oh, good question.
0:37:37 So it’s a creature from German mythology
0:37:41 that lives underground– kind of like a goblin-like creature–
0:37:44 lives underground and controls the mineral wealth
0:37:46 of the earth.
0:37:47 Uh-huh.
0:37:50 And it’s also the namesake for the word cobalt,
0:37:53 for the metal cobalt.
0:37:55 Right, I mean, as I understand it,
0:37:57 people used to think cobalt was bad, right?
0:38:01 And then now we’re like, oh, actually, cobalt is good.
0:38:03 It looks a lot like nickel sulfides
0:38:07 when it’s a cobalt arsenide, and arsenic is toxic.
0:38:10 So it would poison miners.
0:38:12 And they called it the goblin metal.
0:38:14 What’s one thing I should do if I go to Zambia?
0:38:17 Oh, I love that question.
0:38:18 Well, you can’t miss Mosiotunia.
0:38:20 I call it– we call it Mosiotunia.
0:38:22 That’s the traditional name.
0:38:24 It means the smoke that thunders.
0:38:26 You will know it as Victoria Falls.
0:38:29 It is completely spectacular.
0:38:30 Niagara Falls is amazing.
0:38:32 It blows it away.
0:38:33 Totally blows it away.
0:38:39 You just can’t go to Zambia and miss Mosiotunia.
0:38:43 We’ve talked a lot about trying to predict things
0:38:46 and trying to quantify uncertainty
0:38:48 in the context of your company.
0:38:52 Do you think that way outside of work?
0:38:58 Well, yeah, I guess the way I make–
0:39:03 wager on things is a form of–
0:39:06 it’s certainly the way I think of–
0:39:09 I try to be scientific in every aspect of my life.
0:39:16 And I say that what science is not is empiricism.
0:39:18 It is not looking at the data and drawing
0:39:20 the inevitable conclusion.
0:39:21 There’s no such thing.
0:39:23 You can look at– with any set of data,
0:39:27 you can fit many, many, many hypotheses
0:39:28 that explain the data.
0:39:31 That’s always– this is the non-unique thing,
0:39:33 where it’s always true, basically.
0:39:34 It’s always true.
0:39:36 And so science really is about myth-making.
0:39:40 It’s about making up a myth that explains the data.
0:39:41 That’s your hypothesis.
0:39:44 The difference between science and religion
0:39:46 is that we test our myths.
0:39:47 That’s the difference.
0:39:49 That’s the distance of good science, right?
0:39:52 And in your heart, you should want to disprove it, right?
0:39:55 Like, if you’re really the best scientist,
0:39:58 you should want to prove yourself wrong.
0:39:59 Yes.
0:39:59 That’s this.
0:40:01 The best thing a scientist can say.
0:40:02 That means you learn something.
0:40:05 You really only learn when you realize you’re wrong.
0:40:12 Kurt Haus is the co-founder and CEO of Kobold Metals.
0:40:15 Today’s show was produced by Gabriel Hunter Chang.
0:40:18 It was edited by Lydia Jean Cotte and engineered
0:40:20 by Sarah Bouguere.
0:40:24 You can email us at problem@pushkin.fm.
0:40:26 I’m Jacob Goldstein, and we’ll be back next week
0:40:29 with another episode of What’s Your Problem?
0:40:32 [MUSIC PLAYING]
0:40:35 [MUSIC PLAYING]
0:40:38 (upbeat music)
0:00:06 Pushkin.
0:00:13 Digging up metal from out of the ground
0:00:16 is a business that is literally thousands of years old.
0:00:19 But mining suddenly has new importance.
0:00:22 The energy transition, going from fossil fuels
0:00:24 to renewable energy, is going to take
0:00:29 a ridiculous amount of metal, metal like copper and lithium.
0:00:31 The need is so great and so urgent
0:00:33 that we’re going to have to come up with new ways
0:00:36 to find metal buried in the earth.
0:00:39 And as it happens, a new kind of mining company,
0:00:43 a mining company you might call an AI-driven startup,
0:00:47 just made the biggest copper discovery in over a decade.
0:00:49 It’s worth tens of billions of dollars.
0:00:52 [MUSIC PLAYING]
0:00:57 I’m Jacob Goldstein, and this is What’s Your Problem,
0:00:59 the show where I talk to people who are trying
0:01:01 to make technological progress.
0:01:05 My guest today is Kurt Hauss, the founder and CEO
0:01:07 of Cobald Metals.
0:01:09 Kurt’s problem is this.
0:01:13 How do you use AI, machine learning, data science,
0:01:17 to find the metals we need for the energy transition?
0:01:18 As you’ll hear, my conversation with Kurt
0:01:23 goes beyond mining and AI to cover Kurt’s really compelling way
0:01:27 of just thinking about making decisions in an uncertain world.
0:01:30 We started, though, by talking about how he came up
0:01:31 with the idea for his company.
0:01:39 So if you go back about eight years ago,
0:01:43 my co-founders and I were looking at the trends
0:01:47 in the energy transition, seeing the electric vehicle
0:01:49 and renewable energy sort of revolutions coming.
0:01:55 And it’s quite easy to convince yourself
0:01:59 that the material requirements for the energy transition
0:02:03 will be tremendous, the amount of very specific materials
0:02:04 that the world needs–
0:02:09 copper, lithium, cobalt, nickel, graphite, others.
0:02:13 This is basically stuff to build batteries and wires,
0:02:14 essentially.
0:02:17 This is just batteries and electrification, right?
0:02:19 Just an electric motor is a bundle
0:02:21 of copper wire surrounded by, you know,
0:02:23 surrounding a permanent magnet.
0:02:26 Every battery require– every mobile battery
0:02:30 requires lithium and nickel and cobalt.
0:02:31 These are all the–
0:02:34 these are the key materials for which, in some cases,
0:02:35 like the humanity has been using lots of copper
0:02:38 for a long time– there’s a big copper market–
0:02:42 but it needs to at least double from a very large base.
0:02:46 Lithium, humanity has not been using much lithium for very long.
