
The Economics of Artificial Intelligence | Tyler Cowen
What this covers
This is a single-speaker talk from Universidad Francisco Marroquín's UFM Studios, in which Tyler Cowen works through what current AI models mean for knowledge work, schooling, health, and the balance between countries. He sets up a split between professions built on ideas, words, and screens (doctor, lawyer, economist) and physical ones like football, where he sees no near-term threat. His evidence is largely personal: he says GPT can pass the bar exam and medical boards, and that in a small test on economics questions it beats him despite his decades in the field. He also cites a jump in measured IQ from about 90 to 135 in a year, and describes models that write their own tests, grade themselves, and iterate.
The middle of the talk turns to access and everyday use. He says free and open-source models, including Chinese ones such as Deepseek, put expert-level diagnosis within reach of anyone online, and that Kenya's connectivity reaches even remote safari areas. He gives practical cases: reading blood tests, diagnosing a pet, finding galleries, and decoding a Paraguayan menu in an indigenous language. He qualifies the access claim, noting that a diagnosis solves an information problem, not a resource problem, because treatment still has to be obtained. He draws on Ludwig von Mises and the Austrian economics tradition to argue that people are inherently entrepreneurial and will adopt these tools without being pushed.
The more contested claims concern institutions and the long run. He says students at his 40,000-student university are ahead of their professors, that blanket bans treating AI as cheating leave students unprepared for workplaces that expect its use, and that the optimal size of many firms will shrink as individuals become more potent. He compares the disruption to the printing press: net positive, but uncomfortable and unevenly shared. He also argues that students should plan for lives of around 100 years, which makes curing Alzheimer's and dementia at least as urgent as curing cancer.
AI systems like GPT will fundamentally transform work, education, and geopolitics within the next 1-2 years, requiring individuals to learn how to collaborate with AI rather than compete against it, while creating unprecedented opportunities for leverage and impact alongside significant societal disruption.
- AI models are improving at accelerating rates (IQ scores jumped from 90 to 135 in one year) and will soon outcompete humans in most knowledge-work fields
- The critical skill is learning to 'train' AI like a dog trainer, not understanding the underlying technology, making expertise accessible to anyone within months of effort
- Smaller countries face geopolitical choices about which AI systems (American vs Chinese) to adopt, creating new forms of dependency that will reshape global power dynamics
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
AI models can self-teach by being programmed to write their own tests, grade their own answers, and iteratively improve without human intervention, enabling learning at speeds no human can match
“One thing they can do is teach themselves. So, you can program it. You can just tell it, 'Well, keep on writing tests, give yourself the test, and grade your own answers.' Or you can have another AI grade the answers. Once it can grade its own answers, it can just keep on getting better. And it can keep on getting better. Not exactly at the speed of silicon. That's misleading. You need to put some work into it. But it basically can learn at a pace no human can learn.”
For professions involving ideas, words, and computer screens—such as doctor, lawyer, economist—AI will change the nature of work in a new and different way within a very short timeframe, whereas for physical professions like football players, AI poses no near-term threat.
“I predict that if you're competing against this, probably you will lose. Not all of you will lose. If you want to be a football player, this is no threat. Not for a long time. You know, you just go ahead and do it. But if you want to be a doctor, a lawyer, an economist, anything where you work with ideas, work with words, work on a computer screen, this will change your life.”
The printing press brought both wonderful and disruptive changes, and AI will similarly bring benefits alongside disruption, meaning that while the changes will be net positive, they will be uncomfortable and anxiety-producing for many people dealing with new choices and uncertainties
“the printing press brings a lot of change, a lot of disruption. These are never good things for everyone. and simply the stress of having to deal with new choices, new paths, new opportunities. Uh, a lot of people just don't like that. Like I said, it's very intimidating.”
In Kenya, despite being a poor country, internet infrastructure is so robust that people can access world-class AI education and services from anywhere in the country, even in remote safari areas, suggesting that lack of wealth or infrastructure is no longer a barrier to AI access
“Kenya is a very poor country. Uh Kenya though it has good internet connections. You can go anywhere in Kenya and like get good internet. Like my wife and I, we went on safari. We were were like out there with the lions, the elephants. We had better internet than at home in our living room.”
AI systems can teach themselves by writing tests, grading their own answers, and iterating at a pace no human can match, enabling them to learn continuously without external instruction.
“One thing they can do is teach themselves. So, you can program it. You can just tell it, 'Well, keep on writing tests, give yourself the test, and grade your own answers.' Or you can have another AI grade the answers. Once it can grade its own answers, it can just keep on getting better. And it can keep on getting better. Not exactly at the speed of silicon. That's misleading. You need to put some work into it. But it basically can learn at a pace no human can learn.”
GPT and other AI models can pass professional licensing exams like the bar exam for lawyers and medical boards for doctors, outperforming the speaker despite his decades of expertise in his field
“They take uh the desktop version of these models and they give them the bar exam for lawyers. And of course, the model passes the bar exam. They give it the medical exam in the United States for doctors. Of course, the model passes the medical exam. I could not pass either of those exams.”
Having access to diagnosis does not automatically mean a person can receive the corresponding medical treatment; the distinction illustrates that AI solves the information problem, not the resource problem.
“It doesn't, you know, having the diagnosis doesn't mean you get the brain operation when you need it, but it's a big big step just to know what's wrong with you or maybe there's nothing wrong with you.”
The printing press brought both good things (spread of knowledge, ability to defend indigenous rights) and bad things (disruption, stress, new difficult choices), and not everyone benefited from it, though it was historically on net positive.
