Mark P. Mills - Ideas Lab Series - August 9th, 2023

[00:00:07] Mark

We’re at the end of the beginning, not the beginning of the end of the build-out of the cloud. The cloud of course is structurally different than the internet. The cloud uses the internet, just like the internet initially used telephony and telephone lines, literally used telephone lines. We’re at an inflection point in infrastructure for information that threads through everything else. It threads through not just social media, but it also threads through all the other domains of machines, materials, and services.

[00:00:40] Intro

Imagine spending an hour with the world’s greatest traders. Imagine learning from their experiences, their successes and their failures. Imagine no more. Welcome to Top Traders Unplugged, the place where you can learn from the best hedge fund managers in the world so you can take your manager due diligence or investment career to the next level.

Before we begin today’s conversation, remember to keep two things in mind. All the discussion we will have about investment performance is about the past, and past performance does not guarantee or even infer anything about future performance. Also understand that there’s a significant risk of financial loss with all investment strategies, and you need to request and understand the specific risks from the investment manager about their product before you make investment decisions.

Here’s your host, veteran hedge fund manager, Niels Kaastrup-Larsen.

[00:01:35] Niels

For me, the best part of my podcasting journey has been the opportunity to speak to a huge range of extraordinary people from all around the world. In this series, I have invited one of them, namely, Kevin Coldiron to host a series of in-depth conversations to help uncover and explain new ideas to make you a better investor. In this series, Kevin will be speaking to authors of new books and research papers to better understand the global economy and the dynamics that shape it so that we can all successfully navigate the challenges within it. With that, please welcome Kevin Coldiron.

[00:02:12] Kevin

All right. Thanks, Niels, and welcome everyone. We’re in the midst of a revolution, but it’s not the revolution you think. Ever since Chat GPT burst on the scene last year, we’ve been swamped with opinions on how AI is going to change our lives, and it will change our lives but it’s not going to do so in isolation. To get a true revolution, we need advances in three different areas to happen at the same time. Number one, we need advances in our ability to gather and share information, we need advances in machines, and we need advances in materials, the physical stuff we use to make all this happen.

According to our guest today, this is exactly what we’re experiencing now. So here to explain why this is the case and some of the implications of it is Mark Mills. Mark is the author of a new book, The Cloud Revolution: How the Convergence of New Technologies Will Unleash the Next Economic Boom and A Roaring 2020s. Mark, welcome to the show and thanks for joining us today.

[00:03:19] Mark

Delighted to join you and pleased we get to talk about the book. Anybody who writes a book wants to talk about their book. It’s always a lot of work to write a book so I appreciate that.

[00:03:32] Kevin

Well, can you maybe just start off telling a little bit about yourself, your background, what you do for a living and why you felt compelled to write this book?

[00:03:42] Mark

Well, I wrote the book because I spend a lot of my life doing forecasting like all of us do. Forecasts have a lot of characteristics that make them not useful. I’ll tell you about my background in a second, but people can find a lot of about me, probably more than I want them to know, by going to the magic Google machine and putting in Mark P. Mills and all kinds of stuff comes up. I’m a senior fellow at the think tank, the Manhattan Institute, where I write about and do research on technology policy and energy policy. It’s a policy think tank all the more.

I’m a faculty fellow at Northwestern University’s engineering school where I get the pleasure of a lecture a year dabbling in manufacturing science and working with the professoriate there; brilliant, brilliant engineers, scientists. I’m also partner by way of full disclosure because I have a dog in the race, so to speak. I’m an investor and partner in a venture fund that focuses on software in energy domains because of the intersection of what we’ll talk about, the software bits and atoms.

But you know, forecasting is a tough art. There’s lots of books about forecasting. In fact, I may make my next book about forecasting. I have a very long appendix in my book about forecasts of the past and what’s wrong with forecasts, the characteristics of forecasts. I like to characterize forecasters as falling into one of three buckets of characteristics. There’s forecasters that are paid to entertain, which is not a bad thing. It’s a lot of fun. This changes everything. The world’s going to end, this is going to eliminate all the cars in the world, whatever. We’re all going to be living in space as space tourism. It’s entertaining. It’s fun. It’s not fiction, it’s just imaginative and not necessarily grounded in anything.

Then there’s a class of forecasters that are paid to sell their book, so to speak, in investing terms. They have an agenda, they have a story, they have a narrative, they have a business. I’m not saying they’re lying, but they believe in their story. People in that category often engage in cognitive dissonance by virtue of enthusiasm for what they’re doing. Good on them, except they’re selling a story, whether it’s a policy story or telling an idea as in policies or whether it’s selling a product.

The third class of forecasters are those who are paid to be right. You never hear from them because if they’re really good, they’re locked up with golden handcuffs inside of investment funds. They work for Warren Buffett or whoever. So the book is I hope entertaining, but I was trying to lay out an architecture for forecasting the near future based on a truism, and technology in particular.

Forecasts about politics are very different. Forecasts about technology have a characteristic that’s very important. I stole a line from the management consultant Peter Drucker that I only forecast things that have already happened.

[00:06:40] Kevin

It’s a wise decision.

[00:06:42] Mark

Yeah. What he really meant by that was demographics are consequential, the future will have older people in Europe and the United States. That’s already happened, not going to change. So in technology domains, you can make forecasts that have an unusual characteristic. Using the mantra you only predict what has already happened is actually a good metric. If somebody announces this is a favorite clickbait, invention, a new thing, a new chemistry, okay, that’s nice, but it’s not a product. It’s not going to change the near future.

