Embrace Abundance: Navigating the Future of AI and Economy
We're diving deep into the wild world of AI and its exponential growth in this episode. Alex Imas, an economist from the University of Chicago, breaks down how traditional economic models, which are all about diminishing returns, might not hold up in the face of AI's rapid advancements. We chat about how the pace of change isn't just fast; it's accelerating in ways we’ve never seen before, raising big questions about what that means for our economy and businesses. If AI can improve itself quickly, it challenges everything we thought we knew about growth and productivity. We also tackle the potential for AI to completely reshape industries, forcing us to rethink job roles and how companies operate. So, if you're curious about the future of work and the economic landscape, stick around—this conversation is packed with insights!
Takeaways:
- The concept of exponential growth in AI is crucial, as it challenges conventional economic models that are typically based on diminishing returns.
- Understanding that AI's growth could fundamentally alter economic models is essential for adapting to new realities in business and technology.
- The discussion emphasizes that AI's recursive self-improvement could lead to unprecedented advancements, but it also poses risks and uncertainties for the economy.
- To prepare for the future, organizations must embrace AI's capabilities while being aware of the potential constraints and political dynamics that could affect growth.
- The importance of viewing AI not just as a tool for efficiency but as a transformative force that can reshape markets and consumer preferences is highlighted.
- Companies should rethink their structures to leverage AI effectively, moving from traditional roles to more technology-centered approaches for maximizing productivity.
Transcript
Hi, this is Adam Davidson, one of the Co founders of FeedForward and your host of this podcast. There's a word I hear constantly right now, exponential. AI is growing exponentially.
And I think most of us, myself included, sort of nod at that word without fully reckoning with what it means. Because almost everything in standard economics, and I mean everything, is built on the opposite assumption. Diminishing returns.
Growth happens, and then it slows down. That's the baseline for basically every economic model ever. So what actually happens if AI breaks that pattern?
My guest is Alex Imas, an economist at the University of Chicago's Booth School of Business and one of the clearest thinkers I've found on what this AI moment means for the economy and for your organization. Hey, Alex, great to have you on the call. Alex Imas. Well, why don't you say hello and then we'll say who you are.
Alex Imas:I'm Alex Imas. I'm very happy to be here with you, Adam Great.
Adam Davidson:You are an economist at University of Chicago's Booth School of Business and write and think a lot about the economics of AI. Let me tell you what's on my mind, and you are exactly the person I was thinking I want to talk to about it.
The word exponential keeps coming up that this is not linear growth, where you can kind of have a mental model of what we've seen in the last year, which is a lot. And so in linear growth, the next year would be roughly the same amount of change.
Exponential means a year of growth in a shorter period of time than a year. Right. Like maybe we don't know, three months, two months, six months. And then so it's like the pace of change changes.
And it seems to me both like our human brains, but also our economic models kind of break down when we start thinking about exponentials. Is that on your mind lately?
Alex Imas:Yeah, I mean, it's on my mind a lot because there's a lot of discussions around kind of the economics of what exponential looks like. I think you're, you're totally right with the. On the fact that our economic models are essentially all based on kind of diminishing returns.
Like name whatever you want, capital, labor. Let's say you, like, find an asteroid with tons of gold. You get, all of a sudden the price of gold tanks.
All of a sudden, you know, that resource is less, less, less valuable. All of a sudden you have diminishing returns to gold. So, like, basically think of in.
In our economic models has kind of these diminishing returns that we are very used with to working with a negative second derivative.
Adam Davidson:A negative second derivative, meaning. And it's been a few decades since I took calculus, but the pace of, the pace of change slows.
Alex Imas:Yes, yes, yes, exactly, exactly.
Adam Davidson:Might still be speeding up, but they're slowing down in how they're speeding up.
Alex Imas:Right. So you're, you're kind of, you know, you're getting growth, but that growth is slowing down. Right.
We're very used to this and like that Essentially every assumption in standard economics is either linear, which you make for simplicity, just because it makes the math simpler, but then somebody in a seminar will be like, actually, like, we should really have diminishing whatever.
Whatever you want to call it, because that's, that's, that's been the facts on the ground for such, for such a long time that like, except for these, like, brief little speed ups, like, for example, like, you know, if a developing country hits some sort of, you know, gets like a technology bump, and all of a sudden it starts playing catch up with the developed world or something like that, you get, you get what looks like not diminishing returns, but for the most part, we're not for a period of time.
Adam Davidson:And then it gets to be diminishing return. Like, China did have truly astonishing growth.
Alex Imas:So did absolutely.
Adam Davidson:Korea, Finland, Singapore, like, name, name your developing economy.
Alex Imas:At some point, you're kind of like the S curve starts.
Adam Davidson:Yeah.
Alex Imas:Starts hitting.
Adam Davidson:Right. And I think intuitively my mind goes to like, the United States somewhere between the Civil War and World War I. Like, we.
We did this massive transition from primarily agrarian to. To an industrial economy. It was astonishing productivity enhancements, but it eventually was diminishing returns.
And then it seems like after World War II, we hit another level, you know, of astonishing growth in the 50s into the 60s. But then diminishing returns. Like, it just probably.
People can think of their own workpl how you see, you know, I think of podcasting coming to my industry of audio journalism, and astonishing growth followed by diminishing returns. So, yeah, okay.
Alex Imas:And it's, and it's, it's interesting because, like, with America particularly, the really interesting thing about American growth is that it's been linear, right? It's been 2%. 2%, 2%, 2%, 2%.
But if you actually look at what is generating that growth, it's a bunch of exponentials flattening out over and over and over and over and over again.
And then when you get that, when you get like exponential leveling out, exponential leveling out, you get what looks like, you know, what looks like Linear growth.