0:02:50 And now the lithium market needs to grow by well more
0:02:57 than a factor of 10 to fully electrify just the transportation
0:02:59 sector.
0:03:02 So the sort of macro needs were very obvious.
0:03:04 So that’s observation one.
0:03:08 Observation two– say, OK, well, maybe the incumbents
0:03:10 are really good at finding new materials.
0:03:13 And as prices rise a little bit, they’ll find new materials,
0:03:14 and the market will just be well supplied.
0:03:17 And that turns out to be definitely wrong.
0:03:20 And it’s actually really easy to verify that it’s wrong,
0:03:23 because the large, well-resourced mining companies
0:03:26 basically don’t even do exploration, actually.
0:03:31 The big mining companies, they spend $65 billion a year
0:03:34 on dividends and share buybacks and less than half a billion
0:03:35 on exploration activities.
0:03:38 But that half a billion, that’s the deployment
0:03:42 of conventional exploration technologies, right?
0:03:46 Things that would be natural to most geologists
0:03:49 from the 1960s or ’70s, right?
0:03:51 So you can round to zero how much money they’re
0:03:53 spending on research and development
0:03:55 for new techniques and new technologies
0:03:57 to improve the exploration process.
0:04:00 So it was basically those sets of observations,
0:04:02 those two sets of observations.
0:04:04 We need metals, and nobody’s really looking for them.
0:04:05 They’re not looking for them, and they’re certainly not
0:04:06 getting better at it.
0:04:07 They’re getting worse at it.
0:04:13 We call that trend and the increasing cost of discovery.
0:04:17 We call that e-room’s law of mining.
0:04:18 Moore’s law backwards?
0:04:18 Very good.
0:04:19 Yeah, I’m impressed.
0:04:20 Yeah, exactly.
0:04:22 They talk about that in biotech as well.
0:04:24 We borrowed it from biotech.
0:04:29 Meaning, whereas microchips get cheaper and better every year,
0:04:32 mining gets worse and slower and more expensive every year.
0:04:36 It’s specifically exploration and discovery, specifically.
0:04:38 So those were the major needs.
0:04:40 So then you say, OK, what can we do?
0:04:41 How can we do?
0:04:43 What can we do differently?
0:04:44 How can we help?
0:04:50 And the answer is that exploration is fundamentally
0:04:52 an information problem.
0:04:53 Fundamentally, right?
0:04:55 We know for deep physical reasons,
0:04:59 which I can explain in a minute, we know there’s gobs
0:05:03 and gobs of undiscovered rich deposits out there.
0:05:05 We don’t know where they are.
0:05:08 So the gap is the knowledge of where they are.
0:05:13 If God gave you a perfect model of the Earth’s crust,
0:05:15 the location and form of every atom,
0:05:16 you’d be a perfect explorer.
0:05:20 You’d know where all the high-grade, high-concentration
0:05:21 anomalies were.
0:05:23 You’d also be a perfect miner.
0:05:26 The miner’s religious vision is the gift
0:05:28 from God of perfect information, yes.
0:05:30 Perfect, exactly.
0:05:31 But it’s not that.
0:05:32 So we don’t have that.
0:05:34 So we have a huge amount of uncertainty.
0:05:38 But the sort of managing the data that you have,
0:05:42 and then making probabilistic inferences on that data,
0:05:46 is fundamentally an information problem.
0:05:50 We look at it as this is kind of a perfect tailored
0:05:54 application for data science and modern scientific computing.
0:05:57 It’s a little different.
0:06:00 It has some sort of unique, really cool attributes to it.
0:06:02 But it is fundamentally an information problem
0:06:04 and fundamentally a search problem.
0:06:08 And so the thing that could be massively different
0:06:12 would be a company built from the ground up,
0:06:15 a Silicon Valley company built from the ground up
0:06:18 that combines the best existing knowledge
0:06:23 of geoscientists with world-class data scientists
0:06:27 and software engineers coming out of the major tech monopolies–
0:06:31 Google, Apple, Facebook, you name it–
0:06:33 who have never worked in the metals and mining business
0:06:34 before, right?
0:06:36 So it’s fundamentally sort of bringing
0:06:40 the tools of data science, machine learning, AI
0:06:43 to bear on geoscience.
0:06:43 Absolutely.
0:06:44 If I’m going to reduce the idea, it’s that.
0:06:46 Totally, yeah.
0:06:49 It’s amazing that nobody got to it before you did.
0:06:51 There are these giant billion-dollar mining companies.
0:06:54 And it was right there for them, but they didn’t do it.
0:06:57 I mean, why didn’t somebody do it before you?
0:07:00 What you will definitely hear is, oh, we use data science.
0:07:02 Like, we use scientific, right?
0:07:06 And it’s like not totally wrong, but what is definitely
0:07:10 unambiguously different, if not unique to Kobold,
0:07:13 is that we’re a full-stack explorer.
0:07:18 We were started and built on the concept
0:07:23 that applying vanguard scientific computing
0:07:27 techniques to these problems would improve the efficacy
0:07:28 and efficiency of exploration.
0:07:29 That is the goal.
0:07:34 We have– our technical staff is about 60% data scientists
0:07:37 or software engineers and about 40% geoscientists.
0:07:41 So we’re roughly equal numbers across the three disciplines.
0:07:44 And that’s completely– that’s unique.
0:07:47 Let’s talk about data, right?
0:07:51 I feel like discussions about AI, for me,
0:07:54 tend to get more interesting when we get into data.
0:07:56 And it seems like that’s where a lot of the action is.
0:07:59 And from what I understand about the story
0:08:02 of your company, kind of building the data set
0:08:04 is a big part of the story and a big part of what
0:08:05 has differentiated you.
0:08:07 So you have all these data scientists.
0:08:09 What they need is data.
0:08:11 How do you go about building this data set
0:08:12 to find these metals?
0:08:14 Yeah, it’s an incredibly good question.
0:08:19 So most of the data we use was collected
0:08:21 by other people at other times.
0:08:24 Humans have been collecting information
0:08:26 about the physics and the chemistry of the Earth’s crust
0:08:29 for a very, very long time.
