Sam Enright works on innovation policy at Progress Ireland, an independent policy think tank in Dublin, and runs a publication called The Fitzwilliam. Most relevant to us, on his personal blog, he writes a popular link roundup; what follows is an abridged version of his Links for January.  Blogs and short links 1. Henry Oliver on the literary anniversaries of 2026. Henry is also looking to hire an intern to research John Stuart Mill. I hope that said intern will discuss Mill’s essays on the political economy of Indian land and on ranked choice voting with me, which I find more interesting than On Liberty.1 2. From my colleague Seán O’Neill McPartlin: In political discourse, high rents and land prices are frequently blamed on ‘speculators’, but the use of that term is so deeply philosophically confused that I can’t help but think that many such arguments are not even wrong. Peter McLaughlin put it nicely: A fantastically neat insight from Seán that I’d never noticed before: when people blame speculation for high house prices, they often mean two completely distinct practices that would have opposite effects in practice. Sometimes it means land hoarding, refusing to sell and holding out for a better deal; other times it means excessive trading, selling too often because housing has become financialised. Yet people still talk as if ‘speculation’ were a unified thing, with a single effect (upward) on house prices. 3. The European Union’s ‘single market’ is being allowed to wither and die. I talked about this blog post at a drinks reception recently, and then a Dutchman gave me grief for referring to “self-righteous Europeans” in the third person (as if I wasn’t one). Note that the specific figures in this post – that EU member states have de facto 45% tariffs on goods and 110% on services—are extremely misleading. 4. In addition to the Great Firewall, some Chinese provinces are competing to put in place additional censorship, above and beyond the censorship required of them by the central government. Henan is a leader in this area.2 It’s increasingly common for Chinese websites to block access from any IP address based outside the mainland. A friend who consults for multinationals on business in China says that never in his career did he expect to have to VPN in to China. 5. ​​Apropos of Tyler Cowen’s post about the utilitarian track record of American-backed regime changes, this whole page is insane: The U.S. Army turned to psychological warfare [to get Panamanian dictator Manuel Noriega to leave the Vatican embassy], blaring disturbing chicken noises at “deafening levels”, gunning the engines of armored vehicles against the [Vatican embassy’s] fence, and setting fire to a neighboring field and bulldozing it to create a “helicopter landing zone”. Reportedly the version of the song “I Fought the Law“ performed by The Clash was played repeatedly along with “You Shook Me All Night Long“ by AC/DC and “Welcome to the Jungle“ by Guns N’ Roses; other songs in the line-up were “Too Old To Rock ‘N’ Roll“ by Jethro Tull, “Panama“ by Van Halen, and “Never Gonna Give You Up“ by Rick Astley. I can certainly think of ways of deposing a dictator more dignified than a Navy SEAL rick-roll. 6. Congratulations to Jamie Rumbelow for winning a Sidney Award (gated) from David Brooks for his piece in Works in Progress about Manhattan’s elaborate network of steam tunnels. 7. Polling insights: 12% of Americans claim to have a license to operate a submarine. Is Lizardman’s constant on the rise?3 8. Thomas Nagel, what is it like to be a bat, poetry edition. 9. Sam Mendelsohn’s introduction to the Mahabharata and Ramayana. It really does seem that reading one of the illustrated editions is the way to go.4 A section that got cut from the final draft of my essay Notes on Taiwan contained some speculation about why Eastern classics are so long compared to Western ones (I still don’t know). I really like this quote from A.K. Ramanujan: No Indian reads the Mahabharata for the first time This useful comment on Marginal Revolution gives further context on Indian oral culture. Music and podcasts 1. From the Works in Progress podcast: Anton Howes on how Henry VIII accidentally started the Industrial Revolution. There is a lot of great overlap between the discussion here and Anton’s session at our Adam Smith conference. 2. Seun Kuti, Egypt 80, Heavier Yet (Lays the Crownless Head). My favourite track is Dey. I also enjoyed Radiolab explaining Fela Kuti’s critical role in the history of Afrobeat. And here is the full 12-part series about Fela excerpted in that episode. Egypt 80 (formerly Africa 70) was Fela’s band, which is now led by his son Seun. 3. Dave Chappelle has a stand-up bit about how, if jobs moved from China to America, the iPhone would cost $9,000. He had the right order of magnitude: the only smartphone made in America uses decade-old technology and costs $2,000. Naturally enough, it’s called the Liberty Phone.5 4. Charles Lloyd, Zakir Hussain, Eric Harland, Sangam. Excellent Indian-inspired jazz fusion, which I wrote about in January 2025 as a criminally underrated subgenre. Dancing on One Foot is the most accessible track. See also Batson’s first explorations in Indian classical music. 