Why History Can't Prepare Us for What's Coming | Alap Shah

24:01Demetri Kofinas:So I certainly can see how this disintermediates the economy more, how it advantages startups in many ways and reduces the headcount for a lot of companies, at least startup companies that need fewer resources to start up, what that means for job losses at corporations and how the politics at these companies maybe create some inertia, I don't know. Okay, so that kind of begins to address the question of why do you think this time is different? In the piece, you also argue that AI is the first technology that targets the non-routine cognitive work that absorbs every prior wave of displaced workers, and that essentially this is going to be a problem for employment.

24:42Demetri Kofinas:So my question is, separating questions of super intelligence and high-level cognitive replacement, why is it a problem if non-routine work goes away? Because the specific framework that we're referring to here is think about a two-by-two matrix of manual labor and cognitive labor, and then think about routine work and non-routine work. The first quadrant being manual routine work that has begun to be automated since the dawn of the industrial revolution, right? The factory line is sort of the best example of that. So those jobs have gone away. That leaves you the other three quadrants. Since the 1980s and sort of the first sort of IT revolution, you started losing a lot of the routine cognitive work.

25:26This is sort of like, you know, things in accounting and things like that, if you think about it. The reason that that automation did not lead to significant job losses is that most of those folks moved into cognitive non-routine work. This is sort of the work that most of our white-collar work is today. And the problem is that that was the escape valve for all of those white-collar workers. And that escape valve, our belief is, will no longer hold in this AI-first world. Today, you're right, the AI cannot sort of take a full task and do it end to end. Some of them are getting better, but they're not quite there yet.

26:04But ultimately, the reason I think that the risk is so large is we have to just look at the trend lines and how these trend lines are progressing, where we have this exponential trend line of AI improvement that has been, in a lot of ways, going for almost a decade. But really, if you think about ChatGBT as the starting gun, has been going for well over three years now. And we're in a world where as those trend lines continue to trend upward, and this is an exponential, so the growths are tremendous. But most importantly, what are we doing as a society and what are we doing as an economy in response to those trend lines?

26:40We're putting more and more dollars and focus into the AI story and especially AI CapEx. By 2028, AI CapEx will be more than 1 % of global GDP. I think that is a floor and it will continue to go up. I mean, just look at what's happened to semiconductor stocks in the last even month. It's very clear to everyone that the ROI on these things, look at what happened to semiconductors and look what's happened to Claude's run rate RAR in the last few months where it's adding the equivalent of entire SaaS complexes in ARR in a single month because it's becoming that powerful and it's becoming that useful.

27:16And so this virtuous cycle has taken off and it's going to a new level. We have most of the world's most intelligent people all focused on the same problem. And so you put all those things together, you have the compute, you have all the algorithmic improvement, you have the capital that is backing it. And you're in a world where we're going to see continuous accelerating AI capabilities even more than we've seen in the last three years. So just consider what that would mean a year from now, two years from now. You put all those together and that's why I've got a reasonable level of confidence that a lot more intelligence is coming from AI and ultimately that's going to mean that there is going to be a lot more risk around replacement of human labor.

27:57Sure.

27:57Demetri Kofinas:So I think maybe the thing I enjoyed the most, and this was probably in the part two of the three-part series, was your observation about trillions of dollars of enterprise value having built on monetizing human limitations, essentially the frictions in our economy. Walk me through this argument and tell me a little bit about how you guys came to it, this observation, this belief about essentially everything in the middle getting wiped out. Yeah, it's really interesting because I've been an investor and a student of the internet for almost 20 years. And if you think about what the internet ultimately does, it helps you find information and then helps you find goods much more easily than you could before.

28:44So that's ultimately just a form of intermediation. And if you really take a step back and think about what our economy is, it's a ton of intermediation. You know, there's lots and lots of search costs and pre-internet, things were really hard, you know, to get information. And now they've gotten easier and easier. And so a huge amount of our economy and especially a huge amount of our market cap is just a tax on, you know, human time, human laziness and human imperfect information. And agents flip all of that on their head. You know, you and I, to save a buck, are not going to run three or four more searches on our phone and spend an extra few minutes.