“these are never good things for everyone. and simply the stress of having to deal with new choices, new paths, new opportunities. Uh, a lot of people just don't like that. Like I said, it's very intimidating. And that if you're not a bit scared by the AI, you've been asleep is maybe the most fundamental takeaway.”
At the speaker's large university with 40,000 students, students are ahead of their professors in understanding and using AI, yet faculty members who have established writing and working habits are reluctant to change and the market will eventually force institutional adaptation
“I know my students are ahead of their professors when it comes to AI. So sometimes people sit around and talk like, 'Oh, the professors, should we teach our students about this thing?' And I say, 'No, you have it backwards. They should be teaching you about the thing because the faculty can be set in their ways. They've always written papers a certain kind of paper a certain way and they're not really changing yet. At some point the market will force them to change.”
Mises and Kursner's economic theories predicted that human beings are fundamentally entrepreneurial and want to improve their lives, so they will naturally seek out and adopt AI tools for self-improvement even without external pressure
“As we learned from Misus and Kursner, human beings, they're entrepreneurs and they want to do better in life for themselves, for their families.”
In an artificial test comparing the speaker with AI on economics questions (5-10 questions), the AI wins the competition. The AI does not yet win on every single question, but the pace of improvement is striking.
“If I set up like an artificial test, five questions, 10 questions, ask the AI and then ask me who wins that competition. You're too polite to say, but I think you already know it wins. It does not win on every single question yet. But what's striking to me is how fast these systems are improving.”
The speaker used AI to translate and interpret a Paraguayan restaurant menu written in Gorani (indigenous language) and receive restaurant recommendations, a task requiring both linguistic knowledge and cultural understanding that was previously impossible for non-expert users
“I was at a Paraguayan restaurant. Now I've never been to Paraguay and in fact a lot of the menu it was not even in Spanish. It was in Gorani which is the indigenous or the main indigenous language of Paraguay. So I just took a photo of the menu and I asked my little friend GPT what's this? Please translate and also what which are the best dishes? What should I order? And it told me so this is great.”
AI has proven value in travel (finding galleries and restaurants), home health (pet diagnosis), and personal health (understanding blood tests), suggesting AI's practical utility is already mainstream and expanding rapidly.
“we wanted to go to the best and most interesting art gallery in Atitlan...We just asked this. It told us where to go. We went there. We actually bought two pictures. It was a wonderful experience. We ask it about restaurants. Uh we don't necessarily travel with a guide book anymore. We just travel with this how we take care of the dog, our own health.”
GPT and similar AI models are already superior to human experts in knowledge-intensive domains like medicine, law, and economics, where they pass licensing exams that many humans cannot pass.
“They take uh the desktop version of these models and they give them the bar exam for lawyers. And of course, the model passes the bar exam. They give it the medical exam in the United States for doctors. Of course, the model passes the medical exam. I could not pass either of those exams.”
Universities with blanket policies treating AI use as cheating are not preparing students for their actual future work environment, where employers expect employees to work with AI systems and value the ability to use them effectively
“at my university the general policy is if you use the AI you're cheating. Um I understand why they do that but also in the world to come your work will be done with the AI. So for my class this year and last year, I told students they had to use it and I would grade them on how well they managed to use it to produce a good paper...Their future employer, boss, contractor, like they don't have that same notion of cheating. They just want the job done well.”
Everyone in the world now has access to better medical diagnosis than a wealthy American billionaire had in the past, because free and low-cost AI medical diagnostic systems are widely available, democratizing expert-level medical knowledge globally
“right now everyone in the world has access to better medical diagnosis than like a rich person, you know, a billionaire in America. They may not be using it, but this is not science fiction. This is not a prediction. This is a reality.”
Students today should expect to live to approximately 100 years old (or longer in some cases), meaning their education and skills must serve them for 60+ years of productive life, not the 40-50 years previous generations could expect
“you can expect to live much longer than you thought...My guess is you could expect something close to 100, at least if you stay out of trouble...whatever you learn now as an education, it's just going to be there for more years than you thought. Maybe 15, 20 more years in some of your cases.”
In comparative medical diagnosis tests, AI performed better on average than human doctors, and particularly excelled in bedside manner—being polite, clear, and explaining symptoms in an understandable way.
“Someone actually this has been done a few times. They do a test. So they have human doctors diagnose patients and they have GPT diagnose patients and on average the AI does better than the humans. And it's interesting there's one category in particular where the AI clearly does better and that is bedside manner. Is it nice? Is it polite? Does it talk to you all sweet? Does it explain to you the right way? What's wrong with you? What's right with you? AI does that better.”
The optimal scale of enterprises will likely become smaller for many companies due to AI, and individual capability and impact will be 'much more potent' if individuals choose to leverage AI.
“So the optimal scale of enterprises is likely to be smaller for many companies and what you will be able to do as an individual. It will be much more potent if you choose to be.”
If cure for Alzheimer's and dementia aren't achieved, extended lifespans could become problematic for older people, making the medical agenda of curing neurodegenerative diseases at least as urgent as cancer
“we hope there's also cure for Alzheimer's dementia but we don't know you know the different rates at which these cures arrive it'd be great to cure Alzheimer's first if you cure cancer first and a lot of old people get Alzheimer's that itself is a problem”
Mises and Kuznets teach that humans are inherently entrepreneurial and want to improve their lives and families' circumstances, so humans globally will adopt AI tools despite institutional barriers, enabling education and opportunity access in difficult circumstances.
“As we learned from Misus and Kursner, human beings, they're entrepreneurs and they want to do better in life for themselves, for their families. So we will see people around the world, people otherwise in difficult situations, but they will be getting incredible educations probably for free.”