5. Was Michel Foucault a libertarian? As with many of the questions that Rasheed Griffith asks, I suspect that Betteridge’s law of headlines applies. Books and Papers 1. Various, Beyond Reasoning Gains: Mitigating General Capabilities Forgetting in Large Reasoning Models. Read for my AI journal club, which was in an experimental ‘wisdom of crowds’ format. We read this paper in order to come to a collective judgement, before seeing the market rate, on what we should wager on Gavin Leech’s prediction market about whether reinforcement learning harms off-target capabilities. Fair-minded centrist that I am, my answer was 50%. I think this paper was sufficiently above my level that I am still at the “recognise some terms and talk to Claude and what they mean” phase. But you can still learn a lot from doing that! Reading computer science papers is such a different paradigm from reading philosophy, history, etc., in that “reading” is secondary to actually implementing the techniques yourself. I am still a noob at that, although Claude Code helps a lot. Reinforcement learning is probably one of the easier areas of computer science for an economist to learn. The basic mathematical machinery (value functions, dynamic programming, fixed point theorems) will be familiar to anyone who has considered doing an econ PhD at a quant-heavy department. ‘Reinforcement learning’ is also one of those terms, like inference, that is now used by the young ’uns in a very confusing way that is contrary to decades of previous usage. Sometimes I hear people use RL as though it were synonymous with the entire post-training stage of an LLM. At least in the Richard Sutton textbook, RL is a set of methods for learning how to maximise cumulative reward in an environment that can be modelled as a Markov decision process. By this definition, it seems to be debatable whether reinforcement learning by human feedback (RLHF) should count as genuine RL. People also sometimes use ‘RL’ when talking about things like supervised fine-tuning, which is definitely not RL. So I suppose my contribution to this group was very much that of a philosopher: arguing that the initial question was confused because of its failure to parse semantic distinctions that most people find annoying and pedantic. 2. Andrew Brown, J.D. Bernal: The Sage of Science. One of the most underrated science books I have ever read. I’m convinced John Desmond Bernal was one of the great scientific polymaths of the 20th century. I have many thousands of words of notes about this book, and hope to get around to profiling his work for Asimov Press at some point. 3. Allen Newell, Herbert Simon, The Logic Theory Machine. A paper that grew out of the Dartmouth summer workshop on AI. The Logic Theorist was an automated theorem-proving program that ran on the JOHNIAC at RAND. In this paper, the Logic Theorist is given 52 theorems from chapter two of Russell and Whitehead’s Principia Mathematica, of which it was able to prove 38.6 In one case, the proof is more elegant than Russell and Whitehead’s own. Personally, I find this quite astonishing; AI had made an original contribution to mathematics – at least in the sense of simplifying an existing proof – as early as the 1950s! In any case, having spent an inordinate amount of time reading Bertrand Russell last year has already paid off more than I would have expected. Films and video 1. Park Chan-wook, No Other Choice (어쩔수가없다). From the director of Decision to Leave and Oldboy, two of my favourite Korean films. This film definitely won’t stick with me as much as Park’s other films, but it was light and funny, and we had a great time at the screening. The central comedic tool of making a bizarrely specific industry seem like a much bigger deal than it really is economically works very well. The Korean embassy in Ireland should run some kind of memorial event for Kevin O’Rourke, the missionary from County Cavan who translated much of the Korean cultural canon into English for the first time. He became a professor at Kyung Hee University, an honorary Korean citizen, and the first foreigner in history to receive a doctorate in Korean literature. A long time ago, I knew some of his family friends from Busan. We truly have people everywhere. 2. From YouTube, we have Amanda Askell on training Claude’s character and why Opus 3 was so well-aligned. Welch Labs also explains the phenomenon of double descent and how it violated conventional wisdom in statistical learning theory. Finally, there is Jacob Collier and Esperanza Spalding on NPR’s Tiny Desk.   