Adam Davidson:Yeah, that's really. Right. Right. Like, if we only had a agricultural revolution or we only had railroads or we only had industrialization, Right.
We'd be like, how do we look at the last 20 years? Yeah. Or 30 years. Yeah. And that is the case for most of human history. Right.
You would have like the agricultural revolution, the very first one might have been worse for the individual, but. But seems to have created all sorts of new things, pluses, big cities, et cetera.
rom like Mesopotamia's heyday: Alex Imas:Yeah, exactly, exactly. There's almost no growth followed by periods of negative growth. Right.
So you have like the Dark ages where you had actually like severe negative growth.
Adam Davidson:Severe negative growth. Right.
Alex Imas:Okay.
Adam Davidson:So while they're happening, exponentials always feel like they're going to go on forever. Is that probably true? Is that your guess?
Alex Imas:That's true, yeah. That's very true.
You know, when, if you're living through, you know, 20 years in China, it feels like, oh, my God, we're gonna surpass the United States. But at some point you hit the, the diminishing returns with, with the Soviet Union. They're famously, you know, Paul Samuelson, like the.
The biggest names in economics were very worried because the Soviet Union was like. Was basically like. The United States was like this. And the Soviet Union was like this.
Adam Davidson:They were saying it was flat and, and, yeah, like up. But. But like. Yeah, like a simple linear.
Alex Imas:Linear. Linear, exactly. Whereas the Soviet Union was completely in.
Adam Davidson:An exponential, like a terrifying skate park or something.
Alex Imas:It was exactly. Only like you're. You're basically, you know, going to be into a. Yeah. And basically that kind of.
The top economists were like, super worried about this.
They were thinking, you know, maybe we should be thinking about central planning because this seems to be getting exponential growth and we're not having exponential growth.
Adam Davidson:Right. Or Japan in the 80s. I mean, I remember that vividly, there was real fear that like, basically America will become a fiefdom of. Of Japan.
And then, so, okay, so we're done. Then we're in an exponential. But it's going to peter out. It'll be diminishing growth. And we're done. And we're done.
Alex Imas:Absolutely not going to say that.
Adam Davidson:You're not saying that.
Alex Imas:I am describing the past. That's what I've been doing so far.
Adam Davidson:Yeah. Right. Okay.
So there's a thing in economics, this came up a lot during the financial crisis that in fact, one of the most famous books that came out by economists was called this Time is Different. And it was specifically about the risks of thinking this time is different. And that was in a different context.
properly, which in the early:So whenever I say this time is different in my brain there's a little voice saying that's what people say right before bubbles and then the bubble bursts and you're wrong. But can you make the this time is different case? Is AI different?
Alex Imas:Yeah, I mean the case for AI being different is that we've never encountered a situation where we are replicating general intelligence at scale that could potentially improve itself as far as the frontier of its own intelligence.
This is something that's completely different in the sense that if you can get into a cycle where the intelligence, the frontier of intelligence is increasing as the intelligence is working on itself, why do we get periods of exponential growth? That's ultimately the question. Exponential growth comes from technology shocks. Technology shocks come from intelligence shocks.
Like we have figured out something that we have not seen before, like the railroad. Somebody discovered how to put tracks and connect different cities. All of these are really knowledge shocks. And these are coming.
Usually these knowledge shocks are coming from a bunch of scientific ideas fitting together. And all of a sudden they come together into something that's economically viable.
But the speed of that has always been the human speed, which is non recursive. Right? It is, it is building on itself. Right. You have cumulative knowledge.
But it's not like, it's not in the way that we're kind of thinking about RSI with AI, which could really kind of, at least in the models, can kind of explode. So if I was to make like a Steelman case for like this time, it's different.
It's all coming from this recursive property of, of artificial intelligence. Now you know, I can make, just.
Adam Davidson:To be explicit, the models can improve themselves. So they don't need a bunch of slow bags of whatever we are protoplasm thinking about how to improve them.
They can improve themselves and they just operate on a much, much faster cadence.
Alex Imas:Yeah, exactly, exactly. So that computer speed is a lot more, it's just much faster. Right. So.
And then you could also like there's a lot of properties of digital intelligence that human kind of like soft, you know, cellular intelligence doesn't have like for example, like you know, memory scales.
Like you could, I can, you can take a database and give it, you know, you could basically send that package over to another agent and it has the, you can replicate that intelligence right away. It's. The computational processes are happening a lot faster now.
Learning is something that it seems like humans are completely still dominating in like the speed of learning new information. Like you know, you take a teenager, you put them behind the wheels of a car within a couple hours, they're like kind of driving.
The number of bits that have gone into that teenager's head is tiny compared to like the amount of bits that need to go into like an LLM for it to learn a new skill.
Adam Davidson:Right. Or a self driving car to learn how to self drive and make creative decisions.
We recently had Felix Reisenberg from the guy who runs Cloud cowork on and he was. Where his mind was going is ambiguous areas. How, how AI will be able to be useful in increasingly ambiguous areas.
So not just we need a strategy plan, here's our parameters. Should we expand in Germany or move to Africa or something, but rather just give it, you know, how can we grow?
It can kind of figure out the ambiguities and you can increasingly trust it to make decisions that to me like when I think of a teenage, my son's. I'm a little terrified of my son learning how to drive, but that's a separate conversation.
But a human being can handle outlying cases even if they haven't seen it before. Whereas we know computers just need a lot, a lot of. So if they can start making their own judgments both about actual work but also about themselves.
Like what?
Alex Imas:Right, right.
Adam Davidson:What's the next thing that would make me even better?
Alex Imas:Yeah, I mean like that is, that is one of the things people are working a lot on.