29:23For an agent, that is a trivial thing to do. You and I are not going to find every listing. We're not going to be able to go in and understand all the pros and cons, read all of the reviews of a given product. Agents flip all of that on their head where they can be infinitely patient and they can find all that information. And if you think about a world where today, the way we work is we open up our phone and we're going to go to our trusted app to go in and find how to get a pizza, how to get a ride, how to book a hotel. But in the world that is coming, in the world that, you know, the reason we came to this thesis is because I'm kind of living in this world already where I'm running agents for all these things.

30:06It just fundamentally changes how you behave. Where your phone, imagine, let's say, it's a year from now and either this is happening on your Android phone or using Gemini or using ChatDBT. and instead of opening up the DoorDash app, opening up the Uber app, opening up the booking app to book something, you just speak into your AI or into your phone, you know, order me a pizza from Joe's, book me a four-star hotel in Boston for me and my family and it does the rest, right? That's the world we're rapidly moving to. In a lot of cases, you know, I'm able to do a few of those things today and that world will be one where all of those intermediation steps that created some of the largest companies in the world and created some of the big what was historically perceived as some of the biggest moats in the world will essentially be completely flipped on its head are all those businesses going to go away of course not you know they're going to adapt and there's going to be new opportunities but there is going to be a huge sort of aggregation of demand into these companies that are driving the intent.

31:09And it will make humans way more productive, we'll be able to get way more done, we'll be way more efficient. But it will also mean that we don't have to pay any of the taxes of an intermediation layer. And so most of the internet is an intermediation layer. The other place that is a huge intermediation layer is financial services. Most of the financial services that we consume, again, we have our trusted partner. We use our banks for all kinds of things. We use a few other insurance brokers and whatnot to buy most of the information, and the financial products that we require. But it's the same thing.

31:38It's because you and I don't want to read the 20 pages of an insurance contract to understand what the pros and cons are. And last year, I had this insurance contract. I probably didn't use it. I'll just renew and I'll go along my merry way. Whereas instead, the agent's going to be able to find, A, it's going to be able to find the best contract, but it also is then going to be able to iteratively negotiate on your behalf to find the one that is willing to give you the best deal, you know, the most product for the least money. And so you put that together across these industries, like around internet and financial services, it's a huge percentage of our market cap and employment.

32:15And I do believe that a lot of that is going to have a huge, huge sort of negative impact from agents, which the really good news there, though, is most of those surpluses are going to flow back to individual consumers. And that will hopefully cushion some of the blow from potential job losses as well.

32:31Demetri Kofinas:Okay. So I think we're certainly capable of revisiting some of the negative knock-on effects. And we haven't even really talked about the doom loop that you guys wrote about, which you touched on early on in the conversation. What are the skill sets that you envision that will be advantaged or are advantaged in this world and that will become more advantaged? The skill sets, and if you want to, you can describe them also as jobs, but I mean, more fundamentally, who are the individuals that are going to be well-positioned of this world? People who are agentic by nature, who are able to use the AIs and ultimately - What does that mean?

33:09Demetri Kofinas:Because I've heard that term used a lot and I don't really understand it. I mean, it sounds like they're saying basically people that are conscious, but even that doesn't quite track with me. People with initiative, first and foremost, I guess what I would say is if you have a vision of a thing you want to do, a thing you want to build, These AIs are so empowering in a way that nothing else has been. The company that you previously could not have started, now you can start with essentially a Claude subscription. Previously, you'd require people to help you build the product and people to help you market it and people to help you serve your customers.

33:44Those people are still required, but you need a lot fewer of them. And especially to get started, it's really powerful to just be able to come in and say, hey, build me this thing. find me these customers, help sell these customers. That is the most promising and exciting part of where we are today with AI, and it's only going to get more powerful. So people who have that initiative, that have that vision, that in the past really just wouldn't have had the capital to go and build these companies. Now the capital requirements are way, way smaller. At the same time, there are more dollars available for these sorts of entrepreneurship.