The future that most people care about is the next decade, and what you can build in the next decade you already know how to build. It’s either something that’s already fully mature that you can build more of like electric cars, they’re fully mature, you could build more of them, or it’s something that someone is just now building, there is only a few of them commercially available, they’re not widely deployed in markets but they’re obviously possible to build and get on the cost curve of getting cheaper, better, which would be one of my other favorite subjects, this anthropomorphic class of robots.

They’ve been in fiction forever, but they aren’t fiction anymore. They really exist. You can buy them. They are affordable in certain applications. So will there be more electric cars in the future? Well, that’s easy, setting aside subsidies and stuff. We obviously know how to build those. Will there be robots in the future, which people still are skeptical about, very skeptical because it’s been promised for so long? Flying cars is another example, air taxis. We know the answer is yes because they exist now at a commercial phase that’s nascent.

That’s because the idea or the invention that began the path of being commercially viable started 10 or 20 years earlier. Because everything, despite the constant claim of accelerating change… One of the other things I map out of my book is the timeline from inception to a product becoming commercially useful to an inflection where it’s significant is roughly the same and has been for almost two centuries.

It has not accelerated because doing things at scale that are in the atoms world where everything operates, including computers, pretty much the same, 10 to 20 years from inception to a product, 10 to 20 years from the product being an inexpensive, viable, safe, useful, high-reliability product, and then after that, you start to see significant market penetration for most products, most devices, most things. So that’s what the books about. It’s looking across the domains of machines, materials, and information.

What’s going to be different about the next decade has already been invented, and the thread through it, the reason for the title of the cloud, is the thing that’s really different is the infrastructure of the cloud. We’re at the end of the beginning, not the beginning of the end of the build-out of the cloud. The cloud, of course, is structurally different than the internet. The cloud uses the internet, just like the internet initially used telephony and telephone lines, literally used telephone lines.

But we’re at an inflection point in infrastructure for information that threads through everything else. It’s threads through not just social media, but it also threads through all the other domains in machines, materials, and services.

[00:10:01] Kevin

Given that you mentioned that and your book is called The Cloud Revolution, maybe we could just talk about exactly what you mean by the cloud. The reason I want to do that is because what I really enjoyed about your book is that you repeatedly tie things that we call virtual or cyber back to their underlying physical realities. When we’re talking about the cloud, we’re talking about some form of physical space, made up machines which are made up of materials. Let’s say we’re all using the same terms, when you talk about the cloud, explain what that is in physical terms.

[00:10:43] Mark

Yeah. That’s what is fascinating and surprising to most people. It’s obvious what you stated. Nothing exists, no product or service exists without materials that were first mined from the earth, turned into products, whether they’re machines or other costs of products to allow you to do stuff. But most people ignore that because it’s sort of hidden, and it’s meaningful, first as a measure of scale it’s meaningful.

If you’re an investor, it’s meaningful because sometimes there’s new opportunities to invest and expand new domains that are making these things. But cyberspace software AI exists in machines. But for the machines, there is no software. Software is mathematics. It’s a language. It’s mathematical language largely, but not entirely. It’s a semantic language now too with AI, but it’s mathematics. Those ideas existed theorily, even metaphysically. You don’t have to instantiate a number. But if you want it to be useful, you have to instantiate it, you have to have a machine, everywhere and always.

Because the cloud is infrastructure entails three things. It entails end-use machines, devices that either collect or deliver information. It doesn’t have to be a voice. It doesn’t have to be a map. It can be a temperature, it can be a state of being, a velocity, but you have to have sensors at the endpoint collecting, delivering information, then you have to have connections from those to some processing, some kind of mediation, which is the data centers, but they are connected without wired and wireless networks.

The wireless networks are invisible, except the things that allow the wireless networks, the radios, are machines as well, big machines, complex machines, kilowatt consuming machines. Every cell tower is a multi-kilowatt consuming device. There are millions of cell towers in the world that never existed before. You have to build these things. So if you think about that, how would you measure the cloud setting aside what the cloud does, because it does more than store cat videos and annoy people on social media as everybody knows.

It does mapping, which is an inferential task. It provides ways to get advice for both professionals in industries in medicine as well as people. You could shop in the cloud. It does lots of stuff, obviously. But a lot of it is not computation and it’s not communication. It uses communication. The dominant feature of the cloud now and the future is that it does things that I would call inferential instead of computational. Mapping is inferential. You want to probably go here based on where you are, where you want to go in traffic.

But how big is the cloud? How big is the infrastructure that combines the hardware devices that make that possible in a seamless network? Well, you could measure it. You can measure as a network in miles because it is a network wired and wireless and it’s measured in billions of route miles, billions. There’s no other network humanity has ever built that’s measured. It dwarfs the highway networks and dwarfs all telephony networks by orders of magnitude.

You could measure in dollars, because in the end, money matters and every infrastructure we spend money on. The infrastructure of the cloud’s capital spending on data centers and networks globally is now greater than the world’s utility capital spending on electric utility networks and production. So it’s already a bigger utility than the electric utilities of the world. You could measure it in buildings, which is sort of my favorite because the data centers for those who have never seen them, these are buildings the size of a Walmart not filled with people or merchandise, but filled with hot silicon servers, storage devices, communications devices, and those things are just buildings full of machines.

