Intro. [Recording date: February 24, 2026.]
Russ Roberts: Today is February 24th, 2026. Before I introduce our guest, I’d like to share the results from our recent listener survey regarding your favorite episodes from 2025. Here are the top 10.
In tenth place, we have a tie between “The Economics of Tariffs and Trade” featuring Doug Irwin, and “Why Christianity Needs to Help Save Democracy” with Jonathan Rauch.
Ninth is “How to Walk the World” with Chris Arnade.
Eighth is “The Music and Magic of John and Paul” with Ian Leslie.
Seventh is “Will Guidara on Unreasonable Hospitality.”
Sixth place goes to “EconTalk episode #1000,” a solo episode with me.
Fifth is “The Magic of Tokyo” with Joe McReynolds.
Fourth is “The Perfect Tuba” with Sam Quinones.
Third place is “Shampoo, Property Rights, and Civilization” with Anthony Gill.
Your second-most liked episode was:
2. “A Mind-Blowing Way of Looking at Math” with David Bessis.
And, your favorite episode, chosen by 33% of listeners:
1. “What Is Capitalism?” with Mike Munger.
Thank you all for participating in the survey and for your wonderful comments, which I truly enjoy reading.
Now, on to today’s guest, Tyler Cowen from George Mason University, known for his work on Marginal Revolution and “Conversations with Tyler.” This marks Tyler’s 20th time on the show. He last joined us in November 2024 to discuss the significant novel “Life and Fate” by Vasily Grossman, which many of you found enlightening. I appreciate your feedback.
Welcome back to EconTalk, Tyler.
Tyler Cowen: Always a pleasure to be here, Russ.
Russ Roberts: Today’s discussion stems from a recent talk you gave at the University of Austin, which we’ll link to. We’ll also provide a link to the top 10 episodes for those interested in revisiting them. Your talk covered how AI might transform higher education, among other topics.
Before we delve into that, let’s first discuss your thoughts on the disruptive potential of AI in the job market. Many commentators suggest we are on the brink of a major upheaval, with fears of job losses as AI outperforms humans in most tasks—except perhaps offering a comforting smile. Is that our future? Are we headed toward mass unemployment and poverty? There’s been a wave of pessimism on social media regarding this lately. What’s your take, Tyler? Do you share these concerns?
Tyler Cowen: While it’s true that AI will eliminate many jobs, it will also create a multitude of new ones. Take the energy sector, for instance; as AI advances, the demand for energy will soar, necessitating a larger workforce. The same goes for biomedical trials, which will see a surge in innovative drug and device development needing rigorous testing and regulatory navigation.
Interestingly, I believe AI will lead to an increase in the demand for lawyers, which may not be a welcome outcome. We will need to draft a plethora of new laws governing AI, and while I suspect the AIs could draft better legislation than humans, it’s unlikely we’ll allow them that privilege. So, expect a growth in legal positions, particularly in government. Those are just a few examples.
However, predicting the emergence of future jobs can be quite challenging. If we think back to the early days of the Industrial Revolution, would anyone have predicted a future filled with podcasters? Probably not. We’re in a similar situation today.
On the plus side, I believe we’ll enjoy increased leisure time as AI takes over routine tasks more efficiently. This trend is already apparent; many tasks are completed more swiftly, even if management remains blissfully unaware of this productivity boost.
Russ Roberts: You mentioned that the legislation crafted by AI might surpass human-generated laws in quality. Sam Altman once suggested that governance would need to adapt in a world increasingly influenced by AI. I have my reservations about AI’s aptitude for governance, particularly regarding the trade-offs and nuanced decisions that are critical for human society. Governance isn’t merely about optimization; it’s inherently about balancing competing interests. Do you agree with that perspective, or do you see a role for AI in shaping regulatory frameworks in an AI-driven economy?
Tyler Cowen: Personally, I’d prefer AI models like Claude or GPT over most current political leaders or regulatory officials. But the reality is that AI will be used as an aid rather than a decision-maker. The real challenge lies in whether humans will heed AI’s recommendations. While I believe AI will provide superior governance solutions, the effectiveness of its implementation is contingent on human cooperation. In the short term, we may even witness a decline in governance quality due to an influx of AI-generated, albeit irrelevant, comments overwhelming public discourse. We’re already seeing this trend emerge.