And like the frontiers of research is like this like sort of meta intelligence of like understanding what I'm capable of as an, as an AI agent and having a mental model of oneself. And then that requires that essentially facilitates how I can intelligently interact with the world. Like what information?
Like our information acquisition is endogenous. We're not just like fed information and we're updating on that information like without any sort of constraints.
We're saying like look, this is the information I want to update my weights on or this is the information that I want to update my weights on. We're getting like a ton of information constantly. And we're not, we're using like a fraction of it, but it's not randomly determined.
It's the information that's most useful for the, for, for, for what we need it for. And I think that requires kind of a meta model of what is useful. Right.
Adam Davidson:And we know our brains are very good.
If you take two newborns and put one in a hunter gatherer tribe and the other in a big city, as I understand, their brains will actually wire to perceive things differently so that they are in infancy and early childhood, they're actually developing the right pathways.
So AI that's able to do that, like, oh, I'm an AI inside of a large corporation, or I'm an AI design that's really focused on creating compelling artwork or whatever it might be.
Alex Imas:Right.
Adam Davidson:That it can decide, okay, I'm going to extra strength in this thing. One question I have though is for the next three to five years, say, does it matter if it's almost certainly going to be exponential?
I would say for the next several years. Right. Whether it then hits diminishing returns or not. Does that feel like a reasonable.
Alex Imas:Yeah, I think the thing that I want to really clarify here is exponential on what? Right? We need to think about exponential on AI capabilities, on coding, or is it exponential with respect to economic outcomes of AI?
Is it exponential with respect to improving itself and improving the hardware around it? So all of these questions are super important because you can get what looks like exponential, like the famous meter graph, right?
And if I was just to look at that graph and think like, wow, a whole bunch of our economy is kind of looks forward of like coding and this thing is going exponential. I'm going to look at the stock market or the GD GDP and I should see exponential there. But like we don't see like anything like that. Right?
So I think it's really important to define what we're thinking, what we're calling exponential. I think it's very reasonable that and I think there's been absolutely no evidence to suggest that things are slowing down in AI development.
Adam Davidson:Right.
Alex Imas:And Anthropic has just put out a report saying like, you know, they're, they're seeing XR signs of RSI and explain.
Adam Davidson:What rsi recursive self improvement. Maybe just walk through what, what recursives, why that's important, why RSI is.
Alex Imas:RSI is basically the exponential. It results in an exponential, but essentially saying like, look, right now there's just like these wet humans bottlenecks in, in, in AI research.
It's like a person like thinking like here, this is like the sort of architecture that's going to work best. This is the sort of data that I need to curate perfectly in order to get this sort of improvement.
Imagine doing all of that at computer speed where you have an AI agent doing all of these things. These are AI models able to improve themselves. There's no human in the loop.
It's just an AI scientist and an AI scientist is saying this is what we need to give the model. This is the new architecture we need to try and these are the experiments we need to do and running that all on computer speed.
And then the resulting model is now the new AI scientist for the model itself. And then now it's the new AI scientist.
So the AI scientist is getting smarter at the speed that the model is improving in improving the next iteration of the model.
So the knowledge is now arriving instead of it's arriving kind of at this like human speed which is, you know, every couple decades we get some new idea. Now theoretically AI knowledge about AI self improvement is arriving within, you know, hours, days or something like that.
Adam Davidson:So this is where I do want to get into like the thinking failure modes because I think of two failure modes and I feel like I do both of them about 100 times every single day. So one is being unimaginative.
Like you know, I'm old enough to remember when the idea computers were thought of as these giant things that existed in rooms. And I have a memory of telling my mom that my teacher said that one day everyone would have a computer in their own home. And that seemed insane.
And but it really was hard.
Like I wasn't very few people, maybe nobody or almost nobody was able to really imagine if, if we're really going to see that miniaturization where like, of relatively cheap watch has the computing power of like the entire space race.
You know, nobody was thinking about social media or you know, Fitbits or whatever, not, not to mention, you know, RFID chips or whatever, like a million things the way the world is different. So on the one hand we under imagine where we kind of think in a linear way, even if we are trying to think exponentially.
So when you say rsi, you know, my mind goes to well what did I use AI for today? It'll just be better at that and faster at that. But really it's the AI is talking to the other AIs about AI stuff and, and my brain goes blank.
But the other risk is like, so one is under imagining the other I find is we just hurdle at either a totally unconstrained utopia or dystopia like like, then it's like, oh well, none of us will have to work. We'll have everything we want. Agriculture is solved, luxury is solved. We're just a bunch of people living and having all of our dreams fulfilled.
Or it's the opposite. There's, you know, Sam Altman and three other people will have all the wealth and the rest of us are going to be slaves working for the AI.
And maybe eventually Sam Altman will too. And, and it's the Terminator.
And it seems like anywhere in between, like Under Imagining and Utopia slash dystopia, it's very hard for our brains to even know like where, where should I be?
Alex Imas:I, I almost think it's like the, the forecasting to me on an individual level is like just a, I don't want to say a waste of time, but it kind of feels like that. So think of the movie.
Adam Davidson:The, the, the, the.
Alex Imas:My favorite example is the movie Alien. So they have hyperspeed travel. They have androids that simulate all human interaction.
They cannot be, you cannot tell an Android apart from a human being. Perfect, you know, intelligence beyond belief. And they have all of these things which had required exponential growth. Right.
You need exponential science in order to get to hyperspeed travel within, you know, the sort of lifetime that they're, that they're thinking about. Right, but what are they using within those spaceships? They're using computer screens that are running on dos.
Adam Davidson:Right.