You could measure those in square feet and in dollars and in power. The only analogous buildings in an odd way are skyscrapers. Skyscrapers under a single roof can a house a million square feet, so can data centers. So if you sort of counted the world skyscrapers were, which were the icon of the technology of the 20th century fundamentally, they came along in the 20th century and skyscrapers were made possible by, as you said, out of confluence of technologies, electrification to make elevators, high straight steel, building techniques, roads, because a skyscraper is not useful if you don’t have roads and highways and concentrated infrastructure.

So skyscrapers in the world that are roughly Empire State class, there are about 1500 of them. We still build them. Data centers that are Empire State class, like a million square feet and above, 5000, and we’re building them at a far faster pace.

[00:15:47] Kevin

I wonder, Mark, if I could just ask you questions going back to what you were saying right at the beginning about forecasting things that have already happened. Are you saying that the scale of the cloud, as it exists now, is what’s going to propel this revolution? Or are you saying given what we can do what, we can produce now, it’s very, I don’t want to say easy to predict, but it’s predictable that the scale going forward will be of such a size that it will generate this revolution? In other words, is the cloud build-out just getting started but it’s getting started in a way that we can now forecast with some clarity?

[00:16:33] Mark

Yeah, I think it’s more the latter category. The cloud as we have it today, the first cloud data center was 1999 in Santa Clara, but the real build-out of the cloud started, give or take, 2005. Early 2000s, that’s when this data center boom started and the kind of features that are associated with your smartphone were made possible essentially by cloud architecture. So we’re say a decade-plus into the build-out, so it’s now significant. As I said, it’s already equal, exceeding the spending on electric utilities, but it’s just getting going. Many people say, oh, it’s a big utility. You can only use so many roads.

America has a lot of highways, famously the Interstate Highway System, so does Europe, and railroads. No one expects those build-outs to increase tenfold much less twofold in the coming years. They saturate. We’re not even close to saturation in the cloud. We know this because of the velocity of the buildup that’s underway with conventional cloud. You began with mentioning AI and what’s happening. AI portends an even faster and more expansive build-out of the cloud because of the physical nature of the machines needed to instantiate what we call artificial intelligence, machine learning.

[00:17:53] Kevin

So what you’re saying there is that the infrastructure that AI requires is going to necessitate a bigger build-out of the cloud. For it to operate across all the applications that we want it to, the cloud has to get bigger.

[00:18:12] Mark

Yeah, it’s already happening. We already know if you look at what Amazon has announced, Hewlett-Packard, Google, Meta, all of them have announced first either a pause or a redirection and expansion of the cloud spending. It’s not just the service more uses for shopping. That’s not a very predictable linear track now, the invasion of commerce by E-commerce. There’s no step function really coming that we can see. Most people know that in machine learning, both the learning and the inference require incredible compute horsepower. It’s the most energy-intensive use and most silicon-intensive use of computing. Period. Full stop.

It’s orders of magnitude more power and logic intense, if you like, than traditional computation. To learn something, you take billions of data points and you run a computer flat out, a supercomputer. To take supercomputer capabilities, you’re not going to put a supercomputer anytime in the near future in your pocket. Your smartphone can have AI built into it. It does. Your facial recognition is AI-centric. It’s not a linear function because your face is a complex map. So it’s an AI kind of chip that’s doing that.

But to do the kinds of things everybody imagines or is afraid of or excited about requires supercomputers. They have to be easily accessible, hence the network to the cloud, and they have to be easily upgradeable because the pace of change is so fast that what cloud providers do is they’re continually upgrading the hardware to bring you next-generation capabilities. But the footprint of power and machinery to provide the demands are now merging in all the applications that people are imagining.

You’re seeing explosive growth in both potential applications, not just things like drawing and doing movie making, but in everything from medicine to manufacturing. The IT departments of every company are experimenting with Chat GPT or the equivalent to take work and time out of routine tasks. Once they figure out that they can do it, and it takes a little time, I think another year or so before a lot of people get up to speed or two, then they’ll want the service. They’re not going to get that service by and large anywhere except for the cloud.

That’s a step function as big as the advent of the cloud or as big as the advent of the internet. In analogous terms, this is the era of ‘you got mail’. It feels like everybody’s amazed at the Chat GPT and the cloud. Well, they were amazed that you got mail at AOL. It was amazing. Everybody wrote about it, all kinds of babbling, clobbering. Remember that?

[00:21:02] Kevin

Right, yeah.

[00:21:03] Mark

Everyone was blown away. You could shop online too. You could actually see. Wow. So that level of awe for where things were then, in hindsight, it was pretty pathetic computing and email service, that’s analogously where we are with AI and the cloud. We’re just entering an explosive growth era, hence you see this in stock prices, whether they’re properly reflecting growth or whether we’re going to repeat the 1990 bubble, which I think probably happens. But the enthusiasm is properly placed because it is a big deal and it is the structural change that’s now accelerating.

[00:21:45] Kevin

There’s so much in your book it’s kind of hard to know where to drill down. The first part is like a revolution in how we collect and propagate information. My understanding, tell me if I’m wrong here, just talking to people who have some expertise is that the AI type of inference machine learning works best on a database that’s curated in some sense, that’s focused on collecting a particular type of data, obviously, the cleaner the better.

So is there some sense of a downside to connectivity in that you’ve got AI-type logic operating across vast data sets that include a lot of garbage, the “Internet” being the most example? Are we going to see the advances coming as we segment data into use-specific areas and that’s where we really get the advances?