You can read the full version of Sam’s January links here. [1] I saw someone on Twitter recently describe JS Mill’s writing style as “undergraduate with a deadline at midnight”, which is an assessment I probably disagree with less than Henry does. [2] Shockingly, “The Henan Cyberspace Affairs Commission could not be reached for comment.” [3] I got the DK illustrated version of the Mahabharata for an old flatmate from Mumebai. Probably my fondest memory in that flat was trading back and forth strangely translated cultural peculiarities: “I see your Gujarati spiderman, and raise you Irish Spongebob.” [4] Bizarrely, this effect is almost entirely driven by the quarter of Hispanic adults (!) who claim to know how to operate a submarine. Is there a tradition of Latin American pranksterism I was unaware of? [5] I hope this does not come across as snarky; it’s a genuinely amazing engineering accomplishment. [6] See section 3.5 of the Stanford Encyclopedia of Philosophy’s entry on computational philosophy. (0 COMMENTS)

We’ve posted the second of two cross-posted articles with Law & Liberty in response to the Supreme Court ruling in Learning Resources v. Trump. Today, David Hebert explains why the economic fallout from the tariffs can’t be reversed by the Court’s ruling. From the article: Just over a year ago, citing the International Emergency Economic Powers Act (IEEPA), President Trump began unilaterally changing tariff rates with countries around the world. The goal was to restructure global trade. Since this was the first time any president had used IEEPA in this way, it was always going to invite challenges. In May, the U.S. Court of International Trade ruled against the president. In November, the Supreme Court heard oral arguments on the case. Last week, in Learning Resources v. Trump, the Court issued a 6-3 decision, making it clear that IEEPA does not give the president the power to unilaterally impose, rescind, and adjust tariffs as he sees fit. Chief Justice Roberts, who wrote the majority’s opinion, held that tariffs are fundamentally a taxing power and that, because of this, they are different in kind, not just degree, from the trade tools that IEEPA explicitly authorizes. This opinion is certainly an important legal victory, but we should not confuse it with an economic one. The damage of tariffs has already been done and it is continuing to be done. We hope you’ll read the whole article, which you can find here. (If you missed it, take a look at John O. McGinnis’ discussion of the legal implications of the ruling, too.) (0 COMMENTS)

Trade is not just about transactions. It’s about relationships and trust built and earned over time. Just over a year ago, citing the International Emergency Economic Powers Act (IEEPA), President Trump began unilaterally changing tariff rates with countries around the world. The goal was to restructure global trade. Since this was the first time any president had used IEEPA in this way, it was always going to invite challenges. In May, the U.S. Court of International Trade ruled against the president. In November, the Supreme Court heard oral arguments on the case. Last week, in Learning Resources v. Trump, the Court issued a 6-3 decision, making it clear that IEEPA does not give the president the power to unilaterally impose, rescind, and adjust tariffs as he sees fit. Chief Justice Roberts, who wrote the majority’s opinion, held that tariffs are fundamentally a taxing power and that, because of this, they are different in kind, not just degree, from the trade tools that IEEPA explicitly authorizes. This opinion is certainly an important legal victory, but we should not confuse it with an economic one. The damage of tariffs has already been done and it is continuing to be done. Consider that just hours after the Court’s opinion was released, President Trump held a press conference where he said, “[Foreign countries] that have been ripping us off for years are ecstatic. They’re so happy and they’re dancing in the streets, but they won’t be dancing for long. That I can assure you.” True to his word, the president announced later that day that he was using Section 122 of the Trade Act of 1974 to impose a 10% global tariff on all imported goods for 150 days beginning on February 24th. As if there were any doubt about this, the White House also posted on X, “Keep Calm and Tariff On.” Later in the weekend, the president announced that the new tariff rate would be 15%, the maximum rate allowed under Section 122. But now, as these tariffs come into effect, the rate is set at 10%, inviting further confusion. It remains unclear whether this is on top of any trade deals that have been signed or if those countries will be somehow exempted. And these new tariffs are already raising serious concerns about their legality. A 2021 survey of members of the American Economic Association found that 95% of economists agreed that tariffs are economically destructive. In other words, a group of people so famous for disagreement that the jokes practically write themselves has 95% consensus on this