Russ Roberts: Are you at all pessimistic about the economic implications of AI? The doomsayers argue that aggregate demand could collapse as half the population becomes unemployed, leading to a scenario where the benefits of lower prices are inaccessible to a majority. What’s your view on that?
Tyler Cowen: That encompasses a range of issues. When you asked if I’m pessimistic at all, those words carry weight. My primary concern is that AI will alter governance in unpredictable ways. Our political structures often fare worse than our economic models, regardless of one’s viewpoint. If governance deteriorates, the economy suffers, and I don’t rule out that possibility. It’s a significant concern.
On a more conventional economic front, I anticipate an overall increase in wealth. We won’t necessarily see fewer jobs; in fact, many at the lower end of the income spectrum will gain access to a range of services at little to no cost. However, I do foresee a rise in the number of billionaires and mega-billionaires, as fewer individuals will build increasingly large and profitable companies. These new enterprises will generate numerous job opportunities, and I don’t foresee a future of mass unemployment. However, I certainly have my worries.
Russ Roberts: You mentioned that your primary concern is governance, particularly its potential impact on the economy. Is that your foremost worry in economics or your biggest concern overall?
Tyler Cowen: It’s my biggest worry, irrespective of AI. If politics deteriorates, economic outcomes will follow suit. Numerous negative influences are already at play, most of which aren’t directly related to AI. Yet, integrating AI could exacerbate existing issues, particularly if those in power fail to regulate its use effectively. The potential for political decline is a troubling scenario.
When it comes to economics, we have models like the price system, Hayek’s ideas, and Say’s Law, which can often guide our understanding of economic trajectories. In contrast, our political models leave much to be desired. Take the median voter theorem, for example—valuable as it may be, we still have no clear understanding of what the median voter desires regarding AI.
Russ Roberts: I share some concerns regarding a potential future filled with leisure time resulting from a dramatic reduction in the need for work. Keynes envisioned a world where increased wealth would lead to greater leisure, and while he was right about the wealth part, he was mistaken about the leisure part. People still work hard, even if they might enjoy more leisure over their lifetimes.
My worry is that if AI displaces numerous skills too rapidly, the transition could be troubling. Consider the implications of driverless cars for millions of cab and truck drivers. If that shift occurs too quickly, the political ramifications could be quite severe. What are your thoughts?
Tyler Cowen: For jobs like trucking, the role encompasses much more than driving; it involves loading, unloading, and managing various points of contact. I suspect that these transitions will be gradual rather than abrupt.
As for when companies like Tesla will effectively displace human drivers, I expect it within the next decade. However, it won’t happen overnight. Many consumers will still prefer human taxi drivers, especially given the costs associated with automated services. The equilibrium in this market remains uncertain. It’s essential to recognize that many systems may end up with higher marginal costs than anticipated as they scale to handle a variety of situations.
Work has evolved throughout human history, and while it may change more rapidly now, I don’t harbor deep fears about increased leisure time. Some individuals may struggle with that adjustment, as we saw during the COVID pandemic. Yet, for those eager to work, the opportunities to manage and execute projects will be greater than ever before, regardless of whether they generate income.
Russ Roberts: Have you had the chance to experience a driverless car?
Tyler Cowen: Yes, it’s quite enjoyable.
Russ Roberts: I think it’s remarkable. I would opt for a driverless car over a human driver any day in the current landscape.
Tyler Cowen: Keep in mind that you aren’t paying the full price for a Waymo ride.
Russ Roberts: That’s a fair point.
Tyler Cowen: People believe the Tesla network will eventually prove superior due to data accumulation, subsequently lowering marginal costs. Yet, as road conditions and regulations evolve, it’s challenging to predict public expectations for these changes. Perhaps we will find that human drivers can offer additional services, such as package delivery. However, we shouldn’t be too confident in our predictions regarding the future of these jobs. If one particular job does vanish, I believe society will adapt.
Russ Roberts: Cheaper alternatives will likely arise in many aspects of our lives, should that scenario play out.