[00:22:53] Mark

You’re absolutely right. You phrase it really correctly and that’s a really important insight that you picked that I thought about a lot in my book. The quantity of data that we could collect is functionally infinite because information isn’t constrained by the atoms world. So you can imagine collecting any level of granularity or precision infinitely. I can always want more data.

The problem is, quantity doesn’t necessarily have a lot of noise in it, it’s not useful.
It’s the reason you can compress videos because a lot of the information in the photograph is whitespace. That’s essentially what the compression algorithm does. It throws out the stuff you don’t need then reassembles it. As we move into an era where we’re hungry for data, the model of how we can do that, how can we curate the data as it’s being collected literally, so you’re not storing data that’s worthless and then processing data that’s worthless, because that’s actually expensive.

As data volumes go up, you hit physics limits. It actually becomes a problem to store it and transport it. But you have this revolution in sensors, ability to collect data where the cameras are tidy and cheap are a good example. Cameras collect voxels, three dimensional bits, are really data-intensive, but visual information is incredibly powerful, not just for fun and entertainment, but for science, for analysis and for regulations. There is an infinite amount of visual data you can collect especially with how cheap cameras are and how much cheaper they’ll get. They’re sensors for photons.

But the pioneering on this interestingly enough with the Event Horizon Telescope. So this is me going down a deep hole because I wanted to be an astrophysicist. But you probably remember all the news about the first photograph of a black hole. Remember about two years ago, we saw this picture and the picture was assembled by a collection of telescopes that function virtually as a single telescope with a diameter of the planet Earth. They collected exabytes of data, numbers that are staggering.

The quantity of data was so great that it couldn’t be transmitted over fiber, even fastest fibers. They used what they call jokingly the sneakernet. The data were collected in hard drives and carried by people in airplanes. It was the central collecting computer. So astronomers have been trying to…

[00:25:37] Kevin

It’s old-school going back to having floppy disks.

[00:25:41] Mark

They carried cases full of hard drives because transmitting 100 petabyte file, 100 petabytes in astronomy is a small file. It takes 20 years to transmit 100 petabyte file on the high-speed fiber, 20 years. You just leave it in the hard drive and carry it in an airplane, you’re there in two days. But what astronomers and researchers have been trying to figure out is as data are collected, you can use AI and algorithms to as you’re collecting it make a determination whether what you’re collecting is noise or garbage. So you don’t store garbage, you don’t store noise.

You’re not doing compression in the old sense of the word, you’re doing filtering. But it’s a smart filter and it has to be an AI-centric filter because the nature of what you’re filtering is going to dynamically change. So you have to have a filter that’s self-aware of the mission it’s on. Maybe collecting data, you’d say I wish I had it because it would be useful for something else. That’s irrelevant. It’s not useful. It’d be like your doctor collecting data about temperatures outside. What the hell do you care? You want it focused on your personal body temperature, not the temperature of the room or the building.

So, that kind of focus, to your point, is possible with the application of end-use AI that’s now curating and determining what’s noise and what’s not noise. Then you’re using AI at the other end. It’s already a self-curated (call it naturally curated through AI at the front end dataset), which allows even though I still need a supercomputer, I’m now dealing with (I’ll pick a number) a 10 petabyte file instead of a one-exabyte file.

An exabyte is 1000-fold more than a petabyte. So you’ve now changed the compute cost by 1000-fold to do something magical through the application of AI. It’s a virtuous circle. It’s one of these rare times in history where the technology has… You could call it a virtuous circle. There is no analog in the thermo mechanical world, but it’s almost as if the act of driving your car made the process of drilling for oil and refining gasoline more efficient, but just by the mere act of doing it the same machines. It doesn’t, they’re independent magisteria.

[00:28:05] Kevin

I wonder if we could talk a little bit about materials, the second area of advancement. I really liked this section of the book I think because a lot of people don’t think about materials that much. You say, hey, we could draw a straight line from the MRI machine to mines in Brazil. There’s a direct correlation between that. Just yesterday, I saw Norway announced a big discovery of rare earths that I think previously were only available in China. So there’s kind of this political dimension to the materials as well.

You talk about this notion of computational materials and that the technologies that are emerging are dependent on those. Can you explain to us what computational materials are?

[00:29:05] Mark

Yeah. I’m not the first one to use the word because it finds its roots obviously in the scientific domain of using computers to think about how to design materials with characteristics, a product that doesn’t exist in nature. You can do the experiment using known laws of chemistry and physics in silico instead of in a lab. If the computer is powerful enough, you now have accelerated the experimental phase by not just a thousandfold. You can do millions of experiments, in silico, in weeks, in a supercomputer that you couldn’t possibly perform for centuries in real life.

[00:29:48] Kevin

You’re talking about simulations. Is that what you mean?

[00:29:51] Mark

Well, these are simulations based on the quantum mechanical reality of how atoms and molecules combine, so really difficult simulations. But yes, you’re simulating an experiment in a computer. That’s a computational material in that sense. The other class of computational materials is designing materials that integrate features of reactivity or awareness of its environment. So smart materials, materials that can heal themselves or change their property in some fashion because of the conditions they are in, whether it’s hotter, pushed harder, you want them harder, you want them softer, you want them more ductile.

Those characteristics can be built into materials now because you can make them be fully integrated with the sensor and the actuator function if you like. But that’s a computational material, it’s an action.