34:18Demetri Kofinas:Just to make that more concrete and tangible for people, are you essentially saying that people who possess the skill set that's commonly seen in entrepreneurs are going to be advantaged? Or is there something that we can distill even further in this notion of, you know, being ambitious, being driven, being uninhibited, having initiative? People with the entrepreneurial skill set absolutely are going to be hugely empowered. But ultimately, I think this is something that we as society need to embrace and, you know, our education system especially needs to embrace is the skill set of entrepreneurship needs to become much more broadly distributed and taught.

34:53Because that is the skill that I think will become most important in our society is that initiative is the way that you would direct your AI, your set of agents to bring about a new product, new change in the world. And that's not just, in a business sense, it's true for any sort of pursuit where the ability to direct AI in a way where you understand what it's good at, what it's not good at, and how that maps against, you know, whatever the requirements of your customer are, or of, you know, a cause that you're working on, it's sort of the most important skill. And it's not one that's necessarily being taught in school today.

35:30It's not even one that's necessarily being taught in the workplace today, but it is becoming increasingly important. So this idea of sort of AI and agentic literacy is in a lot of ways going to be upstream of just being a productive human in society. Yeah.

35:45Demetri Kofinas:So no, I'm glad you brought up education because I think there are both huge opportunities there and it's also very exciting what can happen. And I also agree, and we're going to have a chance to talk about this, about the need for reform at the state of the social compact, the social contract. But I still don't understand why. Because in a world where individuals who are, let's say, have what we more commonly think of as the entrepreneurial skillset are advantaged, I don't understand why that advantage, which seems to be unique to human beings, based on my extrapolation of your argument around the differentiation of agency, why that doesn't become one of the new pillars around which society is organized, or the gains of the new economy are distributed.

36:30Demetri Kofinas:It becomes a new source of competitive advantage. That's what I'm not understanding. In other words, I feel like there's so much that I don't understand and can't imagine about what the new economy is going to look like. And you're not the only one. This is a common view. So your article hit on a nerve and a lot of people are worried about this. And again, I want to emphasize further that when I have seen some of the things that Claude can do, I've had moments where I look inside and look at my business. I say, am I going to be replaced? When I see Notebook LM having two hosts talking about a topic, I'm like, what is the value that I provide.

37:06Demetri Kofinas:But in each of those situations, I ultimately look inside of myself and I say, okay, I may not know how this thing is going to evolve, but I have to have some faith that I'll be able to navigate it. And I just think we haven't seen anything like this in so long that we don't appreciate how frightening some of the previous waves were, where I would certainly agree is this one does seem to be moving faster. But just why do you feel so confident that these areas that you've already identified as being points of differentiation between human beings and AI are not the new focal point around which the gains of this technology will be organized and distributed?

37:42It's the distribution that's the problem. Because the gains for folks like you and I are going to be very significant. But if you and I operating at the levels that we do are having real existential questions around, what am I going to be doing in a few years' time if this technology continues progressing at a similar rate, then what's going to happen, a great example I'll give you is what's going to happen to the person who just graduated from college at like a second or third tier university with a mediocre GPA within that college, right? We're already seeing a world where historically, young people coming out of college historically had much better unemployment rates than older folks.

38:28And that has very definitively flipped since 2023 and has gotten worse and worse every month. I'm seeing this in my own business. And I, you know, talk to friends who run companies and we see the same thing where historically you would invest in someone early on and they would take five to 10 years to really come up the curve and become, you know, some of your best performers when they got into their thirties. Now that ladder is just being pulled up for folks like that. And, you know, part of that is other parts of our economy having issues, but the ultimate driver of that I think is AI. There's a lot of data to suggest it is AI because, you know, that person in the first few years of their tenure are ultimately a bit of a drag on the organization and you're training them and investing in them for the long-term opportunity.