Russ Roberts: Shifting gears, let’s discuss higher education, an area ripe for AI disruption. However, as you’ve noted, there’s substantial inertia within the higher education system. It’s not the most agile institution in America.
Tyler Cowen: “Inertia” is a polite way to put it, yes.
Russ Roberts: It’s not clear that higher education can be revolutionized overnight, if at all. Students pursue college for various reasons beyond mere education. You proposed a rather startling notion: that a third of college courses should focus on effectively utilizing AI. Could you elaborate on that idea?
Tyler Cowen: This could apply to a third of total course time. Many future jobs will require proficiency in AI, yet most institutions do not currently teach this skill formally. Some individual professors may incorporate it, as I do.
I propose that we allocate a significant portion of the curriculum to this essential skill, which is currently lacking. It’s important to note that teaching students how to utilize AI effectively doesn’t preclude teaching them other subjects. For instance, one could teach students how to leverage AI to better comprehend Homer’s Odyssey while simultaneously imparting knowledge about the text itself. Therefore, dedicating a third of the curriculum to AI doesn’t necessarily displace other subjects; it can enhance them.
The challenge lies in the fact that many faculty members are not equipped to teach these skills. And administrators are likely even less prepared. So, who will take on this teaching role? Perhaps students could even teach professors, given that they’ve been using AI to cheat!
Russ Roberts: When you mention teaching people how to use AI, do you mean instructing them on writing better prompts? What specifically do you envision?
Tyler Cowen: It varies by field. In programming, for example, you might teach students to use Codex for better coding. In the humanities, it might focus on crafting effective prompts. When querying AI about Homer’s Odyssey, how does one ensure they receive the most insightful answers? Which advanced model should they consult? These are the types of questions that will need to be addressed. Over time, I expect that effective prompting will become second nature.
Russ Roberts: At this stage, it seems the biggest challenge is that many people don’t even think to use AI, right?
Tyler Cowen: Absolutely. The first step is simply encouraging people to use it. For example, in a biology class, integrating AI systems into lab work would be critical. I fully acknowledge that we currently lack qualified instructors, but why shouldn’t we aim to provide students with the knowledge they need?
Russ Roberts: Let’s discuss writing, as I find it a crucial issue. You mentioned how we should adapt writing education in light of AI’s presence, particularly in terms of preventing over-reliance on AI for grades. Could you share your insights on that?
Tyler Cowen: The problem of cheating with AI is vastly overstated; we simply lack the will to tackle it effectively. It’s not that detecting AI-generated work is impossible now or in the future, but we need to be proactive. For 2-3% of their written output throughout their college experience, we could lock students in a room for a test. If their submissions diverge significantly from their test performance, we could have a conversation without resorting to expulsion. This approach requires a certain level of commitment to addressing performance discrepancies, which is sorely lacking in current systems.
Russ Roberts: To clarify, you’re suggesting that students write an essay without access to AI or the Internet, and then compare it to an essay completed with unrestricted access. If there’s a significant difference in quality, that could indicate AI usage, correct?
Tyler Cowen: Exactly. It’s a sampling issue. If discrepancies arise, we could conduct additional in-person writing assessments to verify their capabilities. There’s no need for expulsion or a permanent record that could hinder future job prospects. Many students simply experience anxiety in confined settings, but we would likely see a reduction in cheating under such a system.
Russ Roberts: I want to challenge the definition of cheating a bit. In the same talk, you expressed that you don’t use AI for your columns, which I understand. However, many people use it to varying degrees. Some might completely rely on AI to generate an essay, while others might use it for more subtle assistance, like reorganizing content or suggesting edits. Are you advocating for a complete ban on AI in writing courses?
Tyler Cowen: I believe we should separate tasks. A significant portion of writing classes should involve students using AI. In my current History of Economic Thought class, I require students to use AI in their writing. The standard for good papers will be elevated, and they should learn to integrate AI effectively. Ultimately, we should assess the joint product of their efforts.
Moreover, we should teach them to write independently primarily to enhance their critical thinking skills. While many may not need to become proficient writers, they will certainly benefit from improved thinking abilities, which writing cultivates.