[00:30:43] Kevin

Just to understand, you have this line and I think you just said it where there’s potential to create materials that have properties that are impossible in nature. I kind of got stuck on that thinking, well, how is that possible? But now as you’re explaining it, I think maybe I’m understanding because you’re saying that it’s because they’ve got sensors, they can react to their environment so this will dynamically change their properties. Is that what you mean?

[00:31:15] Mark

Yeah. Obviously, it’s a rhetorical flourish because if it were impossible in nature, nature wouldn’t permit it. So it’s that they don’t occur naturally in nature would be the more precise language because nature is permitting it because I’m using natural atoms by definition. So the classic example of that is not just the self-healing materials. Polymers exist in nature because they are organic. But the kind of polymers we manufacture don’t exist in nature.

Some people aren’t happy about that because they don’t degrade naturally. You have to be careful how you dispose of inorganic materials. But you can imbue in them properties, reactivity, intelligence, if you like, limited intelligence, self-awareness. That doesn’t occur naturally. It doesn’t exist in nature. Now, it exists in biology. That’s what animals and plants do. So if you’re sort of expanding there, we’re trying to copy nature. Mimicking nature is the holy grail of all materials. You would like to be able to grow a part rather than machine it from a block.

Life doesn’t exist from a big mass of organic protoplasm, that you chip away at the organic protoplasm and end up with a plant. It grows from a seed and grows into a plant. We do the inverse and we build machines. We take wood, we chip away at it to get the part. Well, 3D printers effectively emulate the growing process and allow you to do things that can emulate growth. There are commercials, as is in my book and people can find this, there are commercial 3D printers that print skin for burn skin grafts. That’s now possible. So we are growing skin the way nature does.

It turns out skin is the easiest organ to emulate and grow. But Dean Kamen is working on, as many others are, growing parts of more complex tissues like the heart and even eventually growing a replacement heart for you from a 3D printer if you have an ailing or damaged heart. This is a distinction with a difference. This is really different than machining and assembling. So that kind of computational class of materials as well as the metamaterials which were discovered only in the last two decades, which is thinking about ways to assemble a surface so that it exhibits a property of what’s called negative refractivity index.

The famous experiment we all saw in high school physics class, you stick a spoon in a glass of water, it looks bent. But it always bends the same way. That’s how refraction works. It never bends the other way. But with a metamaterial, you could make the image bend the other way, which actually means that you can build a surface that will function like an invisibility cloak. So the science fiction idea of an invisibility cloak is now possible to build at certain wavelengths, not all wavelengths. So it’s not crazy to say you can make an invisibility cloak. I can make a surface invisible to certain wavelengths of light or radiofrequency.

Does that have application other than in comic books, in science fiction or military? Yeah, it does. It turns out that that property of manipulating light in a way that no natural material doesn’t use in nature, means that you can start thinking about building something called meta lenses for both radio frequencies and optical frequencies. So we’re on the cusp of seeing (back to my original analogy), not imagining but these products are already being built, lenses that are tiny that actually emulate the properties of big lenses.

The reason most cameras have big lenses is because you can’t fake the optics without a big lens, well, meta lenses can do it. They can fake the optics. Same is true when switching radio frequencies, the trick with high-speed communications. Most of us just discovered that the internet does not move things at the speed of light, it’s slow. If you and I have a link and we send an email each other, you’ll notice it takes a little while depending on the packet. It’s not just a routing issue. Turns out it’s not just a physics issue how fast the photons can go in fiber and the air. It’s dominantly a switching issue.

Switching speeds are really, really hard to make fast, really, really hard. That’s sort of the choke point. It’s limited by the physics of the switches, the optical switches. Meta lenses can change that by a factor of more than 10. So you’re going to get this weird idea of a meta surface that doesn’t exist in nature that might make an invisibility cloak or show up as a practical matter in toys like cameras, if you like, but more importantly, in the speed with which we can communicate, which comes back to full circle. That will really matter because I need to put a lot of data into the cloud if I want to make AI really useful.

[00:36:06] Kevin

Could you, as a material scientist say, hey, I’d like to have a particular property and then could I create an AI algorithm that in some sense searches all the possible combinations to produce that? Could you say, design me something that does this, and given the right data set, it can do it?

[00:36:35] Mark

Well, the short answer is yes. In fact, such companies and programs already exist at a comparatively rudimentary level, because it’s a really hard problem from a computational perspective, not just a computational one plus one equals. Because when you start combining atoms and molecules in novel ways, you’re in quantum physics land, which is not just a computational problem, you need computers that can do AI, that are looking for probable outcomes and high probabilities and could do it very fast. So there are a number of companies.

One of the earliest ones was spun out of Northwestern University in effect, which was designed to do exactly what you said. It alloys. So it was very specifically designed to look at metallurgy. If you said I need an alloy that doesn’t use cadmium, because cadmium is toxic to handle, but it has this ductility, this temperature resistance, and you state your characteristics, and go hunt all the possible combinations of non-cadmium alloys. Again, that’s in silico experimentation based on novel combinations, and of course, what AI does which is different than computation, when you program the original computers to do chess, what they were doing was trying to program all future possible moves because it’s a statistically limited number of moves. It’s a big number, but it’s pretty limited, so you just crunch them fast. AI doesn’t do that.