issue. Plenty of reputable people have asked the question of what the effective tariff rate is, who actually pays the tariffs, and how many jobs will be created or lost. This is important to the work of gathering (further) evidence of the destructive effects of tariffs. But the decades of empirical, historical, and theoretical work on this front fail to capture the real cost of tariffs. It won’t show up in any BLS report, BEA release, or any other economic report one can imagine. The real cost is the destruction of trust on the world stage. Trade is not just about transactions. It’s about relationships and trust built and earned over time. This establishes that trading partners will play by agreed-upon rules and that market access is not a bargaining chip to be leveraged whenever one side, in this case Washington, needs a political victory. “Political leaders and business executives around the world must ask themselves a new question in international trade: Is access to the largest consumer market in the world worth the cost of dealing with a partner who treats market access as a bargaining chip?” Adam Smith understood this. He knew that the wealth of nations wasn’t built on clever tariff schedules or trying to hold the rest of the world hostage. It’s built on expanding the division of labor, broadening the extent of the market, and enabling man’s propensity to “truck, barter, and exchange.” All these things are made possible by stable rules and predictable networks of exchange. Smith understood that tariffs altered these incentives, but even he may have underappreciated the role of trust in international trade, how quickly it can be eroded, and what happens when it is. Rather than breathing a sigh of relief after the Court’s ruling, the rest of the world is getting further confirmation that this administration, and by extension the United States of America, is no longer trustworthy. The first trade deals of 2026 have included deals meant to limit the damage that can be done by the turn away from trade by the United States. Canada and China announced a “trade reset,” with the Canadian prime minister, Mark Carney, referring to China as “more predictable” than the United States. When China, of all places, is viewed as more predictable than the U.S., something has gone very, very wrong. That’s not the only warning sign that we’re seeing. The EU has signed trade deals with Mercosur, which covers 31 countries, and with India. The deal with India is particularly noteworthy because it covers 25% of world GDP and over 2 billion people. This deal is so large that the president of the European Commission, Ursula von der Leyen, referred to it as “the mother of all deals.” And Prime Minister Carney, after his rousing speech in Davos, is leading the charge of “middle powers” to unite around free trade, providing other countries with an alternative to dealing with U.S. trade policy. The rest of the world is not looking to Canada and Mark Carney because they are somehow world leaders in this space, but because if Canada, one of America’s longest-standing and closest allies, with the longest undefended border in the entire world, says that they’ve had enough, surely other countries have had enough, too. Carney’s poll numbers show that Canadians support him, even if standing up to Trump imposes serious economic costs. Political leaders and business executives around the world must ask themselves a new question in international trade: Is access to the largest consumer market in the world worth the cost of dealing with a partner who treats market access as a bargaining chip? Or is it better to work with smaller but more dependable markets whose leaders won’t wake up one morning and decide that they need to alter the deal? International trade used to be about David Ricardo’s insights about comparative advantage and maximizing gains from trade. Now, it’s about Harry Markowitz’s portfolio theory, diversifying away from risk and minimizing losses in the worst case. The risk they’re hedging against is U.S. policy. While the United States is deglobalizing, the rest of the world is reglobalizing around partners who commit to the impersonal rules of an open-access liberal order. The Court’s ruling in Learning Resources does nothing to fix this. Worse, neither will the next election. Even if America happens to elect someone who goes on a world tour promising to be a more reliable trading partner, it’s not that easy to restore trust once it’s lost. To the extent that rebuilding trust is possible, it will be a long process that starts from a worse position. The rest of the world marches on. In boardrooms and government offices, supply chains are being rerouted. Permits to construct factories are being submitted. Long-term contracts are being signed. Investment decisions are being made today with even more uncertainty surrounding American policy, and with this uncertainty taken as a given, not the result of a