Russ Roberts: It’s interesting how intertwined writing and thinking are, isn’t it? People often view them as separate skills, but my ability to think is significantly influenced by my writing skills, and I find it hard to disentangle the two.
Tyler Cowen: I can’t think in the shower! I hear stories of ideas popping up during showers, but for me, it’s just about getting clean. Writing or conversing is essential for my thought process.
Russ Roberts: We had a conversation with Lorne Buchman on this topic, and we’ll link to it. I agree; it likely varies by individual.
Russ Roberts: Returning to your experience at Shalem College, students engage in intimate, small-group discussions over significant texts. They wrestle with the meaning and implications of these texts, which is an extraordinary experience. I can’t speak for you, but I found that kind of engagement lacking in my undergraduate education. The closest parallel was during my graduate studies when small groups tackled challenging problem sets without clear answers. Those sessions were immensely impactful. AI cannot replicate that experience, can it? Or do you disagree?
Tyler Cowen: You and I both admire Adam Smith, right?
Russ Roberts: Absolutely.
Tyler Cowen: Smith proposed that different classes and professors should compete with one another. In my talk at the University of Austin—I want to clarify that it’s not the same as UT Austin—I suggested they offer just 30 classes in a semester, which is quite limited.
I proposed they allow students to take one class with AI, perhaps once every two years, just to see how it goes. Students could form small groups to explore topics, like Tudor England, which isn’t typically covered. When I asked how many were interested in that subject, only a handful raised their hands. So, let those students experiment with AI and see what they think. This approach reflects the spirit of Adam Smith’s market-driven philosophy.
Russ Roberts: It’s crucial to highlight that we pride ourselves on not merely selling a credential at Shalem. While our students do graduate with an accredited degree, our primary focus is on transformation. They come to Shalem to become something, which significantly differs from a competitive, grade-centric environment. I believe this fosters a unique learning atmosphere, making your experiment particularly compelling.
Now, could you elaborate for listeners who may not have heard your talk at the University of Austin? When you mention taking a class with AI, what does that entail? Please outline how it would work, including the syllabus and assignments.
Tyler Cowen: Initially, students would work with a coach, who doesn’t necessarily need expertise in Tudor England but should know how to structure a class effectively. The AI could generate a reading list, quizzes, and other materials. Students would determine whether the class should include papers, quizzes, or a combination of both. They could even reintroduce a human grader at the end for assurance. Essentially, the AI would replicate a human-led class experience.
However, I have more radical ideas that I believe could enhance the learning experience even further. For instance, instead of a traditional class, students could engage in a lengthy dialogue with an AI over several weeks, with another AI assessing their engagement by evaluating what they learned from the interaction. While this may seem too radical, I advocate starting with the more familiar concept of AI-driven classes and allowing students to explore.
Russ Roberts: That’s a fascinating concept. Engaging in a sustained conversation with an AI could be enlightening. However, I wonder about sustaining that interaction over an extended period. While the initial engagement may be captivating, maintaining it over 13-14 weeks presents challenges, especially regarding focus and clarity of purpose. It would be essential to guide students on what specific topics to pursue with the AI.
That said, I believe that creative educational entrepreneurs will innovate beyond the traditional models you propose. This is a crucial initial step for transferring information effectively, particularly in fields requiring foundational knowledge.
However, when it comes to texts like the Odyssey, engaging with fellow students who have shared experiences, such as veterans reading the Iliad, offers a depth of understanding that an AI cannot replicate. The question then becomes how to incorporate group dynamics into the learning experience.
Tyler Cowen: My initial proposal allows for students to work in groups if the topic warrants it, especially in discussions around sensitive subjects like the Iliad. So, that flexibility remains.
Russ Roberts: That’s an interesting point. If we assume that collaborative learning is important, how can we design the group experience in a way that mimics the current educational model? In small groups, how do we ensure that the AI facilitates meaningful discussions and learning?
Tyler Cowen: Students could assist one another with their papers, utilizing AI to deepen their understanding of their topics. They might engage in discussions, record their conversations, and then query the AI for feedback on the points raised. The AI could even simulate what you might say based on your previous discussions.