AI systems have shown intelligence test (IQ) improvements from approximately 90 (average human) to 135 (gifted human range) in roughly one year, demonstrating rapid capability acceleration
“about a year ago, someone tested these systems. They gave them an IQ test. I'm not sure that's a great kind of test, but look, it's what we have, right? They gave it an IQ test and it came in at about 90, which is like would be average for a human. More recently, they gave it an IQ test and it came in at 135.”
Free and open-source AI models (e.g., Deepseek from China) are already available globally with no barriers to access, making professional AI literacy nearly free to acquire for anyone with internet access.
“There's a free version from China. It's called Deepseek. Any of you use Deepseek? Deepseek is free. Just Google Deepseek, one word. It's a lot of fun. You know, GPT, they kind of censor it, so it's not like too rude or too funny. Deepseek is like rude and funny and it knows a lot.”
AI systems have consumed most of the world's written knowledge and significant video content, providing comprehensive access to documented human knowledge.
“it also has not quite all the world's knowledge, but all the world's knowledge that can be written down or put on video. A significant portion of that it has read, listened to, consumed.”
There is not currently a major European AI system that is a serious contender, and European regulations hindering entrepreneurship in tech and AI have put Europe at a competitive disadvantage.
“For Europe, it's a problem. There is not a major European system that is a serious contender. The European Union passes a lot of laws, regulations that hinder entrepreneurship in tech and in AI. So, Western Europe right now is kind of rebelling against the United States. I understand why.”
Over the past three years, the speaker has felt like an idiot because in every domain where he uses GPT, the AI knows more than he does and is smarter than he is, and it improves measurably each week.
“I've spent the last three years feeling like an idiot. I'm here to tell you why I have spent three years feeling like an idiot.”
If you're not scared of AI, you're not paying close enough attention, as it represents a genuine existential challenge to human labor and expertise, even if the ultimate outcome is positive
“I sometimes say if you're not scared of the AI, you're not paying close enough attention because you face this future.”
Most people will choose to be passive consumers of AI benefits rather than competing with it or building with it, enjoying gains without actively engaging with the technology.
“Now a lot of people they don't want to run the whole business. I understand that. The other option is just to sit back, let other people start these businesses for you, and enjoy the gains as a consumer. That's what most people will do. That's fine.”
For professions involving ideas, words, and computer screens (doctors, lawyers, economists, professors), AI will change careers dramatically and quickly—within less than two years—whereas physical professions like football players and auto mechanics will see little disruption from AI
“if you want to be a doctor, a lawyer, an economist, anything where you work with ideas, work with words, work on a computer screen, this will change your life. And it will change your life in a very new and different way...it will be very different less than two years from now...If you want to be a football player, this is no threat. Not for a long time.”
AI legal advice has proven to be comparable to or better than human lawyers, has no financial incentive to overbill clients, and is instantly available at a cost of $20/month versus $200/hour for human lawyers, making it a superior option for many routine legal matters
“anyone who has a legal matter can get free legal advice. When my wife did this, you know, with her retirement, uh, we thought the legal advice, you know, was better than the human lawyers we had worked with in the past. Maybe not perfect, but just really good, free, and it answered all our questions immediately. And unlike some small tiny number of human lawyers, it had no incentive to lie to us. It never said, 'You need more advice. Please come back in two weeks and I'll bill you some additional hours.'”
Learning to work effectively with AI models is similar to training an alien being—you must learn to understand it and teach it, but you don't need to be a computer programmer; rather it requires developing a skill set analogous to dog or horse training
“learning how to work with the AI, it's like being a dog trainer or a horse trainer. It's an alien being. You have to try to understand it. You'll never understand it fully. Mostly, it wants to obey you, but not entirely. You have to learn how to teach the thing.”
Smaller countries like Guatemala, El Salvador, and other developing nations face critical strategic choices about whether to adopt American AI systems or Chinese AI systems, which will determine their geopolitical alignment and economic dependencies for decades
“let's say you're running the education system of Guatemala, the national defense of Guatemala, the treasury of Guatemala, monetary policy of Guatemala through an AI. You can pick the American system of AIS or you can pick the Chinese system of AIS. It's a big big choice.”
Being smart and getting good grades, which historically signaled work ethic and intelligence, now matters 'much, much less' because AI provides these same capabilities to anyone.
“The other option is just to sit back, let other people start these businesses for you, and enjoy the gains as a consumer. That's what most people will do. That's fine. But even in that case, you probably will need some kind of job where you earn income. And the mere fact that you are smart and got good grades, that now matters much, much less than it used to. that used to be often the main thing that mattered like am I smart? Do I work hard? Did I get good grades? Which kind of means you can follow instructions and not be too much you know in your own mind or something that now is greatly devalued.”
The world may return to a global order where smaller countries are in the 'orbits' of larger countries based on AI system adoption, fundamentally changing international relations in a way reminiscent of older imperial patterns.
“the world will return I think to some new order where smaller countries will be in the orbits of larger countries and you will have to choose your larger country.”
Individual responsibility and burden will increase dramatically in the future, as the predictable career paths and life plans that were stable across generations will no longer apply, forcing individuals to make ongoing decisions about direction rather than following established guides
“The life I had where 50 years ago I knew what I wanted to do and 50 years later I'm doing it. Very few of you will have that life. You will be figuring things out as you go. There won't be a guide book you can pick up...the onus of individual responsibility it will be tougher and I would say more intimidating but there will also be much more opportunity.”