Alex Imas:Do you remember that? Do you remember the scene where they're.
Adam Davidson:Interacting with the computer with like the green letters? Yeah.
Alex Imas:So it's like, and this to me is, is forecast. This is the most beautiful picture of four.
Adam Davidson:Right. They had some of the best, most creative minds in America.
They had a huge budget and they couldn't picture Windows, let alone like three dimensional virtual reality.
Alex Imas:Yeah, exactly.
And that, you know, in Blade Runner it's the same thing, you know, you're interacting with these androids and then it's like, oh, this thing comes out of my pocket and there's like a little DOS screen and I'm like, you know, boop, beep, boop, beep, boop. And it's like every sci fi in the world is like that.
So it's like the thing that's really hard to imagine is like, where are we gonna, where are we going to improve? And how the speed of that improvement. Usually other things are improving. Sometimes things are stuck.
Sometimes they're improving it even faster than the thing you're considering. Usually people are just kind of focused on the one thing that's improving and kind of holding things constant, which gives you a very.
But what the future really is in reality is a interaction of all of the things that are changing. It's not the individual things that are changing by themselves. So I get really frustrated when thinking about economics.
Like in like 100 years or something like that, people are like, well, we will have, you know, artificial superintelligence and hedonium seated in the universe. What do you think the labor share is going to be for humanity? I'm like, if we have asi, you think we'll be thinking about labor share?
You think we'll be thinking about the GDP and like money? Like, what are you talking about? Like, the economy is going to transform too. You can't like, say, like, oh, here's the Fed.
The Fed is here and Jerome Powell is in it. But there's Hedonium. And what's Hedonium?
Adam Davidson:That's not.
Alex Imas:I just thought Hedonium is a recent thing I've picked up from my travels within the AI AI world. Hedon is a. It's matter that has a utility function. It's like an intelligence whose only purpose is to be very happy.
Adam Davidson:Wow.
Alex Imas:So it's a simulation of happiness. So like a utilitarian would want to sacrifice. If you. I'm a true utilitarian.
I would want to sacrifice all of the world's resources in order to create Hedonium. Because I'm maximizing happiness, right?
Adam Davidson:Yeah.
Alex Imas:So a utilitarian would want to seed the galaxy with hedonium and that would maximize happiness. Anyway.
Adam Davidson:Gotcha.
Alex Imas:If you talk to some of these folks, this is a.
Adam Davidson:And they're into it, it seems like.
Alex Imas:Yeah, but. But you also have Jerome Powell somehow, you know, doing interest rates at the same time.
Adam Davidson:Right, right, right. And like, by then the Fed will clear your checks and in a day and a half instead of three days. Yeah.
Alex Imas:You'll be able to get through to the irs. There's going to be a phone number, right?
Adam Davidson:Yeah. It's interesting. Your science fiction is generally about the moment, Right. You read Cold War era science fiction.
It's all about Cold War era stuff that you read stuff today. Also the way we look at history as well. I know a lot more about archeology, for example, than I do about science fiction.
And you read:It's a lot about individuals and what did women and minority ethnic groups and how did they have power? And part of that are the tools available. The tools in the 50s lent themselves to looking at giant structures and the tools lately.
So anyway, so it's interesting. I think I would guess that the way we imagine the exponential is a reflection more on who we are and what our own fears and hopes are.
And also our own political position in the world as well. I would guess so. You are an economist and economists the dismal science. It is a lot about constraints and trade offs if we want to.
Because while on the one hand I agree that any effort to predict any with anything like precision a forecast is absurd. When I ran Planet Money I used to joke that I wanted economists when they made a forecast about growth. They should say we think it'll be 2ish.
When they say we expect trend growth to be 2.03% or something. It's like.
Alex Imas:You want to have a fudge factor.
Adam Davidson:Yeah. And we're okay at predicting when things go the way we expect them to go, but we're terrible at inflection points. So 2ish.
With a 50% chance we're completely wrong. But at the same time economists are helpful at thinking about constraints and trade offs.
going to be like in February: Alex Imas:Yeah. So the way that I like to think about economics is that we are quite good at thinking about constraints and trade offs and optimization.
What that means is that our comparative advantage is not saying these are the assumptions that are most likely.
It's to say let's take a scenario and let's work backwards as far as what constraints need to be relaxed, what constrains need to interact, what, what sort of optimization needs to be happening for this scenario to, to come about. And then we can say hey, this scenario is actually pretty silly because this is what needs to happen for it.
So like I was, I, you know, I had a, I wrote a paper up in January showing that this like scenario that we would get negative economic growth from AI is actually very unrealistic. Because you can say, you can take that scenario, let's say we have negative of economic growth, what does it require?
And then you could work backwards and say this is what it requires. And that is a very crazy condition that would never be met in practice.
Basically it requires that like rich people run out of stuff to spend their money on.
Adam Davidson:Right.
Alex Imas:I don't think it's going to happen.
Adam Davidson:Right.
Alex Imas:So you could have positive economic growth, an exploding gdp, but like all of the wealth concentrated in a smaller number of people because they're just finding stuff to spend money money on. So let's go back to your question. Like what is kind of like a scenario of this sort of like exponential growth? I think so.
Antab Kornek and Basil Halperin wrote a really interesting paper about this and it's a theoretical paper, but basically they did this exact extra exercise like what is required for economic exponential growth and what is required for economic exponential growth. It turns out that it's not just software AI rsi. It's not just like, let's say we just get the meter graph and it just keeps going.
You're still not going to get exponential growth in the economy because what you need is the complementarity exponential growth in hardware innovation.
If you do not get exponential growth in hardware innovation, which we haven't seen any of by the way, then you will not see that exponential growth even if you're getting exponential growth in like sort of like the longest time that a coding agent can be working on a problem in order to solve.