AI does much more akin to the learning that we do. It looks at different probabilities and different combinations that might not be obvious to you or to the AI at the outset. That’s a big deal. It’s self-evident. You’re a car designer or a pharmaceutical designer and you want to be able to ultimately ask the question, I need a product that behaves like an aspirin, but for this cohort of patients, this side effect wants to be avoided. I believe that that thing that causes side effects in the dog is this with this biological function, go find something. Some of that was already done, by the way, during the evil lockdowns with the COVID pandemic.

Data lakes were created by a group of companies and being offered for free to clinical researchers to come up with both new vaccines and also new therapeutics. A number of novel therapeutics were discovered that way. They’re in clinical trial now so you still have to go through that. Ultimately, you can even fake the clinical trial in silico. You can say this is what a person will react like, I can do all that on digital twins. That’s really, really hard. We don’t have the computers to do that yet. But you can do some of it through the merging of bits and atoms.

As you know (I put it in my book,), one of the things that you do is you put the relevant tissue that you’re worried about being impacted, the real one tissue, the real heart tissue, on slides and you make a slide matrix of tens of thousands of slides. You use both AI and a computer and actual biological cells, which is done today routinely to look for first-order effects, toxicity effects for example. So those things are real, already happened back to my core thesis.

So it’s not hard to imagine them getting more powerful if you believe computers are getting more powerful still and Moore’s law is not over, and it’s not. All the hype about Moore’s law is ending, yes, eventually. There’s always limits. We’re many generations away from the end of Moore’s law.

[00:40:10] Kevin

I wonder because we’ve got maybe 15 minutes left to talk about the second part of your book, which is probably the part that most people are I guess concerned about, worried about, and you have an optimistic take. The subtitle of your book is, hey, it’s gone be the roaring 2020s. So I wanted to get into that a little bit. I’m nerdy enough to have a favorite or a group of favorite economists that I like to follow. One of them is this guy, Andrew Smithers, who I’ve known for a long time.

His take is, hey, the issue we’ve had is it’s not that, hey, we need to worry about robots taking our jobs, it’s actually we’re not very good at getting the most out of the technology we already have. So if you look at the productivity numbers in the last decade or so, we’ve been below average almost for the last 15 years. So his thing is, hey, we’re not getting as much out of the technology that we already have. Why do you think it’s going to be different? Why are we going to be able to get more out of technology going forward than we have had success doing in the recent past?

[00:41:24] Mark

I’ll use my same metric as I have all along, because we’re already doing that, but only at limited scale and deployment. So the ability to get more out of what already exists, the machines we know how to build that make them more productive, how people operate with the machines. Whether you think there’s more or fewer people doesn’t matter. There’s always some point, there’s an intersection of machines and people. How effective that interaction is correct.

This is no knock on economists, but I do knock the economists occasionally. I like to point out that economists are only marginally good at building models to predict what’s already happened and rarely demonstrate an ability to predict anything that’s going to happen when it comes to technology. They’re not really good at predicting dislocations. But they are correct that new technologies go through these phases of maturity where they change things a lot initially once they’re commercially viable. So this is the first-order problem is that economists look at things as I don’t see any benefit for that technology.

Well, it’s because it’s used in a de minimis way and there’s a learning curve simplistically put for all technologies, and then you get this big ramp. So robots and car factories are a classic example that did welding. The first robot, the Unimate, in 1962 profoundly improved the productivity of General Motors Lordstown factory, and everybody copied it because it was the most productive factory in the car world because it implemented that class of robots. Well, we’ve run a lot of capability on that. It’s our May asymptote.

So can it get better? Yeah. Well, okay, that’s a fair question. Because what most economists would point to is this, as Gordon does at Northwestern that once you have made the toilet, it’s done, once you’ve made the car, it’s done, and once you’ve made enough better, all the improvements are just incremental so there’s not much more we can do. So what you want to look for are dislocations. Are there any profound dislocations?

My whole book is really about the fact that across all markets, transportation and healthcare and education, manufacturing, there are dislocations that are equivalent to, in many cases, the development of the engine itself or the development of electrification and electric motors, that class of dislocations. There’s not just one, there’s a whole set of them. But when you look at history and you look at when dislocations have happened, again, to be simplistic, when the railroads came along, it was a full two decades before society reorganized itself around how to operate with railroads to make the production system, delivery system combinate the capabilities of railroads.

Same was true for cars, same was true for airplanes, same was true for computers. So fundamentally, to assembly, that’s what the cloud is. The cloud is, combined with AI (the last chapter my book is really about AI and research), but the combination of those two things is as or more consequential as the combination of the mode of age and electrification, the steam engine internal combustion engine electrification, circa 1910, 1920. We were not wringing the benefits out of those new technologies anywhere close to what they were going to do at that time.

It took time for management practices to building structures, everything, how you train people, how you train engineers if they can operate. So if you talk to students today in any engineering discipline about what they’re going to do with the tools they have available, you see a sea change from what you would have talked to them 10 years ago, but those students are just now entering the workforce, manager practice are just now adjusting.

[00:45:24] Kevin

Can you give an example that? I’m curious of a specific example of how someone as one of your students, engineering students now would think about using these tools compared to say just a decade ago, a 30-year-old.

[00:45:38] Mark

So the first thing is, you wouldn’t think in terms of designing a manufacturing line that had a cobalt collaborative robot that didn’t have to be bolted down. So if you’re designing a warehouse branch today compared to 10 years ago, you would do two things different in designing the warehouse and it has already happened. So I use this as an example because that’s been going on for a few years. You’re not getting rid of people.