0:08:32 They’ve been doing it for, well, in some sense, for millennia.
0:08:34 But certainly over the last century,
0:08:38 they’ve been doing it in ever more sophisticated ways.
0:08:42 And for reasons I can explain, almost all of that data
0:08:45 is actually in the public domain.
0:08:49 The problem is it is a utter mess.
0:08:53 It is like an n-member hard, messy data problem.
0:08:58 Think of different humans in different decades
0:08:59 speaking different languages in different places
0:09:03 of the world, collecting different types of data–
0:09:05 and I’ll get into the types of data in a moment–
0:09:08 with different measurement techniques
0:09:11 based on the vintage of the era.
0:09:13 And then storing it in all manner of storage media.
0:09:18 Everything from literally handwritten geologic notes
0:09:20 or handwritten drilling notes all the way
0:09:26 to cloud data structures and everything in between.
0:09:28 And so it is this incredible mess of data.
0:09:30 Give me some specific examples.
0:09:31 What are specific–
0:09:34 like, did you find stuff in a drawer or something?
0:09:36 Like, give me some specific examples.
0:09:39 So I’ll give you examples of there’s
0:09:47 geologic libraries, archives, with carefully constructed
0:09:52 geologic maps that might be 100 years old.
0:09:55 And they were a smart, skilled geologist
0:09:59 to make a doing field mapping, which basically means
0:10:03 observing and recording the observations of outcrops
0:10:06 and describing the rock types and those outcrops
0:10:09 and locating them in space.
0:10:11 And the Earth’s crust changes very slowly.
0:10:17 So provided that was well done 100 years ago,
0:10:18 it’s still valid.
0:10:21 It’s just that it’s literally in drawers,
0:10:27 piled on top of each other, and basically not used.
0:10:29 It would only be used by a very industrious human
0:10:34 being who spent countless hours sort of searching
0:10:35 through the old archives.
0:10:39 So we go to various archives, and we
0:10:43 make an arrangement to digitize the information at our expense.
0:10:47 And we give the owners a full digital copy.
0:10:50 It’s almost always public domain data.
0:10:52 And so we have a right to use it,
0:10:55 or we negotiate a specific use right.
0:11:01 So digitizing a geologic map is the very, very beginning.
0:11:03 Then you need to extract the information
0:11:06 from the digital copy of the map.
0:11:09 And you have many different types of information there.
0:11:12 You could have, in the paper records,
0:11:14 you might have chemical assays, so measurements
0:11:19 of the concentrations of the elemental concentrations of samples
0:11:22 taken from different locations on the map.
0:11:24 And that could be a part of the record.
0:11:27 So that’s tabular data, because it’ll say, well, this sample,
0:11:31 sample whatever, had x% calcium, and y% magnesium,
0:11:34 and z% silica, and et cetera, et cetera, et cetera.
0:11:36 That’s all valuable information.
0:11:39 So that’s tabular information that then gets extracted
0:11:43 by our systems and populated into what
0:11:46 we call our universal schema, which just
0:11:50 means that every data type is stored in a consistent format.
0:11:54 You’re standardizing this wildly messy heterogeneous data.
0:11:56 You’re trying to make it more homogenous.
0:11:57 That’s exactly right.
0:11:59 And we should talk about more about what the data is,
0:12:00 because it’s really fascinating.
0:12:02 So I gave you two examples.
0:12:05 I gave you the sort of qualitative, almost like drawn,
0:12:07 geologic map, which is incredibly useful information.
0:12:10 But qualitative and continuous in nature.
0:12:12 Then there’s the sort of tabular data
0:12:14 that would be any kind of assay data,
0:12:16 measurements of composition.
0:12:18 But then you have a whole different classes of data,
0:12:20 like geophysical data, which tells you something
0:12:22 about the physics of the Earth’s crust.
0:12:25 So for example, you probably know
0:12:28 that the Earth’s gravitational field changes from place
0:12:29 to place as you move around.
0:12:32 It changes because you can go up or down an elevation.
0:12:35 Well, that’s easy to adjust for, because you know the elevation.
0:12:37 It also changes because the density of the rocks
0:12:39 below you change.
0:12:43 And so if you’re standing over an ore body that
0:12:47 has twice the density of whose rocks are twice as dense
0:12:51 as the surrounding rocks, that’ll pull on you slightly more.
0:12:52 And you can measure that.
0:12:54 That I did not know.
0:12:56 And let’s go down this rabbit hole, actually,
0:12:57 because it’s super interesting.
0:12:58 OK.
0:12:58 I’m in.
0:13:00 Because imagine you make this measurement.
0:13:02 What are you actually measuring?
0:13:03 You’re measuring the force of gravity
0:13:05 in a particular location.
0:13:07 And you can measure, OK, I’ve adjusted
0:13:08 the elevation and the force of gravity
0:13:10 is a little bit higher here.
0:13:13 OK, that’s all you actually know at this moment.
0:13:14 So what is that telling you?
0:13:17 Is it telling you you have a modestly more dense object,
0:13:19 like just below the surface?
0:13:22 Or is it telling you you have a massively more dense object
0:13:23 deeper?
0:13:26 Well, it turns out that this is a fundamentally degenerate
0:13:28 problem, or non-unique problem.
0:13:30 It would be the way you describe it mathematically.
0:13:32 There are many ways to solve that problem.
0:13:34 There are many different configurations.
0:13:35 You don’t know the answer.
0:13:36 You don’t know the answer.
0:13:41 What you do know is that there’s a very large class
0:13:44 of invalid solutions.
0:13:48 And then there’s a smaller, but still very large class,
0:13:49 of valid solutions.
0:13:51 There’s a lot of things that it is not,
0:13:53 and there’s some things that it could be.
0:13:54 Could– exactly.
0:13:55 Incredibly well said.
0:13:57 That’s exactly right.
0:13:59 So here’s a really cool application
0:14:01 of our technology and our approach.
0:14:05 So the industry standard approach to this is basically,
0:14:07 what does a normal conventional company
0:14:09 do with this gravitational anomaly?