AI companies will exercise significant influence over global ideas, words, answers, and mathematical reasoning without formal 'power,' making them de facto publishers and intellectual gatekeepers on a scale comparable to the post-printing-press transformation of society
“those companies, I wouldn't say they will have power. Power is the wrong word, but they will just have a lot of influence over ideas. They'll be like the world's major publishers in a sense of ideas, of words, of answers, of math.”
Writers now need to think about what books they can write that AI cannot write, or where AI can help them, and what their readers will want to read that cannot be written by a computer. This requires rethinking what value writers provide.
“One thing I can do the computer cannot...I'm changing the way I approach many things. One thing I can do the computer cannot. I can go somewhere, show up and talk to people face to face...I think as they told you like I write books but now I need to think like hey I'm not the only person who will be doing this. What kind of books can I write where people still want it?”
AI systems will accelerate progress in science and biology, particularly through modeling cellular mechanisms and testing ideas computationally, potentially curing most diseases that currently kill people like cancer within 40 years
“I know people who work applying AI to biology to biomedical science. Many of them think that in the next 40 years we can use AI say to model the cell to test out ideas in the computer and that medical science will advance much more quickly. And most of the things that kill us now, things like cancer probably will mostly be fixed.”
Smaller countries like Guatemala and El Salvador face a major geopolitical decision about whether to adopt American AI systems or Chinese AI systems, which will affect their dependency and sovereignty.
“for a smaller country such as Guatemala or smaller yet, El Salvador, I think there will be big, big decisions to be made. So, let's say you're running the education system of Guatemala, the national defense of Guatemala, the treasury of Guatemala, monetary policy of Guatemala through an AI. You can pick the American system of AIS or you can pick the Chinese system of AIS. It's a big big choice.”
Learning to work effectively with AI is analogous to dog or horse training: understanding an alien being that mostly wants to obey but requires learning how to teach it, not understanding its underlying mechanics.
“My hypothesis, this has not been tested, but having met many people, I believe this to be true, is that learning how to work with the AI, it's like being a dog trainer or a horse trainer. It's an alien being. You have to try to understand it. You'll never understand it fully. Mostly, it wants to obey you, but not entirely. You have to learn how to teach the thing.”
Capitalism with private AI companies making decisions based on profit motive and customer service is preferable to state-controlled AI systems, and private companies' incentives are 'the best incentives we have in this world.'
“we're in a world we still have a fair degree of capitalism left. The AIS are from private companies. Those are companies that want to make money by serving their customers. Those are about the best incentives we have in this world.”
Historically, the printing press created a discontinuity between the pre-printing world and post-printing world, allowing people to read the Bible, write pamphlets, and share ideas at scale. AI represents a similar or larger discontinuity, operating like 'the printing press on steroids'.
“You can think of AI, if you view it historically, as a bit like the printing press. So the world before the printing press and the world after the printing press, they're very, very different. So with the printing press, people could read the Bible, you could write pamphlets. You know, even the world of Francisco Maroin himself, the idea of speaking up for indigenous rights, the Spanish friars who defended the rights of the indigenous people, they wrote things that other people could then read because there was a printing press. And that idea had a lot of impact. A major change in the world with AI systems. It's like the printing press on steroids. much smarter, very quick, everyone has it. You can basically press a button, just talk into your phone, get any book, essay, piece of information you want.”
AI could become the greatest social mobility engine in history because people who learn to work with AI can achieve wealth and impact far beyond what they imagined, while others who don't adapt will remain passive and potentially fall behind cognitively and economically
“I think this will be the greatest mobility engine probably the world has seen. Many people will become much wealthier than they imagined they could be or have greater impact or leverage what they do. Other people, I think, will decide to be more passive. They may end up knowing less. There's a kind of laziness you can choose because this will answer all your questions.”
Everyone globally with a mobile phone and internet connection can now access world-class AI education in their language of choice, creating unprecedented access to learning regardless of geographic or economic location.
“so if you're, you know, say in Kenya, Kenya is a very poor country. Uh Kenya though it has good internet connections. You can go anywhere in Kenya and like get good internet. Like my wife and I, we went on safari. We were were like out there with the lions, the elephants. We had better internet than at home in our living room. This was to us was amazing. So Kenya has great internet. That was a wise decision. So you can be almost anywhere in Kenya and if you have mobile phone, you can get the world's best education if you want it and in the language of your choice, including Yucataya.”
The speaker's hope is that AI will make society more objective, though this is speculative and not guaranteed, as future AI systems could be intentionally biased and less objective.
“My hope is this makes us all more objective Ive I'm not convinced that's the future. You could imagine further on people build AIs that are very biased and less objective. There's a good chance that happens. Uh the constraint on that is that if you build what I would call a stupider AI, you have an AI that's stupider. So there's actually some market incentive to want the smartest AI because it's better in other ways. And the smartest AI is more likely to be objective.”
At the speaker's university (40,000 students), the students are ahead of their professors when it comes to understanding and using AI. The speaker suggests that rather than professors teaching students about AI, students should be teaching professors, because faculty members are often set in their ways and slow to change.
“I know my students are ahead of their professors when it comes to AI. So sometimes people sit around and talk like, 'Oh, the professors, should we teach our students about this thing?' And I say, 'No, you have it backwards. They should be teaching you about the thing because the faculty can be set in their ways. They've always written papers a certain kind of paper a certain way and they're not really changing yet. At some point the market will force them to change.”
Most people who use GPT on mobile phones have access to the cheaper consumer version, while a more expensive $200/month model exists that is significantly more capable, and even that is not the frontier—companies have even better models that have not yet been released to the public
“like what you have in your phone, this is the cheaper model...there's a more expensive model. It costs $200 a month. It's a lot of money. I pay it...the more expensive model is much, much better...You know, there's a lag between a model being built and a model being released. And now from the major companies, there are better models yet, which to me is like, whoa, better yet than the one that, you know, kicks my butt across the room when I do an economics competition.”