But you have in the warehouse wheeled and now walking, but mostly wheeled, autonomous or semi-autonomous or fully autonomous machines that are helping the people, but don’t bump into the people, they don’t hurt the people. They’re aware of their environment. Most robots have to be bolted behind cages so they don’t hurt people because they’re not aware, can’t control.

So that plus the fact that you now have sensors and AI to do analytics in real-time that you didn’t have 5 or 10 years ago changes not only how you design the warehouse itself, but this changes how you would think about operating the warehouse, because you can push operational decisions to the edges, if the tool you have allows you to imbue the employee at the front line, the person on the conveyor belt with knowledge and information that they couldn’t possibly acquire because it would take an engineering team a day to analyze the data. But now that person is getting that information in real time.

This might have to do with potential failure of a conveyor, it might have to do with a change in supply chain, it might have to do with the fact that you’re vacuuming up data in your network and you know that the things that are going to be needed in two hours are really different than the things you’re moving now, how you reposition your trucks, how you reposition yourself and your robots, you can do that in real time and they are doing that in real time.

So the reason we know that’s happening, so we’ll talk about productivity that got stalled out, compare the economic productivity of delivering a package after you ordered it on the magic internet 10 years ago to today. Why is it that everybody, not just Amazon, can deliver stuff to you in a day or hours? What’s different today from 5 or 10 years ago with that? To say it’s more productive is a profound understatement. Productivity is a measure of same or more output with fewer inputs.

The productivity of delivery of packages because of the combination of AI and the robotization, if you like it, of the warehouse infrastructure is what’s delivered to you this incredible increase in productivity of delivery of packages. That’s what will happen in the delivery of goods and materials to manufacturing plants. That’s coming next, it’s already started. But it’s harder because the problems are actually much more complex if you’re making a car or making an aeroplane or making a pharmaceutical than they are to deliver a package. But the same modality for the same engineers working in those domains is coming to them.

So how do you predict where it’ll happen next? Well, not that easy, but you know what is happening because it’s already happening in some domains and because the capabilities are getting cheaper fast.

[00:48:43] Kevin

Just going back to your delivering stuff, because I think that’s something that’s salient to everyone. You talk about, I think you said 80% of the time, a driver is out of the car, and so if you can have robots that can do that last leg, that last 50 feet, then all of a sudden, you get something that’s even more efficient than the gains that we’ve seen. As you said, that’s on its way.

[00:49:14] Mark

The social one is always interesting with new technology. The experiment has been undertaken but not been deployed. You have the human or anthropomorphic robot. You have to carry the heavy package delivered to your doorstop. As everybody knows watching a UPS truck, a lot of times the packages are light, trivial packages. It’s easy to imagine a construct where that truck has flying drones that are relatively quiet, that when he stops to deliver the 20-pound package, it’s delivered 10 one-pound packages to the other 10 houses and he drives away.

So now instead of 80% of his time being out of the truck, it drops to 50%. That’s a big deal. If it dropped to 30%, these are consequential because cuts in half the number of people you have to hire to drive trucks, it allows you to raise the salary of the truck driver, which is what’s happening because we have a shortage of truck drivers. So the productivity boom is sort of nascent and waiting. The most common question I get because of the subtitle, The Roaring 2020s is, so when did the roaring start after all? Come on, it doesn’t feel like we’re roaring.

Governments can sovietize economies. I’m not saying that’s our government. Governments can impede rather than accelerate dislocations in markets. The Great Depression, there’s a very good book on this by an economist so I do respect economists, retrospective economic history. It’s very powerful. Sometimes we’ll draw the right lessons from it. But during the Great Depression, some of the biggest advances in the productivity of the manufacturing sector in the United States took place during the Great Depression.

So you would have said at that time, where’s the boom? Where’s the productivity boom. It was happening because the implementation of the technologies that were being touring and invented in the ‘20s happened in the ’30s. But for that, the US would not have been able to mobilize the war machine they mobilized, but it happened during the Great Depression. So I guess my point is, we’re in that kind of stage analogously. We have a simmering recession. We certainly have a lot of political debate about how to ignite our economy.

I’m an anti-industrial policy person. I don’t think there are people smart enough to manage industrial policy. I think we have to be careful. But that’s what we do. We’re human, right? God willing democracies to do experiments rather than dictatorships. But I think we’re in the interregnum where these adjustments are either just below the surface or on the penumbra, to use a different metaphor, being ignored by a lot of economists. They would say, well, how can I measure it?

That’s why I use analogies like the warehouse. So look at that and tell me why that doesn’t apply to mining because mining has been stuck in the antediluvian ages for a long time. This is really, really hard to do mining, and why that won’t apply to healthcare where we have a labor shortage, and a lot of the tasks in healthcare really can be performed by, I’m going to use a new term, I’m running a piece called dumbbots. It’s a robot good enough to do a dumb task so I could stop having a smart person do the dumb task.

[00:52:23] Kevin

Yeah, I think that’s quite important because if we think about the demographics, you talked about The Great Demographic Reversal in your book, and we had Manoj Pradhan on the show, who co-wrote that book. His thesis is, hey, we’re going to need all the automation we can get, because for every job that’s automated away, a new one is going to be created In age-related care.

What I thought was interesting in some of the bits in your book is like, hey, we can get robots to do cleaning and delivery and so on and that frees up people to do more higher-level service-related tasks. So you think there’s a genuine prospect for increasing productivity in healthcare?