39:11And in a world where AI can do especially that entry-level work really well, in a lot of cases better at a tiny fraction of the cost, you end up in a situation where, those folks are just not going to be employed the way they have been in the past. And so I think those folks are at the bottom of the distribution in terms of their skill set and their experience set. And so it makes sense that that's where AI would hit first. And we're very clearly seeing that it is biting. And we can see it in our own day-to-day and we can see it in the numbers. And so if those folks are a proxy of AI today and AI's capabilities continue advancing at a significant rate, then why isn't that going to essentially apply to more and more of the population as AI continues to improve.

39:52That's ultimately, I think, what I'm most concerned about and what gives me some confidence that this is a real risk.

39:58Demetri Kofinas:So let's talk about what are some of your proposed interventions, which is what you cover in part three of the global intelligence crisis. What do you think needs to change? What policies need to be instituted on the part of government in order to address the social fallout that will stem from the advancement of these technologies? So the way that I've tried to structure this in part three of my piece is first and foremost, to think about this, not as if you were to rebuild our sort of economic system and taxation system for AI, tabula rasa, there's a lot of aggressive things that you would do.

40:37And I think we will end up in that world and we will be forced to do that at some point. But I wanted to lay out a plan that was much more of the moment that might have an opportunity to be passed in the next few years as society and our political economy wakes up to where things are moving very quickly. And so most of what we've proposed is more of an incremental, let's change our existing rules in this way so that we might get better outcomes. And so the first sort of obvious here is that we tax labor even today more than we tax AI. We have social security and Medicaid taxes on labor that doesn't exist at all when you instead tax corporations.

41:15And so we should fix that. That That should be a pretty easy fix.

41:19Demetri Kofinas:When you say fix that, what do you mean? You mean reduce income tax versus capital gains tax? No, it's just, you know, if an agent does some work for you, you know, it doesn't really, it doesn't get taxed in the way that if a human does work for you, you know, there's a seven or 8 % FICA plus Medicare tax on that. And so the easiest way to do that is like measure an agent's work and put a similar tax on that in the same way that you would an employee. because essentially the way the tax system works today is you have incented corporations to outsource work from humans to any form of automation.

41:53Historically, it was software and other forms of automation, including, you might even say, sending things overseas. And now we're increasingly incenting the use of AI rather than humans to do work. And so that's a very simple fix that we think we should make. And then the other side of that is we haven't really responded to the way the job markets have shifted over the last 20 years, where it's no longer, you know, a world where you come in and you have a job for many decades and you build a career. It's much more, we have portfolios of careers and people are doing different things and side gigs and there's lots of just gig work.

42:30That gig work doesn't come with any sort of benefits. It doesn't come with any sort of social safety net. And so a world where I think more and more of that is coming with AI, we should recognize that and we should build some sort of benefit system that moves with you from job to job. I think these are kind of very obvious, no-nonsense things that we should get done soon.

42:47Demetri Kofinas:And I would argue that that suggestion also, I think, applies even to the pre-AI world. The gig economy, I think, was itself made the case very well for something like that. Yeah. It's been a long time coming. Obviously, there's a bunch of vested interests that are pushing against that. But I do think that the world that we're moving into with AI is going to give that a lot more momentum to actually get something done. That's sort of step one, really obvious, no nonsense things. The thing that I think ultimately matter more is what happens when AI starts accelerating from here. And so the way that we've structured the plan is to focus on what is labor share of income.

43:23And so labor share of income, you know, started out in the 60 plus percent range, you know, in the 80s and 90s, and it's been trending down to around 54 % today. we say, you know, if that breaks definitively to the downside, so close to 50 and certainly below that, then a bunch of sort of new things should be triggered. Because essentially what that means is, you know, AI is continuing to zoom. Corporations are replacing human work increasingly with AI. And in the process, if we don't really focus on it, it's going to lead to a lot of the sort of scary scenarios that we're worried about. And so in that world, what needs to happen is there's a few different ways you could tax this, right?