Adam Davidson:And you keep referring to meter which we feed forward. Members should know. It's the graph Ethan shows all the time.
But metr it's basically how long can and agent an AI coding agent go and have a 50% chance of being accurate. So it's not even, I mean you could, you could make it 100% and the graph would just be a little bit shorter. It's not that big a deal.
But it's, you know, not that long ago it was line by line.
Like when I would write code, you, you get one line, you copy and paste it, you put it in, you know, or maybe, you know, you get 30 lines and the likely to be error prone. And then once you have like 300 lines, it doesn't remember the first 30 lines and blah blah, blah.
And now there's people meaningfully using AI for hours and hours and hours. And you know, I would say my experience is about 50% of the time it works well.
But what if it's days and months and then if we throw an RSI and there's what it's able to do in that time period is much more advanced. It might be, you know, in seconds it's doing what today would take 10 hours. But, but that's a software thing.
When you say hardware, I'm guessing you don't just mean chips. You mean like drones.
Alex Imas:No, I mean chips too. No, I mean chips. I mean chips. I mean chips too. Like you look at, look at the amount of, of compute that you need to train the next better model.
It's exploding too. The resources that you need then that's a hardware issue. Right.
For getting the next model to improve prove in the way that we're used to is blowing up. So we're literally going to hit a resource constraint.
ssing the names but basically: Adam Davidson:Because we'll have used every resource on Earth and that we can't until we find that asteroid with lithium or whatever. We can't.
Alex Imas: ctly. Yeah, yeah, that's what: Adam Davidson:Wow.
Alex Imas: re either going to hit AGI by: Adam Davidson:Wow.
Alex Imas:And I think, and that's because of this sort of constraint that you're, you're, you're, you're moving, you're, you're going exponential. Given the resources that we have access to where we have slack, we're constantly diverting resources to AI, right?
So it's not like it's a zero sum game. We're not like finding new resources out there.
We're, we're using the resources that we have and we're diverting them to AI and what the resource flow looks like because it's also increasing exponentially.
At some point we're going to hit that constraint and if we have not reached AGI or some sort of ASI that allows these systems to become very much more efficient at an exponential rate. Which I mean they are becoming much more efficient. They totally are. It's still just not enough.
Adam Davidson:We need it to be at the same or faster rate than the software. So we need to be redirecting a lot of that self improving AI software at the problem of creating better, more efficient chips and that.
Alex Imas:And then so you need like rs, you need this exponential growth and like in the hardware. And to me, you know, and I, I'd love to speak more to the, and I've spoken a bit to, to, to some of the co, to some of the authors.
I think they, they, they at least mostly share my view. I think that's the biggest bottle Is the fact that we are really not seeing, seeing these.
We're seeing exponential in certain domains and that could go on for a while, but unless the other. We're going to start seeing exponentials in many, many other domains. I think that would naturally generate a slowdown. Now, maybe not.
Maybe that expo, maybe that exponential. We're going to hit a point where that exponential is going to allow us to hit exponentials in the other domain. Right?
You have these complementarities between different domains and then all of a sudden you have speed ups all over the place. And then we really hit, you know, I don't rule this out at all.
Adam Davidson:Right.
Alex Imas:If we get enough intelligence with software, then all of a sudden the software could start getting us exponentially all.
Adam Davidson:And then electricity is another. I have a friend who's getting her PhD in this and her view is we are not on any kind of path to have enough power to do what we.
You know, that's why we're suddenly talking about nuclear and, and that's why people.
Alex Imas:Are talking about Dyson spheres all over the place.
Adam Davidson:Yeah, yeah.
Alex Imas:You know, that's what you need. Like this is like the conversation with, had we, we had with particle physics, you know, 10 years ago.
Like what do we need to unlock this next discovery? We need a particle accelerator the size of the galaxy. That is a conversation that physics physicists have. Maybe one day.
Adam Davidson:Maybe one day. That's very big though. That is rather large.
Alex Imas:It's big. Yeah, it's a little bit, yeah.
Adam Davidson:However, going at the pace we're going, and again the exponential pace we're going. If we do have another three or four years of this, of, of the not just growing, but growing. The pace of change growing.
Even if we hit a wall in: Alex Imas:Oh, absolutely. I mean, don't, you know, we're talking about you kind of like you're asking me to talk a lot about constraints and bottlenecks and things like that.
If you ask me to like kind of take a step back. And I think Ethan's made this point already. So this is not a new point. Just getting the capabilities we have now diffused through the economy.
Let's say progress stopped today, which it isn't. It's exponential. Let's say progress stopped today. We have a transformative economic technology on our hands just with what we have now.
The reason that we haven't seen it have this sort of completely transformative effect is because it hasn't diffused.
Adam Davidson:Yeah, I remember someone this Is years ago pre AI, but a economist saying we always think of Silicon Valley and these wild disruptors, but if we could get every business in America 5 or 10% more efficient, just learning the basic stuff you learn in business school, that would have more wild growth potential than a dozen Facebook or Googles or whatever. I don't know if that. I forget the math and the numbers, but that kind of idea. But as we diffuse it.
So getting to the practical, the feedforward member constraint, people are all over the place and thinking about this. It's hard to think about. What I find must be so hard is walking around your office, all right, Frank does this, Joanne does that, Bill does that.
Some jobs I'm sure you can look at and be like yeah, AI could basically do that. Now others you're like, oh, we need an entirely different way of approaching that problem.
The whole way we approach whatever the challenge is developing new products or interbank lending or whatever. The thing is, AI could be a totally new way, but rethinking the organization itself. What does a CEO do? How many people?