Russ Roberts: That’s an intriguing idea. I can see how it could enhance collaboration and deepen understanding.
Tyler Cowen: Think of it this way: each student could take a class with the president, querying the AI on what you might say about a particular topic. The possibilities are vast!
Russ Roberts: Let’s take a closer look at your History of Economics class. How do you conduct your sessions? Do you lecture?
Tyler Cowen: Yes, I do lecture, but I also encourage engagement.
Russ Roberts: Why do you feel that’s important?
Tyler Cowen: There’s a unique value in vivid, face-to-face communication. Recently, I assigned students to use AI to teach themselves the Ricardian model. Later today, I’ll lecture on it, and I’ve asked them to mentally compare the AI’s teaching with mine. This exercise is a significant part of the learning experience.
Russ Roberts: That sounds fantastic. Your humor may be an edge, but it’s possible that even that advantage could diminish.
Tyler Cowen: Indeed, I recognize that my advantages may erode over time. I am ready to adapt; I’ve already shifted significant portions of my work to podcasting due to competition from AI in writing. I now focus more on personal appearances, which AI cannot replicate.
Russ Roberts: You mentioned a company called LearnLM, which aims to enhance AI tutoring quality. What are your thoughts on their approach?
Tyler Cowen: I’m only somewhat familiar. Various projects aim to modify a base AI model to enhance its tutoring capabilities. However, my intuition is that few of these will succeed. Most people will likely default to using the foundational model they are accustomed to, which may not be the best approach.
When I teach with AI, I emphasize using the base model and showing students how to leverage it effectively.
Russ Roberts: Let’s explore a more radical concept. Suppose you start a college where students can design their own major and curriculum entirely centered around AI, with human coaches and opportunities for social interaction. I wonder if we could create an environment where students have the freedom to craft their educational paths.
Imagine an 18-year-old student arrives, eager to learn about economics. They could approach you and say they want to understand the world through an economic lens, but they don’t know where to start. Could you envision a setting where they can explore their passions and demonstrate mastery creatively? How would you certify that mastery?
Tyler Cowen: Mastery could be assessed through testing and paper grading. The British education system continues to utilize a tutoring approach, which works effectively. With AI assistance, we can enhance that system significantly. We can build on existing models and gather data to refine our methods further.
Russ Roberts: When education was once reserved for a select few, these issues weren’t as pressing. Nowadays, education encompasses so much more, and the pursuit of wisdom seems overshadowed. Is there space in today’s landscape for a college dedicated to certifying transformative learning experiences? Would that model be appealing?
Tyler Cowen: Isn’t that what your college aims to achieve? You’re far more familiar with its mission than I am. Does the University of Austin not share a similar vision?
Russ Roberts: I’m not sure we can claim that distinction. While we graduate students with accredited degrees, we are primarily focused on transformation—helping students discover their ambitions and become effective agents of change in their country.
It’s an ambitious goal, and one that, if successful, could resonate widely. However, societal anxieties linger regarding job prospects, despite our students performing exceptionally well in their careers.
Tyler Cowen: Many individuals pursue transformative learning outside traditional institutions, using the Internet and AI. We don’t learn in rigid 15-week structures; we learn flexibly, adapting to our interests and needs. Yet we impose a different format on everyone else.
Russ Roberts: That’s true, but isn’t that due to the necessity of certification? We need to provide employers with assurance of competence in knowledge and skills. Once we intertwine education with that certification process, it complicates the educational experience.
Tyler Cowen: AI can vastly outperform us in certification. We are not there yet, but we will reach that point. Simply have someone spend time with an AI, assess their performance across various areas, and you’ll obtain far more useful information than traditional grading systems provide.
Russ Roberts: Recently, I asked a former high-ranking member of the Israeli military how they would approach selecting candidates for leadership roles. I care about intellectual aptitude, which combines brainpower and curiosity, as well as a genuine ambition to improve the country. I asked how they would conduct interviews differently. Their proposal involved taking candidates into the woods for three days. While I find that intriguing, I doubt it would be an effective marketing strategy for my college.