0:14:10 Well, they do one of two things.
0:14:13 Most of the time, maybe 90% of the time,
0:14:15 they’ll just look at it and say, OK,
0:14:16 here is a gravitational anomaly.
0:14:18 This is higher here than here.
0:14:19 That’s interesting.
0:14:22 And they just see it on a 2D map.
0:14:24 So that doesn’t do anything for the non-uniqueness.
0:14:27 It just tells you what the measurement is.
0:14:28 So they would just use it.
0:14:29 They would just say that.
0:14:31 Well, it’s interesting because it’s higher.
0:14:34 OK, so what Kobold does is very, very different.
0:14:40 And impossible to have done even 10 years ago, maybe even
0:14:41 five years ago.
0:14:44 What we do is we solve, conditioned
0:14:47 on a set of geologic hypotheses that we find interesting,
0:14:53 we solve for the full set of possible subsurface.
0:14:57 So we might actually test like a billion subsurface
0:15:06 and say 999,999,900,000 of those don’t match the data.
0:15:07 We tried, they don’t match it.
0:15:09 So they’ve been rejected.
0:15:11 But we still have 100,000 now that do.
0:15:13 So now we have all of the good–
0:15:15 we’ve rejected the bad possibilities
0:15:17 and we’ve narrowed on the good possibilities.
0:15:19 But it’s still an incredibly large subspace.
0:15:19 Yes.
0:15:21 Impractical still.
0:15:24 It’s got to get a lot smaller for you to do anything like it.
0:15:27 But there’s a lot of information in the uncertainty
0:15:28 that we’ve now quantified.
0:15:30 We’ve now quantified the uncertainty.
0:15:31 And so we apply something that we
0:15:34 call efficacy of information, which
0:15:36 is a phrase that we coined.
0:15:37 And you can read scientific papers about it.
0:15:40 It’s a very general and really cool concept.
0:15:42 And it’s also kind of obvious, actually,
0:15:43 although it’s hard to formalize.
0:15:45 But you look very happy right now.
0:15:46 I get really excited about this.
0:15:48 Because this is an audio medium.
0:15:50 Like, your face is full of delight right now.
0:15:51 I don’t want to interrupt you.
0:15:51 Keep going.
0:15:52 That’s good.
0:15:53 Yeah, I’m excited.
0:15:55 I love talking about this stuff, because it’s super cool.
0:15:58 So we say, OK, the basic idea behind EOI
0:16:02 is, you’re going to collect some information next.
0:16:02 OK.
0:16:05 And that’s really what every exploration process always is.
0:16:05 Right.
0:16:06 Is a sequential set of decisions.
0:16:08 You’re going to go try and get more information
0:16:10 to figure out what is going on in the rocks
0:16:11 under the surface.
0:16:15 So the question you obviously want to ask
0:16:18 is, what next piece of information
0:16:21 will tell me the most?
0:16:26 Given, sort of, per unit of dollar that I expend,
0:16:27 what am I going to learn the most from?
0:16:30 What has the highest return on investment?
0:16:33 Yeah, and from a knowledge and information perspective,
0:16:35 what am I going to learn the most from?
0:16:36 And so here’s a way to think about that.
0:16:38 It is the piece of information that
0:16:40 decreases your uncertainty the most.
0:16:46 And because we have the 100,000 possible answers,
0:16:49 the piece of information that will decrease our uncertainty
0:16:52 the most is, in fact, the piece of information
0:16:57 that tests the most number that falsifies the highest number
0:16:58 of those 100,000.
0:17:01 So for instance, say we’re going to drill
0:17:03 in different directions.
0:17:04 And so we’re going to drill, and we’re
0:17:08 going to intersect the various predictions of concentration.
0:17:12 If we can do one drill hole and it would test 100 of the 100,000.
0:17:13 Yeah.
0:17:18 That’s only– we’re only testing 0.1% of the possible answers.
0:17:20 Don’t drill there.
0:17:20 Don’t do that.
0:17:21 We’re going to learn very little.
0:17:23 We’re going to end up with basically
0:17:24 the same amount of uncertainty.
0:17:27 But if we could drill a different hole, a different core,
0:17:30 that tests 50,000, say, half of them,
0:17:32 we massively reduce our search space.
0:17:37 We falsify 50% of the possible answers.
0:17:40 And sometimes we can actually falsify like 80% and 90%
0:17:41 of the possible answers.
0:17:44 So we massively reduce our search space.
0:17:48 And fundamentally, it’s the most possible information
0:17:49 you can get per unit dollar.
0:17:52 So every time we go to collect any information,
0:17:58 we try to formally quantify the uncertainty
0:18:01 and then calculate this EOI content of term, which
0:18:03 is the piece of information that we have the greatest
0:18:07 expectation will reduce our uncertainty the most.
0:18:08 Compelling.
0:18:09 I’m glad you think so.
0:18:13 It would be nice to be able to do that in life more generally.
0:18:14 It’s super, super hard.
0:18:16 And it comes– the really hard part
0:18:19 is quantifying the uncertainty correctly.
0:18:21 Once you have that quantified correctly,
0:18:24 then calculating the EOI, if you know for sure,
0:18:28 you’ve correctly quantified the uncertainty,
0:18:31 calculating the EOI is kind of an engineering optimization.
0:18:33 It’s kind of straightforward.
0:18:35 In this instance, quantifying the uncertainty
0:18:39 is basically how many ways could the rock under the surface
0:18:39 be?
0:18:40 Exactly.
0:18:40 Yeah.
0:18:41 That’s exactly right.
0:18:43 So you’ve been talking about sort
0:18:45 of gathering this very old school data
0:18:49 and making it useful to you and then what you do with it.
0:18:53 There is another piece of your data gathering operation
0:18:57 or another set of pieces that are more high tech
0:18:59 and that involve going out into the world and getting new data
0:19:01 that doesn’t already exist.
0:19:03 And some of those are kind of fun.
0:19:06 And so I want to talk about that a little bit.
0:19:06 Sure.
0:19:08 Tell me about detecting muons.