Is it the same number of people? Is it a hundredth as many people? Is it a hundred times as many people? I find my brain starts to break.
I think a lot of members a little bit start to break. What do you.
Alex Imas: me. He started the company in: Adam Davidson:They're an insurance company.
Alex Imas:They're an insurance company. So basically the idea is like starting with automation as like a first principle, like having like chatbots, however supported by people.
But like basically the org was designed around technology rather than around people who are using the technology.
So there, you know, if you look at kind of the history of like technological change or whatever, if you a company is growing revenue, it is increasing its payroll because you're hiring more people, you're like getting more capital. But who people need to be using that capital. Lemonade has exploded in growth.
I mean it's a big, big billion dollar company but it has shrunk headcount since its inception. And the reason that is is because the company was never organized around people. It was organized around using technology in the best way possible.
This is, these are like these AI first companies.
Jack Dorsey had a post about this like trying to make block A AI first company and like the organization around that and how difficult this was going to be. And people are still kind of playing around this idea.
But this is all to say that I think it's very difficult to say, here's my company, here's everybody who's employed. How do I organize them around AI? That's very difficult.
It's much easier for a company to start from the ground up and organize as it's growing around AI.
I think for companies, for existing companies, there's this mentality that my workers are becoming more productive, therefore I should fire them or things like that. I think to me that's a failure of the imagination. Each person can now do more.
That means you should ask them to do more, as in not just more output of the way that they're doing. But imagine for a long time this is how the economy was working. You have, you know, 350 million people in America, however many.
I have a big company like Walmart or whatever. I'm manufacturing something and everybody has a latent preference. So I like this type of product. You like that type of product.
Walmart cannot cater to my latent preferences. It can't cater to your latent preferences because taking a product to scale requires a lot of fixed costs.
So they give us an average product that kind of gets us some way of the way. There's. What is AI able to do? It allows companies to cater to the full heterogeneity of people's preferences.
It allows me to get what I want and you to get what you want and for us to pay more for the product as a result, because we're not compromising.
Adam Davidson:Right.
Alex Imas:And that is a completely different business.
Adam Davidson:I mean, that is something I think about all like, for people remember from micro econ 101 that we price on the indifference curve. We price at the point where you're like, I got a buck, you got a candy bar.
I kind of don't give a crap, but I guess I'll give you a buck for the candy bar. But what if you could create a candy bar where I'm like, oh my God, I need that. Here's 20 bucks. I sometimes think about that. The marginal revolution.
Everyone's talking about Jevons these days. But that embedded in. When you walk through a supermarket, there is a lot of that embedded in it. That we are pricing for the marginal consumer.
Alex Imas:Yes. Yeah, for the marginal consumer, not the distribution of consumers.
Adam Davidson:Yeah.
Alex Imas:And if you can price for the distribution of consumers, each company can Be making way more money. And guess what? The consumers are much happier.
Adam Davidson:And you're happier, right? Exactly. You want.
Alex Imas:And the reason you couldn't do it is because scale, you needed a lot of scale in order to produce a Snickers bar.
Adam Davidson:Yeah, that's a point I often make. I actually went pretty deep into the history of Snickers for this book I wrote the Passion Economy.
And the point I made is for the Mars family to sell Snickers bars to someone in Philadelphia, they basically had to sell to everyone between Chicago and Philadelphia. And now they can find Alex Emos in Philly and Adam in Vermont and give us just the thing we want. And now we do have that.
It's called artisanal food, but it's a terrible business. It's a great way to lose a lot of money.
Yeah, it's also, I mean, the other mental model I have is if you have a factory and I said, by the way, I have this magic thing, it's like magnetic and you just put it on a mach on any of your factory equipment and it will instantly be twice as productive for the same input cost. Would your first thought be, oh, then we better throw out half these?
Alex Imas:Yeah.
Adam Davidson:Your first thought would be like, oh, let's run them 24, seven. That's amazing.
Or if you have a retail store, and I said, oh, I have this little trick and it's going to quadruple your sales per square foot of this retail store. Your first thought wouldn't be great, we're going to make the store a quarter of the size so that we can make the same amount of money.
So that's one constraint.
The other constraint, which I think in the past, I think economists are much more sensitive to this now, is the political economy, that there are market forces, but then there are political reactions.
This might be an overstatement, but my sense is that for decades the argument on trade from economists was generally trade is good because it increases the overall pie for everyone involved or for all the looking at for the countries. And it took a while to understand why are people for NAFTA and trade restrictions and stuff.
And eventually some pretty compelling work said, well, yes, it increases the overall pie, but it has distributional consequences that are a minority suffer, but they suffer so much that it brings down the average for people, although capital does much better. And so that's the other piece, I think, is that a lot of what we're talking about.
I mean, at some point, if it's energy infrastructure, if it's raw materials, we're talking about political, geopolitical, probably ultimately military.
You're talking about a whole sphere where a frictionless economy of perfect information is not the right model to think about how things are going to play out.
I remember a friend of mine, well, Simon Johnson, Nobel prize winner, when he was chief economist of the IMF and he was in charge of their forecast, he just said we just assume something catastrophic and crazy happens that we couldn't possibly model. Although you can kind of model political dynamics. Like we couldn't get it precise, but we could.
We're already seeing, you know, oil and hydrocarbon energy, you know, clearly China, us, you know, the rare earth metals. There's a lot going on. How do you think about that stuff? To me, those are just most likely going to be frictions.
They're not going to be accelerants of, of growth and transition.
Alex Imas:Yeah, I mean, political economy to me is like the big kind of like the big thing that we are completely under investing and understanding with respect to AI. So like let's say AI causes a 3% increase in unemployment. You think it's going to go to 4% after that. You think it's going to go to 5% after that.