AI systems will advance science and make humans smarter and more productive if we choose to use them that way. The rest of the outcomes are 'up to us'—dependent on human choice, not technological determinism.
“The AI systems will do wonderful things for us, advance science, and if we choose, make us much smarter and much much more productive. The rest of it is up to us.”
AI systems learned to play a game called geoguessing (where a photo is shown and the player must guess the location from visual cues like plants, roofs, sky) without being explicitly taught this skill. The AI is now the greatest geogesser in the world because it has seen so many images and learned this capability emergently.
“There's a game it's called geogging. Do you all know that word geogging? Like someone shows you a photo and from the photo you look at the plants, the roofs, the sky, whatever, and you try to guess where it is. No one taught this thing how to geog. But it turns out it's the greatest geogesser in the world because it's seen so many images. So it just learned this skill on its own.”
The speaker has restructured his own professional work in response to AI competition: he now writes less, travels and gives talks more frequently (since face-to-face interaction is an AI-resistant advantage), reads books differently (using AI to generate customized learning from one best book rather than reading 10-12 books), and focuses on work he can do that AI cannot
“So, I'm changing the way I approach many things. One thing I can do the computer cannot. I can go somewhere, show up and talk to people face to face. So I do that much more. I make many more trips. I give more talks...I write somewhat less. When I read books, I now read books very differently...I will just try to find maybe the best book and I start the best book and then I keep on asking questions of this...So I save a lot of money on books. I get a better book.”
The speaker's daughter faced no meaningful career uncertainty when they discussed occupations 23 years ago, but now faces immense uncertainty due to AI-driven professional disruption.
“My daughter, now she's 35. She and I started talking about careers. She was 12. That's 23 years ago. I told her about different careers, what they were like. Everything I said to her, it's still true. But we're now for the first time in a long time at a point where all that is changing.”
AI systems are relatively objective compared to mainstream media, and this objectivity constraint is market-driven—stupider AI that is more biased will be less useful, creating incentives for smarter, more objective systems
“if you ask the AI about the topics many of you have studied at Marukin you studied right Mises Hayek Rothbard this all this is from Austrian economics who yeah uh it knows all those things you can ask it a detailed question about misus and get a pretty good answer you can't do that with mainstream media so my hope is this makes us all more objective...the constraint on that is that if you build what I would call a stupider AI, you have an AI that's stupider. So there's actually some market incentive to want the smartest AI because it's better in other ways. And the smartest AI is more likely to be objective.”
In group chats, the prevalence of false claims and nonsense has declined in the past 1-2 years because people can now easily fact-check using AI, creating a subtle improvement in group rationality without fundamentally changing human nature
“it used to be people would just say like BS in the group chat...the last year or two, the people I know, they don't say nonsense in the group chat anymore because you can just check it and see that it's nonsense. So it's I wouldn't say we're fundamentally more rational, but in some ways we're a little reigned in and I would say less ridiculous.”
Professional writers must adapt by writing content that AI cannot produce or by focusing on uniquely human value propositions (e.g., personal presence, subjective curation), rather than competing directly with AI on writing task productivity.
“Over time, you know, I'm a writer. I think more and more what are the things I can write that it cannot write or maybe it can help me or what might my readers want to read that I can do and it will not be the same if the computer does it.”
AI is making academic skill assessments more challenging to administer because distinguishing student mastery from AI-generated outputs has become difficult, and traditional integrity-based grading systems no longer map to professional competence.
“So my university the general policy is if you use the AI you're cheating. Um I understand why they do that but also in the world to come your work will be done with the AI.”
Asking AI which AI model is best is an unreliable process because models will either claim superiority or refuse to answer, making model comparison require external evidence (user testing, benchmarks).
“That's one question. You can't ask any model. What's the best model? They'll either say they're the best model or they'll just refuse to answer.”
The speaker and his co-blogger Alex Tabarok have fundamentally changed how they conduct intellectual arguments—instead of yelling at each other, they now send links to different AI models and debate which model is correct, using AI as evidence in philosophical disputes
“Alex and I get in arguments he's been whitilled on AI but our arguments are totally different now we used to like yell at each other now we just send each other links back and forth from the AI. The AI is on my side. It's like, no, you wrote the prompt wrong. It's on my side. That's our argument now.”
The speaker's personal approach to staying ahead of AI disruption includes: face-to-face travel and talks (where AI cannot compete), reading only the best book on a topic and asking AI targeted questions, paying $200/month for premium AI, and following the latest models.
“One thing I can do the computer cannot. I can go somewhere, show up and talk to people face to face. So I do that much more. I make many more trips. I give more talks. It cannot compete with me. I write somewhat less. When I read books, I now read books very differently. It used to be if I wanted to learn a topic, maybe I would read 10 books, 12 books on the topic. Now I will just try to find maybe the best book and I start the best book and then I keep on asking questions of this.”
American AI companies will function as 'the world's major publishers in a sense of ideas, of words, of answers, of math,' exerting influence over global information and thought through their AI systems.
“those companies, I wouldn't say they will have power. Power is the wrong word, but they will just have a lot of influence over ideas. They'll be like the world's major publishers in a sense of ideas, of words, of answers, of math.”
AI provides better legal advice than human lawyers in some cases, without the perverse incentives human lawyers have to bill additional hours, making AI more market-aligned than human professionals.