0:19:14 Yeah, so this is kind of one of our frontier R&D projects
0:19:16 within the company, something we would–
0:19:19 The opposite of a 100-year-old map in a drawer somewhere.
0:19:19 Yeah, totally.
0:19:20 Exactly.
0:19:21 Exactly.
0:19:23 And we have a lot of physicists at the company,
0:19:24 so they love this stuff.
0:19:25 So what is a muon?
0:19:27 Let’s start with that.
0:19:29 And then I’ll get to why they can be useful.
0:19:34 So cosmic rays are hitting air molecules
0:19:37 in the upper atmosphere all the time.
0:19:39 And when they collide, sometimes
0:19:42 they produce muons in the reaction.
0:19:47 So it’s a subatomic particle, a muon.
0:19:50 And it travels very, very, very fast.
0:19:52 It’s a sort of relativistic particle.
0:19:54 It travels near the speed of light.
0:19:57 So right now, muons are showering through us.
0:20:00 You and you in your studio and me in my home,
0:20:01 muons are coming through us.
0:20:04 And I think about– if you put your hand flat,
0:20:06 you can expect about one muon per second
0:20:08 to be going through your hand.
0:20:10 You don’t notice they mostly go right through you.
0:20:13 It doesn’t cause you any harm.
0:20:15 They do interact with electrons.
0:20:18 And it turns out that every time they
0:20:22 interact with electron, they slow down a little bit.
0:20:25 And when they slow down, eventually they slow down
0:20:27 enough that they decay into other things.
0:20:29 So they disappear.
0:20:32 So let’s say you have a muon detector
0:20:34 and it’s sitting at the surface.
0:20:37 And you’re listening– you listen to its detection.
0:20:39 So it’s like click, click, click, click.
0:20:41 That’s just telling you, OK, muon’s going through,
0:20:43 muon’s going through, muon’s going through, right?
0:20:46 Now, I drill a borehole.
0:20:50 And I start lowering the muon detector into the borehole.
0:20:51 And you know the rate it was at the surface.
0:20:55 Then as I get to say 100 meters, now it’ll be like click,
0:20:57 click, click.
0:20:59 And then as I go to like 500 meters,
0:21:04 it’ll be like click, click, OK.
0:21:06 So what’s happening is getting fewer and fewer muons
0:21:07 are hitting that location.
0:21:10 And the reason is because you’ve got so many more atoms
0:21:13 between you and the top of the atmosphere
0:21:16 that the muon just aren’t surviving.
0:21:17 They’re hitting the rocks.
0:21:18 They’re hitting the rocks, OK?
0:21:22 And so now you think of the journey of a specific muon
0:21:24 as it’s going through the rock.
0:21:29 The muon that interacts with the fewest electrons
0:21:31 is the most likely to hit you.
0:21:35 And the muon that interacts with the most atoms, we’ll say,
0:21:38 is less likely to hit you, because it’s likely to decay,
0:21:41 because it’s likely to basically lose its energy.
0:21:46 Think of a ball bouncing, hitting a bunch of other balls.
0:21:49 So now if you’re sitting at the muon detector that’s
0:21:52 been lowered into a location underground,
0:21:54 and you’re looking up in kind of a cone,
0:21:58 and you can look in three dimensions,
0:22:02 you’re seeing muons say they’re coming from the right a lot,
0:22:04 but they’re not coming from the left very much.
0:22:05 Yes, yes.
0:22:08 That’s telling you something about the relative number
0:22:11 of atoms to the left, which tells you about the density.
0:22:14 That tells you the rocks above you
0:22:18 and to the left of you are denser than the rocks
0:22:19 up and to the right of you.
0:22:22 And denser might mean an ore body.
0:22:25 It might mean a high concentration set of metal
0:22:26 in the rock.
0:22:31 And so it allows us to probe, in really sophisticated ways,
0:22:33 the density of the earth.
0:22:36 In the same way we were talking about the gravitational force
0:22:37 changes around the earth.
0:22:39 It’s the same thing, but it’s a much higher precision
0:22:40 measurement.
0:22:43 So we’ve designed our own novel muon detector.
0:22:46 We did it in collaboration with Occidental College.
0:22:50 And it’s working, and it’s in a pilot hole collecting muons.
0:22:52 And then we have a bunch of new ideas
0:22:57 about how to use this new kind of new data type.
0:23:00 There’s a few other companies doing this,
0:23:01 but it’s very new, very new concept.
0:23:08 After the break, from theory to practice,
0:23:13 Kurt talks about Kobold’s huge copper discovery in Zambia.
0:23:16 [MUSIC PLAYING]
0:23:23 Let’s talk about Zambia.
0:23:25 Sure.
0:23:28 I want to talk about Zambia because it suggests
0:23:34 that your hypothesis for the company is worked, right?
0:23:36 Yeah, to a large degree, yes.
0:23:40 Let me first say, why are we in Zambia in the first place?
0:23:43 So we look around the world, and we
0:23:49 evaluate jurisdictions on four dimensions, OK?
0:23:51 Our physical prospectivity, or how
0:23:53 we perceive the physical prospect,
0:23:54 the likelihood that there’s going to be something new
0:23:58 to discover in a particular location.
0:23:58 That’s thing one.
0:24:01 Thing two is the rule of law.
0:24:04 If we make a discovery, we have a property, right?
0:24:06 How robust is that property, right?
0:24:07 That’s thing two.
0:24:10 Thing three is access to markets, infrastructure, right?
0:24:13 If you find something in the middle of Antarctica,
0:24:15 you’re not going to be able to get it to the market,
0:24:16 no matter how great it is.
0:24:21 And then thing four is, how much resistance or/support
0:24:23 will there be to building a new industrial project
0:24:25 in that location, right?
0:24:26 If you find something in Palo Alto,
0:24:28 no one’s going to ever allow you to build it, right?
0:24:29 So it just doesn’t matter.
0:24:31 You can’t even build an apartment building there,
0:24:32 much less a lithium mine.
0:24:33 Exactly.
0:24:34 Exactly.