[00:53:08] Mark

I think it’s the first time we’re going to have a chance for productivity going to healthcare. As you know, I put it in my book because it was delicious when I found this research paper from some NHA guys, awhile back. They renamed the healthcare technology trend is Eroom’s Law by just flipping the word Moore’s law and said that you could map out that the increase in spending is mapped with a decrease in output, output meaning healthy patients, drugs. So we have this inverse trend.

So all across the healthcare domain from discovery of drugs to doing clinical trials, which will be profoundly enhanced and accelerated by AI and by sensors into the delivery of everyday care, which is profoundly enhanced by Edge AI and sensors in your smartphone, all manner of things that can be diagnosed more quickly and easily remotely with the assistance of physicians, saving you that visit and time therefore productive, and curating instead of in a waiting room of a hospital or your doctor’s office curating whether you should go there or not, that curation, even if we just cut 30% of the visits down would be profoundly productive.

Eldercare is the most interesting and challenging while given the demographics. The one example that I like is not just that I can do remote diagnostics. Lots of people I meet who have health issues have the EKG in your phone, the add-on device. It’s very common and it communicates with the doctor. When you add AI to that, it’s mapping it continuously so you get real-time alerts. You’re on a bad trajectory call, call a doc. Even the cleaning thing, look, Rodney Brooks had a great line saying that it’s far easier to train a robot to do a lawyer’s job than it is to do a plumber’s job, and that’s true because AI in a machine doesn’t have to do any manipulation. It turns out to be easy.

A lot of law firms now have fewer lawyers because their jobs have automated. But folding laundry turned out to be a much harder cognitive task and mechanical task than everybody realized for robots. We’re getting close to ones that could do that. It’s been demonstrated finally. It used to be that sine qua non, could a robot pick up a dirty piece laundry, put it in a machine and fold it the way a human would fold it. It’s doable. It’s now expensive. But the key rubicon has been crossed, the doable, machines that can do that.

So what does that mean? Well, you want people to be able to live at home not in institutions as long as possible. A lot of the tasks that people have trouble with as they get older are mechanical tasks, opening things, picking things up, moving things around, cleaning. It sounds science fiction-esque and dystopian to say robots will do that. Well, they’re already doing it and there’s many applications and on a limited basis. We see the kinds of anthropomorphic machines that Boston Dynamics and Agility Robotics make. These look amazing to us today.

I’ll use my history again. Well, people were amazed at the first cars in the 1890s. They were blown away. Look at technology. From our perspective, we think they’re kind of goobers. That was obviously not an advanced car. It was pretty amazing. Well, robots are- I guess they’re not 1890s, but I would put them pre-Model T. we haven’t seen the Model T robot. But this is a prediction that I made then, and I’ll say it again, we will see a Model T robot. We will see a robot that was functionally equivalent of the Model T robot, that will have widespread use across many applications. It will probably be first in eldercare areas, probably.

[00:56:46] Kevin

That’s interesting. We’ll wrap it up now. You have this great line of the book where you said, if inventors of cars had asked people riding horses what they wanted, they would have said a faster horse. It’s difficult to imagine a future that’s vastly different from our own. If you want to try to do that, I think this book is a really good place to start. It’s called The Cloud Revolution and its author is Mark Mills.

Mark, I really appreciate you taking the time. We’ve only just literally scratched the surface of a lot of the stuff that you cover in the book. So thank you for helping us understand that a little bit and I really wish you all the best.

[00:57:32] Mark

Thanks for the platform to expatiate on my optimism that we’ll get through the dark days and have a booming future. It’s still possible, but you have to believe it’s possible, not in a science fiction way. You have to believe it’s possible, which makes it worth fighting for and living for. But thank you. I appreciate it very much.

[00:57:52] Kevin

Well, that’s great. With that, we’re going to pass it back over to Niels.

[00:57:57] Niels

Thank you so much Mark and Kevin for a very enjoyable conversation on a topic that is likely to dominate for some time to come. Most of the talk we hear on mainstream media about technology’s impact on the future just focuses on AI. But Mark really helped me see that AI is only a part of the story. We are also experiencing a revolution in machines and materials, and it is the convergence of advances in information, machines and materials that is both powerful and underappreciated. He believes this type of convergence hasn’t happened since the early 20th century.

It’s obviously nice to hear someone like Mark who is so optimistic about the future for the economy and especially because everything he forecasts is based on technologies that already exist. It was also interesting to learn how large the physical footprint of the cloud is. A cloud data center has as much physical space as the Empire State Building, and there are 5000 of them already in the world and this number is going to get much bigger. It was also fun to hear about some of the stranger things we’re going to see, like materials that can bend light in a way that makes them invisible, but I guess we won’t really see that one.

Anyway, that’s it for today. Make sure you go and follow Mark’s and Kevin’s work as well as getting a copy of their books, because as you can tell from today’s conversation, some of these ideas and topics are not being discussed enough on mainstream media. From Kevin and me, thanks so much for listening. We look forward to being back with you on the next episode. In the meantime, take care of yourself and take care of each other.

[00:59:45] Ending

Thanks for listening to Top Traders Unplugged. If you feel you learned something of value from today’s episode, the best way to stay updated is to go on over to iTunes and subscribe to the show so that you’ll be sure to get all the new episodes as they’re released.

We have some amazing guests lined up for you and to ensure our show continues to grow, please leave us an honest rating and review on iTunes. It only takes a minute and it’s the best way to show us you love the podcast. We’ll see you next time on Top Traders Unplugged.