“When my wife did this, you know, with her retirement, uh, we thought the legal advice, you know, was better than the human lawyers we had worked with in the past. Maybe not perfect, but just really good, free, and it answered all our questions immediately. And unlike some small tiny number of human lawyers, it had no incentive to lie to us. It never said, 'You need more advice. Please come back in two weeks and I'll bill you some additional hours.' Right? Very different incentives. I would say more market friendly incentives.”
Everyone in the world now has access to better medical diagnosis than a billionaire in America had in the past, regardless of whether they're using it
“Right now everyone in the world has access to better medical diagnosis than like a rich person, you know, a billionaire in America. They may not be using it, but this is not science fiction. This is not a prediction. This is a reality.”
African nations have genuine geopolitical leverage in AI infrastructure choice because their alignment is uncertain, giving them better negotiating position than geographically proximate nations
“Kenya might get better terms because no one really knows which way Kenya will go. But China might just think like, h Guatemala, they're going to go with us. We're not going to really bother making them a competitive offer.”
It is now possible to run a research center or think tank with just 2-3 people and have it achieve the scale, output, and impact of a much larger traditional research institution, if done well.
“you will be able to run a research center, a think tank, I don't know, maybe not with one person, but I believe with two or three people and it will have the heft and the output and the impact of a much larger research center at least if you do it well.”
The choice between working with AI versus competing against it is not one the speaker's generation faced, and people must decide which path to take, representing a fundamental difference from his career trajectory.
“There's also the path, do I learn how to work with this or do I compete against it? And those are decisions you will have to make. They're decisions I never faced. There are decisions also uh my daughter never faced. She'll face them now as I face them now, but I'm obviously older than she is.”
Individual responsibility will be tougher and more intimidating in the future because of increased unpredictability and decision-making burden, but there will also be much more opportunity for individuals who develop the right skills and mindset.
“But you'll have to make those decisions. So the onus of individual responsibility it will be tougher and I would say more intimidating but there will also be much more opportunity.”
GPT can identify Mayan languages, knows there are over 30 distinct Mayan languages in Mesoamerica, and can translate text into specific variants like Yucatec Maya, demonstrating multilingual capability beyond common languages.
“I said, "Uh, hello. Do you speak Mayan at all? Can you say that sentence back to me in the language Mayan?" Of course, it speaks Spanish, right? But does it speak Mayan? It's also a smart guy. It says, "Hello, Mayan isn't a single language, but a family of over 30 distinct languages spoken in Meso America." It then lists different kinds of Mayan...And then it says, well, okay, you know, it basically calls me an idiot in a polite way. And it says I will answer your question in Yucatech Maya. And then it proceeds to do so.”
The optimal scale of enterprises will likely shift toward smaller organizations, and individuals with AI skills will be able to run research centers, think tanks, and other institutions with 2-3 people that have the impact and output of much larger traditional organizations
“you will be able to run a research center, a think tank, I don't know, maybe not with one person, but I believe with two or three people and it will have the heft and the output and the impact of a much larger research center at least if you do it well. So the optimal scale of enterprises is likely to be smaller for many companies and what you will be able to do as an individual. It will be much more potent if you choose to be.”
According to Sam Altman (OpenAI CEO), billion-dollar revenue companies will soon be operated by a single person plus AI systems, and the speaker agrees with this prediction based on his knowledge of current capabilities
“Sam Alman, he's the head, the CEO of OpenAI. He predicts that very soon there will be companies with sales of a billion dollars or more and the whole company will be run by one person and a bunch of AIs. We're not quite at that point yet, but I and most people I know, we agree with Sam.”
AI models like GPT are relatively objective compared to mainstream media outlets, and people can ask AI detailed questions about Austrian economics topics (Mises, Hayek, Rothbard) and get good answers, which mainstream media cannot provide.
“So, if I think, should I ask the AI or should I look at the New York Times, to me, the AI is pretty objective. It's also the case if you ask the AI about the topics many of you have studied at Marukin you studied right Mises Hayek Rothbard this all this is from Austrian economics who yeah uh it knows all those things you can ask it a detailed question about misus and get a pretty good answer you can't do that with mainstream media”
Universities currently enforce policies treating AI use as cheating, but this is misaligned with market reality: employers care about work quality, not about whether AI was used, making current educational policies obsolete.
“So my university the general policy is if you use the AI you're cheating. Um I understand why they do that but also in the world to come your work will be done with the AI. So for my class this year and last year, I told students they had to use it and I would grade them on how well they managed to use it to produce a good paper. This is PhD level. I just thought that's the skill they need. Their future employer, boss, contractor, like they don't have that same notion of cheating. They just want the job done well.”
Using AI in group chats has made people 'reigned in' and 'less ridiculous' because they can fact-check claims in real-time, making communication modestly more rational.
“And it used to be people would just say like BS in the group chat. Is that like an acceptable word here? I won't say the word, you know, nonsense in the group chat. Like the last year or two, the people I know, they don't say nonsense in the group chat anymore because you can just check it and see that it's nonsense. So it's I wouldn't say we're fundamentally more rational, but in some ways we're a little reigned in and I would say less ridiculous.”
The speaker's core professional practices (writing, teaching, giving talks, grading) have remained fundamentally unchanged for his 50-year career, with only surface-level technology shifts (typewriter to computer, added email), representing stable career trajectory impossible for future generations.
“What I thought that life would be like when I was 13 or 14 and what that life is like now, it's kind of the same. It's not exactly the same, but I write things. Instead of a typewriter, I have a computer. I teach. I give talks. What am I doing now? I'm giving a talk. What did I used to do? I would give a talk. Go different places, grade papers. Really has not changed.”
The speaker follows the latest AI models and exchanges tips with friends who use AI, indicating that staying competitive requires active, continuous monitoring and peer learning about frontier AI capabilities.