0:24:36 So those are the four dimensions we look at.
0:24:38 And we looked around the world early on.
0:24:40 Zambia rose to the very top on all four of those.
0:24:43 It’s a fantastic jurisdiction.
0:24:46 It’s the most consistent and stable democracy
0:24:47 in Southern Africa.
0:24:49 The physical prospectivity is tremendous,
0:24:52 because there’s been mining there for 100 years.
0:24:55 And so you might look at that and say, well, sure,
0:24:58 it was a good place to look 100 years ago,
0:25:00 but isn’t it all picked over?
0:25:02 And the answer is definitively not.
0:25:07 This is easy to verify, because basically all the deposits
0:25:10 that were mined over the last 100 years
0:25:12 were actually sticking out of the surface.
0:25:15 They were known about a century ago.
0:25:17 And there’s been almost no exploration
0:25:20 into the deep parts of the basins,
0:25:22 what we’d call blind exploration, right?
0:25:24 This is not directly evident, like you
0:25:26 can see it at the surface.
0:25:27 There’s been almost none.
0:25:30 It was this kind of perfect location in that sense.
0:25:33 It’s right on the central African copper belt, which
0:25:36 provides a significant majority of the world’s copper.
0:25:37 So it’s easy to get it to market, right?
0:25:40 And it’s a legacy mining country that’s
0:25:42 very supportive of development, right?
0:25:43 And so it’s like perfect.
0:25:46 It rose to the top across the board, and we’d love it.
0:25:49 And we love our Zambian colleagues.
0:25:51 And we think it’s one of the best jurisdictions
0:25:54 in the world for us to operate.
0:25:56 So that’s why we were there in the first instance.
0:26:04 And we started exploring in 2020 there in very modest fashion
0:26:08 loosely.
0:26:10 But we were exploring in areas right
0:26:13 in the heart of active mining basins.
0:26:16 Because again, there were active mines.
0:26:19 But if you went out into the deeper parts of the basin,
0:26:21 deeper than 500, 600 meters, there
0:26:24 were areas that just had never been probed or explored.
0:26:28 Yeah, so briefly, what did you find and how did you find it?
0:26:31 Looking at all the data, some of our geoscientists
0:26:36 had really, really clever ideas about how the mineralogy was
0:26:40 changing and how we actually might have very, very distinct
0:26:41 mineralogy.
0:26:43 So we might have areas where it looks
0:26:47 like it’s all this kind of one distribution, one sort
0:26:48 of set of statistics.
0:26:50 But actually, as you cross this boundary,
0:26:53 it’s a totally different geochemical system.
0:26:55 Different sets of geochemical reactions occurred.
0:26:58 So if you’re able to draw that boundary and then only go
0:27:00 and explore within that complicated three-dimensional
0:27:02 boundary, then you’d consistently
0:27:04 have high-grade and thick.
0:27:06 That was the assertion.
0:27:11 What you have in the reservoir, in the legacy data,
0:27:17 95% of the data is this low, modest-grade thin stuff.
0:27:20 And then there’s 5% of the data is this higher-grade stuff.
0:27:23 And if we drill there and within that boundary
0:27:26 and only drill there, it’s going to be good hole, good hole,
0:27:29 good hole, good hole, good hole.
0:27:30 So that’s the hypothesis?
0:27:31 Yes.
0:27:33 You test the hypothesis, and does it happen?
0:27:33 And we’re right.
0:27:34 Yeah, exactly.
0:27:35 So you found it.
0:27:35 Yes.
0:27:37 And we proved it, and it’s there.
0:27:41 And it’s gone from marginally economic to very economic.
0:27:44 For the same unit of rock that we move,
0:27:49 we sell 10 times as much copper as the average copper mine
0:27:49 today.
0:27:50 Yes.
0:27:50 OK.
0:27:51 So you found it.
0:27:51 Correct.
0:27:53 The hypothesis was true.
0:27:55 The system worked.
0:27:55 You found it.
0:27:56 Correct.
0:28:01 Now you are going to become a wildly different company,
0:28:02 it seems to me.
0:28:04 Like you are in Silicon Valley.
0:28:06 You are working with data scientists.
0:28:07 You have a technical background.
0:28:11 You have been running, essentially, a high-tech startup.
0:28:13 You are about to be running a mining company,
0:28:16 where your problems are not just being very clever
0:28:17 and hiring the right AI people.
0:28:21 They’re like getting the US government
0:28:25 to finance a railroad in Zambia and making sure
0:28:27 that the Zambian’s like you.
0:28:30 Like that seems entirely different than what
0:28:33 you have been doing so far.
0:28:37 Partially true.
0:28:39 It’s a little more continuous than you might have implied.
0:28:41 Yes, I’m just trying to make it a good story.
0:28:45 But look, it’s not a regular startup anymore.
0:28:46 Correct.
0:28:48 Like maybe it never was.
0:28:50 But it’s a super different job.
0:28:52 It’s a super different skill set.
0:28:54 I mean, I’m sure to some extent you’ve been dealing with this.
0:28:56 But it’s really different than what
0:28:57 we have been talking about.
0:29:00 It’s a whole different universe of problems and hard things
0:29:02 to deal with and a very different domain.
0:29:03 Sure.
0:29:06 No, these are excellent questions.
0:29:08 So you could think about the company now
0:29:11 as there is the discovery machine.
0:29:13 The discovery machine is everything
0:29:14 we’ve been talking about.
0:29:17 And the discovery, the unit economics of exploration
0:29:22 are great because you can make 100 times your money
0:29:25 or even more on proving the deposit.
0:29:28 Because it’s worth so much once it’s clearly economic.
0:29:28 Right.
0:29:30 So you could just sell the rights in some fashion, right?
0:29:32 You don’t have to be a mining company.
0:29:34 You could be a discovery company, right?
0:29:35 And it’s correct.
0:29:38 So the most important part of the company, the heart and soul
0:29:41 of the company, is the discovery machine.
0:29:44 Now we have at least one deposit that
0:29:47 is sort of unambiguously going to be a mine.
0:29:51 And the question is what happens from here with that mine?