short-term, recognized error. New factories around the world aren’t going to be packed up and moved to America because the next president holds a press conference and apologizes. Supply chains being built now won’t be rerouted through the U.S. because a social media post promises that the politics in Washington have changed. Trump showed the rest of the world what is possible in the American system. And the rest of the world is responding predictably. The word to describe this moment in American history is “hysteresis.” The idea is that something that looks like a small or temporary event, such as a temporary layoff, can have an effect that is much larger than would otherwise be predicted. Hysteresis is a term often used in economics to describe unemployment, where a worker who is originally only temporarily laid off due to a recession never returns to the labor market, or does so only in a limited capacity. Now, we have another example to use in the classroom. The Court’s ruling in Learning Resources is a genuine victory for free trade and for constitutional limits on executive power. But it does nothing to rebuild the relationships that have already been strained. Every workaround, every legal maneuver, and every new emergency declaration will send the same message to the world: The United States can no longer be trusted. You can’t build lasting trade relationships on that foundation, and the rest of the world is learning not to try. The Trump Administration wanted to restructure global trade. They got their wish, just not the way they imagined. The rest of the world is restructuring, too, and it’s doing so around the United States, not with it.   This essay has also been published on Law & Liberty, part of the Liberty Fund network. *David Hebert, PhD, is a senior research fellow at AIER. He has also been a fellow with the US Senate Committee on the Budget and has worked for the US Joint Economic Committee. He also serves as an associate director of The Entangled Political Economy Research Network. (0 COMMENTS)

Learning Resources reaffirms that taxation is Congress’s responsibility, and declaring “emergency!” does not rewrite the separation of powers. The Supreme Court’s decision in Learning Resources v. Trump will have immediate political effects, substantial economic effects, and more subtle but long-run effects on the shape of the law. Doctrinally, its significance may seem limited because the opinions fracture on nearly everything beyond a single issue under a specific statute—the International Emergency Economic Powers Act (IEEPA). Six members of the Court agreed only that IEEPA does not authorize the president to impose tariffs. Even on that conclusion, however, the justices split into two camps: one relied on the major questions doctrine, namely that because of the extraordinary power claimed, Congress had to speak more clearly than it did, while the other concluded the president’s lack of authority was manifest without reliance on any clear-statement rule. Nevertheless, the case is still significant for the separation of powers. Underlying all opinions of the justices in the majority is the shared premise that tariffs function as taxes and thus are within the purview of Congress’s power of the purse. A declaration of emergency does not shift authority to the president. In this sense, Learning Resources reasserts Congress’s primacy. Politics First, the political impacts are consequential for both the president and the Court. The president has consistently put tariffs at the heart of his economic program. Moreover, his trade policy reflects his general populist mantra: that America is threatened by people and goods coming from overseas. Beyond the legal subtleties of the opinion, many citizens will take the case as a rebuke to that vision. Like other second-term presidents, Trump is entering the lame-duck stage, and the case will also reinforce the perception that he is losing power. And power in politics is dependent on the perception of power. Given the president’s past behavior, one may look for him to assert his continued centrality. Foreign affairs and military action may offer such opportunities. The opinion is also important to the politics surrounding the Court. Many Democrats have charged that the Supreme Court is abetting the authoritarianism of the Trump administration by upholding, often on the emergency docket, a wide range of actions, such as the firing of commissioners of independent agencies. These criticisms may be laying the foundation for changing the composition of the Court either by expanding its size or by imposing statutory term limits the next time the Democrats have unified control of government. After all, President Biden and candidate Harris endorsed the latter. But the Court’s invalidation of a core part of the Trump program will make that move harder to defend, particularly if, as I suspect, he suffers other major defeats at the Court’s hand during his administration. By