44:01So Dario and others have talked about a token tax. And, you know, I think there's a world in which that makes sense, especially kind of the end state. But for now, the way we think about it is let's be really direct. Like Americans don't really like income taxes. They don't like increased taxes on the rich and they don't really like wealth taxes. They have a much less of a negative association with corporate taxes. And in this world, corporate taxes are the sort of ground zero of where the impact is coming from, right those corporations are the ones that are you know replacing all those jobs with ai and so we should tax them more heavily we should specifically tax corporations that are doing that and the easiest way to figure that out is the places where margins are just exploding right so if you're a company that is hiring more people your margins are steady your taxes shouldn't change at all but in a world where overall you know jobs are disappearing really fast then let's go directly to the source of where the jobs are disappearing the places where corporations are aggressively cutting humans.

44:56And let's go and tax right there and take those taxes and make some of the humans that have lost their jobs and can't find new jobs because AI is continuing to accelerate, you know, essentially use the earned income tax credit framework to help those folks. And initially, the idea is don't make that a giveaway, but instead make that just like the EITC focus on folks that find new jobs or new jobs are not available, who are looking for new jobs, who are reskilling or doing sort of community service work. So very much conditioned on continuing to work in some sense, but support those folks because you're likely in a world where the jobs are just not out there.

45:36And so that we think about as sort of step one is tax the corporations and use that to make some of the folks that lose their jobs whole. In a world where things get worse than that, we advocate for more of the same sort of taxation and focus that taxation on not just earn income tax credit type job support, but more significant handouts. And then what we call kind of an AI dividend fund, which is like the Alaska permanent fund where you use some of the taxation and some of the spoils from AI to essentially kind of give every American some sort of a stake and something that is an account that actually they can use.

46:14This all will come from a world where GDP is zooming, corporate profits are through the roof, and we'll be able to afford more of those things.

46:24Demetri Kofinas:So what are the investment implications? Obviously, you don't want to be, as we said, in general, you don't want to be a company that has benefited from the friction. But how do you formulate this into an investment thesis? And how have you done that? So the way that we talk about it in the piece is I believe there is a large synthetic short on the consumer economy that is inherent in the AI trade today. So the flip side of semiconductors and anthropic going up every day is that the reason they're going up is because they're substituting for human labor. And so today we're not seeing that yet because we're still in the first inning.

47:05If we're in the third or the fourth inning of that, and we're really starting to see that, you're going to see that show up in consumer spending. You're going to see that show up in jobs being reduced, and especially spending retrenching, even for folks who still have jobs. And so that will be the other side of that barbell where semis keep going up, but the consumer economy, which is the bulk of the US economy, will suffer. And so from an investment implication perspective, before any of the policy gets passed to fix that, the way I think about it is semiconductors and other parts of the AI complex are very clearly winners and will continue to be winners.

47:40And anything that is focused on human wages and labor leading to consumer spending will be on the other side of that. And so most of retail, most of consumer, most of the leisure economy, and all the other sectors that serve those sectors, like trucking and things like that are all going to suffer. And so that is, I think, the clear sort of investment implication of V1. Now what we need to think about are what are the knock-on effects of all this happening. And that's where I'm reasonably positive that as a society, we'll be able to address this problem, figure out the taxation, and in the process, kind of close that synthetic short, which I think will be good for society, will be good for the markets.

48:21Because ultimately, it's unlikely we're going to have a market that keeps going up when semiconductors keep doing really well and most of the consumer economy continues slowing down. That's going to lead to all kinds of weird recession implications as well as ultimately the risk of AI CapEx and AI funding drying up.

48:41Demetri Kofinas:So one more question for you, Alup. You mentioned to me, I think you mentioned to me that you have another piece coming out. Is that right? Do I remember correctly? You're working on a part four? Not necessarily a part four, but a new piece around. Yeah. Can you give us a preview of what your next piece is going to be about? Sure. Thinking through what we've just talked about in terms of where our political economy will go and how we navigate this transition if it comes to pass the way that we expect it might, the piece is really focusing on what we believe to be a new sort of alignment, a political alignment that's likely to arise from AI.