You know, Andy hall has this research where at 2%, once unemployment hits 2%, whoever is in party is going to be voted out.
That puts a lot of pressure on the political system in terms of like what they're going to be, what sort of thing that they're going to be doing with AI and what does it look like?
Okay, we're going to let AI grow and then we're going to redistribute those gains, like set up like a universal basic income, universal basic capital program or something like that. Or are we going to say we are going to ban the use of AI so we're going to basically kind of handicap it so it cannot allow it.
It is not able to get people to be unemployed. So China's kind of going on that, you know, in that direction.
So you, there's, it's, it's not law, but it's their kind of directives that are starting to be in place now where you can't fire people because of AI.
Adam Davidson:For example, in America, it almost seems like we have a law that you have to fire people because of AI because people keep firing people and saying it's because of AI even if it's not. But I have heard of labor unions, big powerful labor unions beginning to have.
We saw it with the Writers Guild, really strong anti AI rules and someone Told me about a recent case of healthcare workers where the contract just said, no AI, we're not negotiating no AI. Which is impossible. I mean, AI is so embedded even.
I mean, if you're talking about brand new generative AI, maybe, but even there that's going to be hard. But you know, Google Maps uses AI.
Alex Imas:Like, it's, it's, it's, it's gonna be, it's gonna be super messy. It's gonna be super, super messy. Right. But it's just these are not, as you said, these are not accelerants to growth.
Like, we don't know exactly what's going to bind and how, what's going to be enforceable, what the regulation is going to look like, what the policy is going to look like. All we can say is that they're not, it's not going to increase the speed of growth.
Adam Davidson:Growth.
Alex Imas:Right. These are all constraints.
Adam Davidson:These are constraints.
Alex Imas:Yes, these are all constraints. We don't know what the constraints are exactly. And to the extent to which they slow down growth, we don't know what that percentage might look like.
So like UBC will constrain growth because, you know, more universal is basic capital. Ubc, basically, it's like, it's essentially. Remember privatizing Social Security under Bush. It's like that for the entire population. Right.
Everybody gets a portfolio.
Adam Davidson:Everyone gets a portfolio. So it's like universal basic income, but for retirement savings.
Alex Imas:Yeah, yeah. And, and that you live off the dividends, basically, in the mean.
Adam Davidson:Yeah.
Alex Imas:So, you know, all of these things are. Well, you be universal. Basic capital is unique because it probably will not slow down growth by much, but it's like, but almost everything else will.
So the political economy is just like the big red herring here. And I mean, these are not hypotheticals, you see, you know, like what's going on with anthropic now in the government.
You don't, we don't have to think about hypotheticals. Although now anthropic just, you know, there's announcements that they managed to train another model somehow through all of this.
Adam Davidson:Right. So.
Alex Imas:So it's not really clear what it's doing for growth, but it's certainly like fable. Not being out there in the economy is certainly not increasing economic growth because of AI. This is a constraint.
Adam Davidson:That's a whole other thing that is worth a separate conversation. But I started to imagine a regime where, because this is another thing we haven't talked about, maybe this is another conversation.
But how much do the big AI companies eat of the pie? And so like Anthropic OpenAI, Google.
And if we do have a regime where the only people allowed frontier models are US Citizens, and the only way to give it to them is they work for those big companies, does that mean that the big AI labs are permanently five months or eight months ahead of the rest of us, and so do they just build everything? Like, they become all the law firms, they become all the banks, they become all the consulting companies and movie studios and everything else.
Alex Imas:Right. And I mean, then it. Then it, like.
And then you're asking the question like, you know, the open frontier is kind of like, you know, a few months behind, and now it's closing, you know,.
Adam Davidson:With GLM, GLM 5.2, which is amazing. It's now my.
Alex Imas:Which is amazing. Yeah, it's amazing, right? And it's an open model, like.
And now the question is, like, did they come up with that model through independent research and compute and scaling, or are they kind of somehow, you know, distilling most of it essentially from the frontier?
Adam Davidson:Right. Yeah.
Alex Imas:And the answer to that question determines what if the frontier, the US Frontier, becomes completely closed off and insulated? Will the open frontier keep it keep expanding? Or is it going to be just kind of degraded because the closed frontier is more closed than before?
Adam Davidson:So I want to leave our members. Here's how I picture our average member.
And maybe I'm wrong or only right for some of them, but at least three times a week, they have to go into a meeting and sound smart about AI and they hear from us that the smartest thing to say is nobody really knows, but here's some ideas. And in the market of big corporations, that's not always an acceptable answer. So what can we give them to say, or what can we give them as a guide?
t can't be. Let's find out in:You got to deploy capital today. And I know you're an economist, even though you're at a business school and I went to University of Chicago.
There's like a unspoken or sometimes spoken rule there that if you have. If any of your research is practical in any way, then it must be deficient. It's sort of, I think, of the.
Alex Imas:Not at the Booth school of Business.
Adam Davidson:Not at the booth school, though. Yeah, you're a different part.
But I think of it as the temple of that Reagan quote that economists are people who see something happening in practice and, and wonder if it could happen in theory.
Alex Imas:No comment. My friends are way too close to us across the street.
Adam Davidson:Okay, all right, so all right, what do you do? What's your advice?
Alex Imas:I think it's really hard to plan for exponentials. That's what we started the conversation with.
And because it's so hard to plan for exponentials, you need to deploy your workforce in a way that takes advantage of the kind of windfall from those exponentials while minimizing the downside.
And I think to me, you know, if I was going to advise somebody, it's to, to, to empower your workers with AI in the sense that have them, have them think of ideas like let's say they're developing software or developing product. Think, Think less about the marginal customer, think more about the distribution of customers.