0:29:53 And the odds are very high that we’re
0:29:57 going to bring in a partner with complementary capabilities
0:29:59 to sort of help bring it to production
0:30:01 and bring in a partner.
0:30:03 It’s a simple way to think of this
0:30:06 as we own 80% of it now.
0:30:10 The Zambian Peristatal Mining Company owns 20% of it.
0:30:11 It’s worth a certain amount.
0:30:15 You could imagine what someone would pay to own it entirely.
0:30:18 And that order of magnitude is billions of dollars?
0:30:18 Correct.
0:30:19 Yeah.
0:30:21 Yeah.
0:30:24 And so someone could come in and pay and contribute
0:30:27 to capital and capabilities to bring it into a mine.
0:30:30 And we could, in principle, spend no more money.
0:30:34 And we still have a large share of all the future cash flows.
0:30:37 It’s worth a lot to have the stake you have in this thing.
0:30:39 You have a set of choices about what to do with it
0:30:42 and how much sort of money to take
0:30:43 and how much of your interest to sell.
0:30:44 Correct.
0:30:45 That’s exactly right.
0:30:47 So is the answer you’re trying to figure out
0:30:49 how much you’re going to be a mining company
0:30:51 versus a discovery company?
0:30:52 I mean, is that–
0:30:52 Yeah.
0:30:55 And we know for sure that the most important thing is
0:30:59 to not weaken the discovery engine.
0:31:00 That’s the most–
0:31:02 and there’s a lot of culture around the discovery engine.
0:31:03 There’s a lot of attention.
0:31:04 That’s the most important thing.
0:31:09 We also know that it’s very, very important
0:31:14 that the value potential gets realized in the Zambian
0:31:15 deposit.
0:31:17 Is there about to be some deal?
0:31:19 If this show comes out in two weeks,
0:31:21 will there be some news between now and then?
0:31:23 No, and I can confirm that we’re not actually
0:31:26 looking for a partner for a couple of years, actually.
0:31:28 We’re not– we won’t formally partner with anyone
0:31:29 for a couple of years.
0:31:30 And the reason they’re actually really obvious,
0:31:33 it’s just that what we have discovered
0:31:36 sits on about four square kilometers.
0:31:38 And there’s another 150 square kilometers
0:31:41 on the license that we own that are totally untouched,
0:31:44 totally unexplored, completely unexplored.
0:31:47 So we are going to fully explore that whole area.
0:31:49 And totally know what we have before we formally
0:31:51 do any partnership.
0:31:53 But we’re also building the capabilities
0:31:57 to take the project as far as we want to.
0:31:59 So we recently hired–
0:32:01 we’ve stood up an amazing Zambian leadership
0:32:08 team of project developers and engineers,
0:32:10 metallurgists, et cetera, hydrologists
0:32:13 to continue to do the engineering optimization of what
0:32:16 a mine will look like in this location.
0:32:18 And so we’re doing all of that because that
0:32:19 has to be done anyway.
0:32:22 It adds a ton of value to figure out exactly how you’ll
0:32:25 optimize the operations.
0:32:28 And it just moves the project forward.
0:32:32 And our goal, our stated intention,
0:32:36 is to start construction on the mine within two years
0:32:39 and to have it in production in the early part
0:32:40 of the next decade.
0:32:43 So I know that we need a lot of copper.
0:32:47 And it’s great that you just found more.
0:32:50 It is also the case that mines have often
0:32:52 been bad for the places where the mines were
0:32:55 and for the people who worked in the mines.
0:32:59 Like, how do you deal with that, harming people and the world?
0:33:02 What’s super exciting about this particular deposit,
0:33:04 and in general, the deposits we’re looking for,
0:33:06 is it’s super high grade, 10 times higher grade
0:33:09 than the average copper mine around the world.
0:33:13 That means 10 times less waste for the same amount of copper.
0:33:16 It’s also an underground mine as opposed to an open pit mine.
0:33:18 So when you consider the overburden
0:33:20 that open pit mines, big holes in the ground,
0:33:22 would otherwise have, it actually ends up being about 30
0:33:25 times less waste.
0:33:27 And we’re going to take almost all of that waste.
0:33:33 And as we excavate the locations underground,
0:33:35 we take the waste and we put it back in.
0:33:37 We backfill is what it’s called.
0:33:39 We stuff it back into the area.
0:33:42 So at any given time, there’s only a modest volume,
0:33:45 modest cavity that’s open.
0:33:50 And then in terms of the, we are passionate and obsessed
0:33:51 with skills transfer.
0:33:53 And it’s just the reason that we are really building
0:33:55 a Zambian mining company to develop this project.
0:34:04 Our CEO, Makai Makai, she’s the CEO of Kobold Africa.
0:34:08 We have 90% of the employees in country are Zambian.
0:34:10 Chief metallurgist, chief mine engineer, chief project
0:34:18 director, a site geologist, all of them are Zambian.
0:34:20 And they’re extraordinary.
0:34:23 And we’re investing tremendous amounts
0:34:28 in helping them be the best professionals they can be,
0:34:31 because we’re going to be there for 50 years, right?
0:34:33 And we want to be there for 50 years.
0:34:38 So what’s the next big discovery?
0:34:45 So I will predict nickel in Canada,
0:34:57 lithium in Australia, and another copper discovery in Zambia.
0:35:00 I appreciate the specificity of that.
0:35:03 I love falsifiable predictions.
0:35:04 God bless you.
0:35:08 Something I make wagers all the time.
0:35:09 Weird wagers.
0:35:12 But it’s not because I’m a gambling man.
0:35:14 I’ve never played a spin of roulette in my life.
0:35:16 But what I love to do is people make vague predictions
0:35:16 about the future.
0:35:19 And I try to pin them down on something
0:35:20 that is clearly testable.
0:35:24 And then we usually bet a nice bottle of wine or a dinner
0:35:24 that we enjoy together.
0:35:26 So it’s fun.
0:35:28 But the loser pays, obviously.
0:35:34 And yeah, so we have, in fact, a big part of our company
0:35:36 culture is what we call the culture of falsification.