underscoring the independence of Republican-appointed justices, President Trump’s denunciation of them may paradoxically safeguard one of his greatest achievements: a Supreme Court most friendly to defending the Constitution’s original meaning in at least a century. Economics The economic effects are also likely substantial but not entirely clear. Although the Court did not directly say that tariff revenues must be repaid, any unliquidated tariff payments will be reversed. Some have suggested that the government will be on the hook for approximately $175 billion. The next question is the extent to which the president can replicate these tariffs under other statutes. No authority, however, has the economy-wide breadth, the discretion as to amount, and the lack of requirement for specific findings that made IEEPA so attractive to the president. Section 122 of the Trade Act is closest, and thus it is no surprise that in defeat, the president immediately invoked it. That provision authorizes economy-wide tariffs, but only at a rate of up to 15 percent ad valorem for 150 days. That is in itself an advantage for long-term economic growth because Congress is unlikely to approve them for the long term. It is also not nearly as useful as IEEPA for Trump, who wants to deploy tariffs as leverage for negotiations, because other nations can wait him out. Section 232 allows the president to adjust tariffs on any article, but requires a national security finding, consultation, and investigation. Trump may try to speed up compliance with all these requirements and claim that national security requires tariffs on all goods, but such actions are certain to be challenged. Section 301 has also been mentioned, but that provision requires specific factual findings after an investigation that individual countries are behaving economically unfairly as a prelude to tariff retaliation. One economic advantage, even during their pendency, is that these substitute authorities result in more predictable, rule-based tariffs. While all tariffs can harm economic growth, the arbitrary and mutable impositions authorized by IEEPA can do the most damage. Law What will be the long-term effects of the decision, other than the holding that IEEPA does not authorize the president to impose tariffs? One legal implication is that the Court’s historic solicitude for deferring to the president when “foreign affairs” is invoked may be waning, in part because of the turn toward originalism. In cases such as United States v. Curtiss-Wright, the Court previously deployed a kind of atextual foreign affairs exceptionalism to protect exercises of presidential power that the Court might otherwise have struck down (there, under the non-delegation doctrine). But here six justices joined in that part of the chief justice’s opinion that distinguishes tariffs from foreign affairs because the authority over foreign commerce lies in Article I, not Article II. There is thus no constitutional reason to interpret presidential statutory authority broadly, when that authority depends on a legislative grant from Congress. As the Americans have known since the Revolution, a tariff trenches on our domestic liberties, even if it affects foreigners too. A second legal consequence is likely to be the further cementing of the major questions doctrine as central to statutory interpretation in administrative law. The doctrine is an obstacle to the executive’s broad interpretation of delegations in important regulatory cases, at least outside statutes that, like IEEPA, can be alleged to touch on foreign affairs. “The strengthening of the major questions doctrine pushes core political decisions back to Congress, where the Constitution expects them to be made.” First, the chief justice and Justices Barrett and Gorsuch in the majority, and Justices Kavanaugh, Thomas, and Alito in dissent, all speak very favorably of the doctrine. Justice Kavanaugh calls it an “important canon” of statutory interpretation, and his principal reason for not applying it here—that tariffs touch foreign affairs and fall in the “president’s wheelhouse”—will not translate to domestic regulation. Even for the dissent, the question is no longer whether the major questions doctrine exists, but the circumstances in which it applies. Thus, for six justices, the major questions doctrine will continue to preclude agencies and even the president from relying on arguably ambiguous language to impose substantial restrictions on citizens’ liberties when these restrictions have large economic and political effects. Second, in a tour de force concurrence that may be his best since joining the Court, Justice Gorsuch places the major questions doctrine on a sounder footing. The complaint against the doctrine is that it is fabricated—made up by current conservatives to frustrate the administrative state. But Gorsuch shows that antecedents of the major questions doctrine antedated the Constitution, let