Given your experience with talent selection for your philanthropy project, Emergent Ventures, have you considered using AI in that process? For example, could candidates spend a day with AI and submit a transcript to help you understand them better?
Tyler Cowen: They already utilize AI in their processes!
Russ Roberts: Of course.
Tyler Cowen: Candidates may ask AI what questions I’ll pose during interviews, making my job more challenging.
Russ Roberts: Indeed.
Tyler Cowen: In fact, AI may soon surpass most human interviewers in effectiveness. It might not yet rival the best, but if it outperforms the majority, we’ve made significant progress.
Russ Roberts: The challenge seems to be that AI can come across as overly ingratiating.
Tyler Cowen: That can be easily adjusted. The key lies in teaching people how to interact with AI effectively, including how to elicit different responses.
Russ Roberts: I recently came across a post by former EconTalk guest Noah Smith, who expressed that the pleasure he derives from using AI mirrors the initial excitement he felt when he first engaged with social media. However, he cautioned about the potential societal ramifications, similar to the impact social media has had on our institutions. Do you think about these concerns regarding AI? While it’s evident you find this technology fascinating, do you believe it will ultimately lead to a positive future?
Tyler Cowen: I think the term “confidence” is subjective in this context. People generally resist change, and these shifts are substantial. Did individuals embrace the Industrial Revolution? Not necessarily. Yet, it has arguably been one of humanity’s greatest achievements. Thus, I believe AI will follow a similar trajectory.
As I noted in a previous interview, the more discomfort people express, the more likely we are to be on the right track. However, we must also acknowledge the potential political implications of these changes, both in the U.S. and globally. Many nations will face complex decisions regarding AI integration.
Russ Roberts: Your book, Stubborn Attachments, champions the idea of growth. It resonates with your outlook: we can expect more resources in the future, perhaps significantly more. When you assert that the Industrial Revolution was one of humanity’s greatest achievements, I assume that’s what you’re referencing.
Tyler Cowen: I would omit “perhaps.” It’s not just about material wealth; it encompasses creativity, opportunities, women’s liberation, and human rights. These are all integral to the message of Stubborn Attachments.
Russ Roberts: Please elaborate.
Tyler Cowen: Economic resources are crucial for enhancing human lives in various humanitarian aspects. Societies that are impoverished often exhibit lower tolerance for human rights and gender equality, resulting in a less favorable living environment. GDP per capita and non-GDP gains correlate significantly, which is noteworthy.
Thus, economic growth is essential. For a nation like Israel, national security is also at stake. Without AI, the country could be in jeopardy. In contrast, Brazil might be secure without it, but Israel’s situation is distinctly different.
Russ Roberts: I recall our first conversation about AI, where you emphasized the importance of Israel developing its AI initiatives. Reflecting on the last two years, it’s clear that there’s significant AI research and innovation happening here.
Israel is a remarkable hub of technological advancement and innovative thinking. While we may not always make the best choices, the potential for progress is undeniable.
Tyler Cowen: Many smaller nations lack such advantages. Most do not have the same opportunities.
Russ Roberts: That’s indeed a striking contrast.
Tyler Cowen: Geopolitical dynamics will undergo substantial changes as a result.
Russ Roberts: As we conclude, let’s consider some advice. Several years ago, I pondered whether my youngest child should pursue college. Would he be better off dedicating those four years to extraordinary pursuits instead of the traditional college experience? Ultimately, he chose to attend, and I believe he benefited both academically and personally. However, it’s a compelling question for today’s young adults: Is college still the right choice?
One clear takeaway is that some of the guidance we provided five years ago may no longer hold. For instance, the advice to ‘learn to code’ might not be as relevant today. There was even talk about making coding a core requirement at Shalem, which seems misguided in hindsight.
So, how should we guide today’s 18-year-olds navigating this rapidly changing landscape? What advice do you offer?
Tyler Cowen: Encourage them to learn about AI. Advise them to seek out what Luis Garicano refers to as “messy jobs,” which involve diverse tasks that can’t be easily automated, require substantial face-to-face interaction, and necessitate solving complex problems involving other people. That would be my counsel, and I receive this question almost daily.