0:35:38 So when you go out to test the hypothesis,
0:35:40 your job is not to collect information
0:35:42 to confirm that hypothesis.
0:35:43 Because you can always do that.
0:35:45 You can always paint a new story.
0:35:46 That’s inductive reasoning.
0:35:47 It’s invalid.
0:35:49 Your job is to go out, is to tell me
0:35:51 how you’re going to test it, how you’re going to prove it
0:35:53 wrong, and go falsify it.
0:35:54 And one of two things happens.
0:35:57 You either successfully falsify it, in which we move on,
0:35:58 and we celebrate that.
0:35:59 We celebrate falsification.
0:36:03 Or you fail to falsify it, which means, OK, so it’s not
0:36:05 dead yet.
0:36:06 Now how do we test it now?
0:36:07 How do we test it again, right?
0:36:10 Yes.
0:36:11 And what’s the most efficient way?
0:36:14 What’s the highest return we can get on the question?
0:36:15 What’s the EOI?
0:36:15 You got it.
0:36:21 We’ll be back in a minute with the lightning round.
0:36:35 OK, we’re going to finish with the lightning round.
0:36:40 And so I got to ask, what is one weird bet you made?
0:36:45 Since we just had the Olympics, I bet that Usain Bolt’s
0:36:49 9.58 seconds world record in the 100 meter
0:36:52 will still be the world record in the year 2036,
0:36:53 by the end of the year 2036.
0:36:54 That’s a long bet.
0:36:55 You’re playing the long game.
0:36:58 Yeah, that’s going to be a longstanding record.
0:36:59 He said it in 2009.
0:37:02 It’s crazy for a record to last that long.
0:37:05 Yeah, would you like to make that wager with me?
0:37:06 I don’t have enough information.
0:37:08 I know enough to know that I’m ignorant.
0:37:10 But what made you make the bet?
0:37:12 It’s a dramatic outlier time.
0:37:14 It’s a total outlier.
0:37:16 Yeah, and he has the four fastest times.
0:37:18 So nobody– the next fastest human
0:37:20 is going to be the fifth fastest time in the world.
0:37:24 It’s like he is an outlier, and that time is an outlier for him.
0:37:25 Correct.
0:37:29 Correct.
0:37:30 What’s a cobalt?
0:37:32 Oh, good question.
0:37:37 So it’s a creature from German mythology
0:37:41 that lives underground– kind of like a goblin-like creature–
0:37:44 lives underground and controls the mineral wealth
0:37:46 of the earth.
0:37:47 Uh-huh.
0:37:50 And it’s also the namesake for the word cobalt,
0:37:53 for the metal cobalt.
0:37:55 Right, I mean, as I understand it,
0:37:57 people used to think cobalt was bad, right?
0:38:01 And then now we’re like, oh, actually, cobalt is good.
0:38:03 It looks a lot like nickel sulfides
0:38:07 when it’s a cobalt arsenide, and arsenic is toxic.
0:38:10 So it would poison miners.
0:38:12 And they called it the goblin metal.
0:38:14 What’s one thing I should do if I go to Zambia?
0:38:17 Oh, I love that question.
0:38:18 Well, you can’t miss Mosiotunia.
0:38:20 I call it– we call it Mosiotunia.
0:38:22 That’s the traditional name.
0:38:24 It means the smoke that thunders.
0:38:26 You will know it as Victoria Falls.
0:38:29 It is completely spectacular.
0:38:30 Niagara Falls is amazing.
0:38:32 It blows it away.
0:38:33 Totally blows it away.
0:38:39 You just can’t go to Zambia and miss Mosiotunia.
0:38:43 We’ve talked a lot about trying to predict things
0:38:46 and trying to quantify uncertainty
0:38:48 in the context of your company.
0:38:52 Do you think that way outside of work?
0:38:58 Well, yeah, I guess the way I make–
0:39:03 wager on things is a form of–
0:39:06 it’s certainly the way I think of–
0:39:09 I try to be scientific in every aspect of my life.
0:39:16 And I say that what science is not is empiricism.
0:39:18 It is not looking at the data and drawing
0:39:20 the inevitable conclusion.
0:39:21 There’s no such thing.
0:39:23 You can look at– with any set of data,
0:39:27 you can fit many, many, many hypotheses
0:39:28 that explain the data.
0:39:31 That’s always– this is the non-unique thing,
0:39:33 where it’s always true, basically.
0:39:34 It’s always true.
0:39:36 And so science really is about myth-making.
0:39:40 It’s about making up a myth that explains the data.
0:39:41 That’s your hypothesis.
0:39:44 The difference between science and religion
0:39:46 is that we test our myths.
0:39:47 That’s the difference.
0:39:49 That’s the distance of good science, right?
0:39:52 And in your heart, you should want to disprove it, right?
0:39:55 Like, if you’re really the best scientist,
0:39:58 you should want to prove yourself wrong.
0:39:59 Yes.
0:39:59 That’s this.
0:40:01 The best thing a scientist can say.
0:40:02 That means you learn something.
0:40:05 You really only learn when you realize you’re wrong.
0:40:12 Kurt Haus is the co-founder and CEO of Kobold Metals.
0:40:15 Today’s show was produced by Gabriel Hunter Chang.
0:40:18 It was edited by Lydia Jean Cotte and engineered
0:40:20 by Sarah Bouguere.
0:40:24 You can email us at problem@pushkin.fm.
0:40:26 I’m Jacob Goldstein, and we’ll be back next week
0:40:29 with another episode of What’s Your Problem?
0:40:32 [MUSIC PLAYING]
0:40:35 [MUSIC PLAYING]
0:40:38 (upbeat music)
Moving from fossil fuels to renewable energy will require huge amounts of copper, lithium, and other metals. Kurt House is the co-founder and CEO of KoBold Metals. The company recently made a huge copper discovery in Zambia, and is looking for other metals in other places. Kurt’s problem is this: How do you use AI – machine learning, data science – to find the metals we’ll need for the energy transition?
See omnystudio.com/listener for privacy information.