“What I do now, I follow the latest models. I exchange tips and advice with my friends who use this. It's changed how I write, what I write. It's changed how I read. It led me to make more trips when I travel.”
Older people have the option to retire rather than adapt to AI, but younger people must choose between learning AI or being displaced, creating different opportunity costs by age cohort.
“of course, a lot of people who are older, they're just going to stop and that's fine.”
The historical period of stable, predictable career paths that characterized the 20th century is ending, and younger generations will face radical uncertainty and constant career change
“This notion that we have these unbroken, highly predictable lives with your generation... that is coming to an end.”
Some people will choose a 'lazy path' where they let AI answer all their questions and become more passive, potentially knowing less than they could, as an alternative to the ambitious path of maximizing leverage with AI.
“Other people, I think, will decide to be more passive. They may end up knowing less. There's a kind of laziness you can choose because this will answer all your questions. So, there will be these two paths. The greater impact path, the lazy path.”
The speaker has been an economist for 50 years (since age 13-14), and the fundamental nature of his work—writing, teaching, giving talks, grading papers—has remained remarkably unchanged, but this period of stability is coming to an end within the next two years.
“I'm 63 years old and I started wanting to be an economist when I was 13 or 14. So I've been working on being an economist for 50 years...What I thought that life would be like when I was 13 or 14 and what that life is like now, it's kind of the same.”
American AI systems are built by private companies seeking profit by serving customers, which the speaker characterizes as 'about the best incentives we have in this world' for ensuring beneficial development.
“But, you know, we're in a world we still have a fair degree of capitalism left. The AIS are from private companies. Those are companies that want to make money by serving their customers. Those are about the best incentives we have in this world. The AI systems will do wonderful things for us, advance science, and if we choose, make us much smarter and much much more productive. The rest of it is up to us.”
Being intelligent and getting good grades, which historically were primary determinants of success, have become much less valuable as predictors of future earnings and impact, as the ability to work effectively with AI ('dog training') has risen sharply in value
“the mere fact that you are smart and got good grades, that now matters much, much less than it used to...that used to be often the main thing that mattered like am I smart? Do I work hard? Did I get good grades?...that now is greatly devalued. What has risen in value is the question like are you a good dog trainer, horse trainer but with this”
Europe currently lacks a major competitive AI system, and EU regulations that hinder tech entrepreneurship mean European nations will likely adopt American AI systems rather than developing indigenous European AI, further concentrating power in American AI companies
“For Europe, it's a problem. There is not a major European system that is a serious contender. The European Union passes a lot of laws, regulations that hinder entrepreneurship in tech and in AI...So, Western Europe right now is kind of rebelling against the United States...but that too will come to an end. So you will have these AI systems in Western Europe, probably not from China, probably from the American companies”
If you're not somewhat scared by AI, you're not paying close enough attention to how transformative it will be, making some level of fear a rational response.
“And that if you're not a bit scared by the AI, you've been asleep is maybe the most fundamental takeaway.”
No one is currently an expert in AI use; everyone is at the frontier of learning, making it possible for students today to catch up to and potentially exceed the knowledge of older adopters within 2-3 months.
“The first is no one is really an expert. So you might think, 'Oh, this is intimidating. This is terrible. There are all these geniuses out there who know this and I don't.' But this is all so new. So you might today know zero. I doubt if that's the case, by the way. But even if you know zero today, you're not really very much behind.”
Someone with no prior AI expertise can catch up to the frontier of AI knowledge and practical skill within two or three months of dedicated application.
“I don't know. I would guess if you apply yourself with two or three months work and you're just there doing what the top people are doing in terms of using this to use this.”
Many people are already using AI systems as therapists, paying $20/month for GPT-based therapy as an alternative to human therapists who charge $200/hour.
“I know many people, I mean really many people, they use this as their therapist. Uh I've never done that. They keep on doing it. In the United States, a therapist might cost $200 an hour. It's a lot of money, right? $200 an hour. Seems crazy to me. like go to a friend, go to your priest, go go go somewhere for free. But no, people pay. Uh but a lot of people have stopped paying and they do this. This is always available 247 in the United States. Uh this costs $20 a month.”
Sam Altman, CEO of OpenAI, predicts that very soon there will be companies with billion-dollar sales revenue run entirely by one person and multiple AIs, a prediction the speaker and most people he knows agree with.
“Sam Alman, he's the head, the CEO of OpenAI. He predicts that very soon there will be companies with sales of a billion dollars or more and the whole company will be run by one person and a bunch of AIs. We're not quite at that point yet, but I and most people I know, we agree with Sam.”
The speaker's personal use of AI includes consulting it frequently for economic policy questions, and he discusses AI with his co-blogger Alex Tabarok in ways where they now ask which AI model agrees with them rather than making unsupported arguments.
“I have a co-blogger Alex Tabarok I wrote a textbook with him he and I do those videos together Alex and I get in arguments he's been whitilled on AI but our arguments are totally different now we used to like yell at each other now we just send each other links back and forth from the AI. The AI is on my side. It's like, no, you wrote the prompt wrong. It's on my side. That's our argument. Now,”
AI will accelerate progress in biomedical science such that within the next 40 years most major fatal diseases like cancer will become controllable or curable, shifting the primary cause of death to old age.
“I know people who work applying AI to biology to biomedical science. Many of them think that in the next 40 years we can use AI say to model the cell to test out ideas in the computer and that medical science will advance much more quickly. And most of the things that kill us now, things like cancer probably will mostly be fixed. You might still get cancer, but it will be controllable or maybe curable.”