alone the rise of the modern administrative state. Corporate charters granted by Parliament were read narrowly. For instance, an English court required an express statement before interpreting the broad authority of the Company of Cutlers to take and forfeit the wares of its members because they were defective. As the example suggests (and scholars have shown), state-authorized corporations often had much in common with administrative agencies; in any event, interpretive rules governing public charters illuminate how grants of authority were construed. Gorsuch also points to English case law suggesting that “the legality of executive action depended on the relationship between the size of the asserted power and the clarity of the underlying authority.” Here, he quotes a very recent English law review article, showing that Gorsuch’s practice of employing law professors as law clerks is paying dividends. Gorsuch is making the case that the major questions doctrine reflects a long-established method of interpretation, applicable to the interpretation of provisions empowering the executive within our constitutional system from early on. He finds what he calls an “extraordinary power” clear-statement rule and says it “looked strikingly like” the major questions rule. The proper interpretation of legal texts depends on legal context. Gorsuch supplies a long-established context that justifies the major questions doctrine no less than other settled, clear-statement rules, such as the rule governing waivers of sovereign immunity or that against retroactive legislation. To be sure, such clear statement rules were originally grounded in values, but once they were accepted into the law, they operate as part of the legal context. Thus, statutory meaning in documents written in the language of the law includes settled interpretive conventions (even if originally value-laden), not just dictionary semantics. Third, Gorsuch calls out Kagan and other liberal justices who reject the major questions doctrine for their inconsistency in interpreting their administrative statutes. In Learning Resources, they insist there is no need for a major-questions clear-statement rule to read an arguably broad delegation “to regulate … importation or exportation” as excluding the power to impose tariffs. In this case, they are happy to construe the text narrowly. But, as Gorsuch notes, that is not the way two of them previously read arguably broad grants to authorize agencies to deploy unprecedented sweeping powers. For instance, Kagan and Sotomayor read the Occupational Safety and Health Administration’s mandate to create “safe and healthy working conditions” as authority to impose a vaccine mandate for a virus that had no particular connection to work. Those two justices also concluded that a statute allowing the Centers for Disease Control and Prevention to issue regulations to prevent the “transmission of communicable diseases” authorized a nationwide eviction moratorium. All three of the liberal justices held that the vague and ambiguous grant of authority to the Environmental Protection Agency to employ “the best system of emission reduction” was enough to close down coal plants, although the EPA had only previously used the authority to mitigate pollution from power plants. Most recently, they even construed a statute that allowed the Secretary of Education to “waive or modify” statutes or regulations relating to student loans, leading to an unprecedented cancellation of hundreds of billions of student loan debt. Gorsuch makes a powerful case that the liberal justices appear to be applying a different mode of statutory interpretation to the Trump administration’s tariffs than they have to the broad authority previously exercised. Even if the charge will not prompt Justices Kagan, Sotomayor, and Jackson to embrace the major questions doctrine, it does undercut the strongest selling point of their position: that they are principled textualists, while the justices invoking major questions are merely result-oriented manipulators of meaning. The strengthening of the major questions doctrine pushes core political decisions back to Congress, where the Constitution expects them to be made. That shift promises greater stability and less polarization because Congress is more likely to accommodate diverse and conflicting interests through bargaining and compromise than presidents or agency heads are to act by executive fiat based on ambiguous statutes. When sweeping burdens are imposed on citizens and when the economic and political stakes are immense, statutory interpretation should not turn on ambiguous language stretched to provide the executive with more power. Learning Resources reaffirms that the legislative authority is the responsibility of Congress and that declaring “emergency!” does not rewrite the separation of powers.   This essay has also been published on Law & Liberty, part of the Liberty Fund network. *John O. McGinnis is the George C. Dix Professor in Constitutional Law at Northwestern University and a senior writer at Law & Liberty. He is the author of Accelerating Democracy (2012) and coauthor, with Mike Rappaport, of Originalism and the Good Constitution (2013). He is a graduate of Harvard College; Balliol College, University of Oxford; and Harvard Law School. He has written for leading law reviews, including Harvard, Chicago, and Stanford, as well as The Yale Law Journal, and in journals of opinion, including National Affairs and National Review. (0 COMMENTS)