What new products are you not making that you would if you had more productivity, another, you know, 10 more employees on your team. Guess what? You do have 10 more employees on your team.
Adam Davidson:Yeah, no, I like. Yeah, yeah. And also what, like what information is being constrained by.
I know a guy who's in venture that supports CPG products like beverages and snack foods and stuff. And he uses AI. He said before AI it was like $300,000 minimum to do due diligence on a target.
And he sees the cost of due diligence falling to something close to free. Maybe it's $8 today or whatever.
And what that opens up is you could start imagining whether venture or private equity or something investing in a corner bodega or investing in. They're separate things. To wonder if we want PE firms taken over all our bodegas. But that's a separate conversation. But yeah, not yeah, I like, we.
Alex Imas:Need, we need the loose cats.
Adam Davidson:We need the, we need the cats we need and the looses, the loose cigarettes too.
But you know, I imagine like you're going into a meeting to discuss say acquisition targets or to analyze your competitors and you're looking at X amount of information. What if you could look at a million times X? So you're not just going in and evaluating the top 10 competitors.
You're, you're looking at social media and evaluating every pop up business in India and China, even Argentina. You're evaluating trends beyond trends and you're getting it synthesized into clear reports.
Like what is the 10,000 x 1,000,000 x 100,000,000 x version? I just created a tool this morning. Like I watch all these YouTube videos on how to use AI. There's like many. So smart people and stuff.
I decided it's taking too long and some. And you don't know. So I just created a tool this morning.
Took me like five minutes that just scours YouTube for good AI videos, summarize, downloads the transcript, summarizes it, compares it to a database of things I already know about, and then tells me, here's the top five things you should try. And then I can sync that to another tool to see what are people saying about this new thing. So embracing the.
I guess that's the way to plan for an exponential. It's to embrace the exponential.
Alex Imas:Embrace the exponential. Like. Like today I was. I'm, you know, I'm writing something where I need, you know, think about market research, right?
Market research was like, here's a product that I want to take to market. I need to do quality control. I need to do tons of surveys. I need to get people in a room, all of that before I can even think about things.
Taking this to market instead, what I can do is like simulate the whole thing in a sandbox. Get some. Get a bunch of AI agents, Give them Personas of the demographics that you're targeting with the product. Maybe you're trying to target.
Maybe you're trying to ask the question, what demographics should I target with this product? Then you ask the different question with AI agents, simulate the whole thing and automate that whole pipeline. So you're take.
You're bringing not one product to market.
You're bringing, you know, a whole, you know, the same basic underlying product, but you're differentiating it and hitting seven markets at the same time.
Adam Davidson:Yeah, yeah.
Alex Imas:The scope of what you can do is just incredible.
Adam Davidson:And it's gonna take. And this is stuff Claudine Gardenberg has talked about. It's gonna take a while to get your head around this edition.
Like, I worked at Sony Music for a few years, and a story they talk about is when Napster and BitTorrent and file sharing came out. Their model was built on artificial scarcity that, you know, they kept. They printed CDs and albums and cassettes.
I mean, it was kind of like De Beers diamonds or something. Like, they kept it scarce to keep the price artificially high or OPEC or whomever.
And then their mental model didn't allow them to handle free or abundant. And it took them more than a decade to get their head around abundant. You know, they were suing their customers.
They were, you know, every time someone listened to their Music. They were mad about it. And then eventually they realized, oh, we can't get 12 bucks once.
We can only get a fraction of a penny, but we can get a fraction of the penny all day long over and over and over again. And that allows us more information about who they are, what they want, so we can give them more of what they want.
You know, I remember not that long ago, you know, Sony might sign 100 artists to hope that one makes it big. And my understanding is now it's a much higher hit rate because they know more.
We can have a separate discussion about whether the music is actually better. And, you know, everyone knows I have.
Alex Imas:An opinion about this.
Adam Davidson:Yeah. So, yes, it is not better. Yes.
Alex Imas:Not better.
Adam Davidson:It's not better.
Alex Imas:It's absolutely not better.
Adam Davidson:Yeah, fair enough.
Alex Imas:We're two old dudes talking about music, so.
Adam Davidson:Exactly.
Alex Imas:You know, well, you need to bring in one of your Gen Z friends.
Adam Davidson:Yeah. Or my 14 year old son who plays me songs and says, don't you love it? I'm like, that is the worst song I've ever heard in my life. Except for the.
Alex Imas:We got it. We got. We got to do the three of us. I love that.
I had a PBS news hour where I was getting interviewed by the host and his college grandson in his dorm room and that was like. That was the best. He had, like a bunch of like, loose trash in the corner.
Adam Davidson:That's awesome. All right, yeah, I'll get action here. All right, so embrace the exponential. How hard is that? Yeah, just go for it.
Alex Imas:Just go for it.
Adam Davidson:Yeah.
Alex Imas:Embrace. Embrace abundance. That's a great phrase.
Adam Davidson:Embrace abundance. Yeah, yeah, there will be abundance. There will be other things too, but there will be abundance. And our hope is there's more abundance. Great.
All right, Alex, thank you so much.
Alex Imas:Okay, thanks, Adam.
Adam Davidson:I hope you enjoyed that conversation with Alex Emis. We covered a lot of ground.
The economics of exponentials, recursive self improvement, hardware constraints, what it would actually take to see exponential GDP growth. But I keep coming back to where Alex ended. Embrace abundance. Most of us are nowhere near using what we already have.
I'd love to hear your thoughts on this one. In the Discord, as always. I'm Adam Davidson and this is the Feed Forward podcast.