49:15And the idea is, historically, you've had the left and the right, and they both have very different beliefs on both cultural issues as well as economic issues. And so on the economic side, the right was focused on lower taxation, less redistribution, and generally kind of more open laissez-faire economics. And the left is focused on the opposite of that, more redistribution and more regulation. We have this very interesting situation arising right now where if AI continues progressing and jobs do get hit in a way that we expect they will, the cultural alignment will not change. But I do believe that there could be a significant sort of realignment when it comes to economic policies.

50:05In a lot of ways, you can think about folks that are voting for the right who are not wealthy, who are the majority of the electorate on the right today, as folks that are voting primarily for social issues at the expense of economic issues. And that's something that you can afford to do in a world where you have a stable economy, you have a job. In a world where your jobs are gone or are not stable, you have to start focusing on that first and foremost to put money on the table day to day. And so in that world, we expect a real opportunity from a redistribution perspective for both the left and the right to move to the left to essentially say, you know, if you're a politician on the right, you know, we're going to be focused on traditional cultural values.

50:47But we need to react to AI by having a real social safety net, because if not, you know, the wheels are going to fall off and ultimately, you know, you're going to lose the majority of your voting base. And so in the process, I think the left is likely to move significantly more left in terms of redistribution. I also believe that the right is going to move significantly left on those redistribution issues. Certainly not as far left as the left will, but both parties will move significantly to the left because that's the only logical replacement AI. And in a lot of ways, that's the only way that the thing that we're talking about in part three can actually come to pass.

51:24because today's political alignment probably won't get that done. But as jobs continue to erode, it's something where, you know, politicians are going to become much, much more responsive to that. And so the belief there is, you know, I think already AI has gone from issue number 10 or something in terms of voters focused, I think three or four today. And I believe, you know, by the midterms, it's going to be, you know, top two, top three issue. And by 2028, I think it's going to be the only issue that matters because it's moving that fast. It's that clear to everyone what's happening. And it's just a fundamental risk to sort of everything underlying our jobs, our economy, and ultimately kind of security of an individual.

52:08Demetri Kofinas:So I really appreciate you coming on the show, Alup. And I especially appreciated part two of your three-part series. If I didn't already say this, I should have. despite remaining unconvinced about some of your assertions, including the determinism around job losses and the substitutability of human general intelligence for machine intelligence. Because I think engaging in thought experiments like this is invaluable. And the fact that it triggered so many people and generated such a deep emotional response is in and of itself, I think, a very powerful piece of information that speaks to just how significant this issue is and how little air time the topic seems to be getting among the political class.

52:51Demetri Kofinas:In fact, I feel like politicians are treating this technology the same way they've been treating social media, which is to say that they're paying lip service to people's concerns, but doing nothing about them. When the correct response should probably look more like our response to the pandemic or to the 2008 financial crisis, though that would require a degree of preemptive action that I think is rarely seen in politics. If people want to follow your work, Alip, or read any future issues that you publish, how can they do that? Primarily through X. I'm Alip Shah 1 on X, A-L-A-P-S-H-A-H, and then also on Substack, Alip Shah, A-L-A-P-S-H-A-H.

53:31Appreciate you having me. It's been great chatting.

53:33Demetri Kofinas:Absolutely. Thank you so much for coming on the show. Cheers. If you want to listen in on the rest of today's conversation, head over to hiddenforces.io slash subscribe and join our premium feed. If you want to join in on the conversation and become a member of the Hidden Forces Genius community, you can also do that through our subscriber page. Today's episode was produced by me and edited by Stylianos Nicolaou. For more episodes, you can check out our website at hiddenforces.io. You can follow me on Twitter at Kofinas, and you can email me at info at hiddenforces.io. As always, thanks for listening.

54:14Demetri Kofinas:We'll see you next time.