The speaker acknowledges that he spent three years 'feeling like an idiot' because in every domain where he uses GPT, the AI knows more and is smarter than he is, with monthly or weekly improvements making the gap larger
“I've spent the last three years feeling like an idiot...Any area where I use this, it knows more than I do and it's smarter than I am. It's not smarter in every way, but each month, sometimes even each week, it gets a bit smarter yet.”
The speaker predicts that life expectancy for his young audience will increase to approximately 100 years or more, conditional on avoiding dangerous accidents and criminal circumstances.
“if you're a student here, you can expect to live much longer than you thought. I I don't know. Life expectancy in Guatemala. In the US, it's in the low 80s. 76. 76. Okay, that's like pretty good. My guess is you could expect something close to 100, at least if you stay out of trouble.”
Central planning typically fails because no individual can predict markets in advance; the diversity of outcomes that AI creates makes top-down policy coordination even more difficult than in pre-AI eras.
“Part of the wonders of the market is uh no one can predict it in advance. That's one reason why central planning typically does not work.”
There is a more expensive GPT model available at $200/month that is 'much, much better' than the free/cheap version, with fewer errors, faster responses, and greater knowledge, though even this premium model is not the frontier of what has been built internally.
“And the more expensive model is much, much better. And it makes fewer mistakes. It's quicker. It knows more. It's more brilliant. So, whatever impression you have of how good this is, like there's already something much better. How many of you pay $200 a month? Show of hands. Zero. That's I'm not saying you should right now, but you know, maybe your employer will or maybe you will have to or maybe the price will come down. But even the $200 a month model is not the frontier. You know, there's a lag between a model being built and a model being released. And now from the major companies, there are better models yet, which to me is like, whoa, better yet than the one that, you know, kicks my butt across the room when I do an economics competition.”
AI models improved from an IQ score of approximately 90 to 135 in a single year, demonstrating rapid and ongoing advancement in capabilities.
“So about a year ago, someone tested these systems. They gave them an IQ test. I'm not sure that's a great kind of test, but look, it's what we have, right? They gave it an IQ test and it came in at about 90, which is like would be average for a human. More recently, they gave it an IQ test and it came in at 135. It's basically a year later.”
DeepSeek is a free Chinese AI model that is functionally similar to GPT but less censored, making it more 'rude and funny' while retaining significant knowledge.
“There's a free version from China. It's called Deepseek. Any of you use Deepseek? Deepseek is free. Just Google Deepseek, one word. It's a lot of fun. You know, GPT, they kind of censor it, so it's not like too rude or too funny. Deepseek is like rude and funny and it knows a lot.”
GPT can identify plants, birds, and other objects from photographs taken by the user, and can translate foreign language menus (including indigenous languages like Gorani) and recommend dishes from restaurants.
“I can take a photograph of a flower, a plant, a bird, whatever. And I ask my little friend here, what's that? And it tells me.”
Unsolved problems in mathematics may be solved by AI systems that can approach them computationally, expanding what humans can discover beyond what human mathematical intuition can access
“a lot of the unsolved theorems in mathematics we might solve”
The speaker is older (63) and could stop working, but he chooses to continue and stay ahead of AI trends because he is excited about the opportunity and wants greater impact.
“Like I could just stop now and call it quits. I'm not going to do that. Like I'm very excited about all this. I want to be ahead of these trends and have more impact myself. But of course, a lot of people who are older, they're just going to stop and that's fine.”
In debates about AI, you cannot ask the AI itself which model is best, as it will either claim it is the best or refuse to answer, so model preference becomes a new ground for debate.
“That's one question. You can't ask any model. What's the best model? They'll either say they're the best model or they'll just refuse to answer.”
The speaker and his wife use AI to diagnose their dog's health problems instead of visiting a veterinarian, because AI can provide quick, accurate diagnosis advice at near-zero cost while reducing unnecessary vet visits
“My wife and I, we take care of a dog. The dog's fine. If there's something wrong with the dog, we take a picture of the dog, ask the GPT, so we don't have to go to the veterinarian because the GPT tells us what's up with the dog.”
The speaker and his wife use GPT to diagnose their dog's health issues instead of visiting a veterinarian, because the AI system is sufficient to identify the problem.
“My wife and I, we take care of a dog. The dog's fine. If there's something wrong with the dog, we take a picture of the dog, ask the GPT, so we don't have to go to the veterinarian because the GPT tells us what's up with the dog.”
The speaker and his wife use AI to identify the best and most interesting art gallery in Atitlan while traveling, and subsequently made art purchases, demonstrating successful practical use of AI for subjective consumer decisions.
“We wanted to go to the best and most interesting art gallery in Atitlan. So, the old days we would Google that's not like a very transparent piece of information like what's the best, what's the most interesting, it's sort of vague, subjective. We just asked this. It told us where to go. We went there. We actually bought two pictures. It was a wonderful experience.”
The speaker personally would pay 10 times the current $200/month price for the more expensive GPT model if necessary, indicating extremely high personal valuation of the tool
“It costs $200 a month. It's a lot of money. I pay it. I'd actually pay 10 times that if I had to.”
Central banks and economic policymakers are already consulting AI frequently to inform decisions, indicating adoption of AI in government policymaking.
“many you know central banks in economic policym people already as do I I consult the AI frequently”
The speaker's wife has used AI to help her retire and manage the terms of her retirement contract, indicating that spouses and family members are also adopting AI for major life decisions.
“We were going to call a lawyer. Uh my wife retired recently, had to sign a contract for the terms of her retirement. It's a very complicated contract. My wife is a lawyer. There were some things she wasn't sure about. Who do we ask?”