This morning we’re hosting the first of two cross-posted articles with Law & Liberty in response to the Supreme Court’s decision in Learning Resources v. Trump. The first, by John O. McGinnis, provides an overview of the legal aspects of the ruling. From the article: The Supreme Court’s decision in Learning Resources v. Trump will have immediate political effects, substantial economic effects, and more subtle but long-run effects on the shape of the law. Doctrinally, its significance may seem limited because the opinions fracture on nearly everything beyond a single issue under a specific statute—the International Emergency Economic Powers Act (IEEPA). Six members of the Court agreed only that IEEPA does not authorize the president to impose tariffs. Even on that conclusion, however, the justices split into two camps: one relied on the major questions doctrine, namely that because of the extraordinary power claimed, Congress had to speak more clearly than it did, while the other concluded the president’s lack of authority was manifest without reliance on any clear-statement rule. Nevertheless, the case is still significant for the separation of powers. Underlying all opinions of the justices in the majority is the shared premise that tariffs function as taxes and thus are within the purview of Congress’s power of the purse. A declaration of emergency does not shift authority to the president. In this sense, Learning Resources reasserts Congress’s primacy. Check out the whole article here, and check back tomorrow for the economic angle from David Hebert. (0 COMMENTS)

Crown Prince Mohammed bin Salman has been dragging Saudi Arabia into the modern world over the last decade. Journalist and author Karen Elliott House lays out the Saudi leader’s motivations, hopes, and contradictions. Listen as she and EconTalk’s Russ Roberts explore the crown prince’s mix of cultural liberalization and political dominance and where his balancing […] The post The Man Who Would Be King of Saudi Arabia (with Karen Elliott House) appeared first on Econlib.

Even people who are otherwise enthusiastic about a free market in labor can get cold feet about immigration once redistribution enters the picture. Some are fond of quoting Milton Friedman, who famously (or infamously) said: “It’s just obvious you can’t have free immigration and a welfare state.” On this view, immigration is fine under fully free market institutions, but in the actual world with its abundant government-provided benefits, immigration restrictions are justified to protect taxpayers from the added expense that could arise if immigrants consume these benefits. But this conclusion is too quick, and even Friedman’s position is more nuanced than people on both sides of the immigration debate tend to realize.  An initial point, though: the concern about the fiscal cost of immigration is overstated. For one reason, in the United States, most welfare spending goes to the very young or the very old. Immigrants, by contrast, are disproportionately of working age.   Setting that point aside, Friedman’s own view wasn’t that immigration as such is harmful. He argued that legal immigration is the problem, precisely because it allows immigrants to access government benefits. By contrast, he thought illegal immigration was beneficial. As he put it: “It’s a good thing for the illegal immigrants. It’s a good thing for the United States. It’s a good thing for the citizens of the country. But it’s only good so long as it’s illegal.” Friedman’s reasoning was that illegal immigration enables mutually beneficial market exchange while limiting immigrants’ access to government benefits. Now, many fiscal conservatives balk at Friedman’s recommendation—namely, if the overconsumption of government resources is the problem with lawful immigration, the solution is to encourage people to break the law. I understand this reaction, but I admit I don’t share it. In my view, whether it’s okay for someone to do something doesn’t depend on whether lawmakers give them written permission. For instance, did you know that it’s against the law to drive on Cape Cod’s National Seashore’s beach if there’s not a tire-pressure gauge in your car?  Nevertheless, I have no moral objection if you drive on the beach gaugelessly. Regardless of whether government officials approve, this is just a peaceful activity that doesn’t violate anyone’s rights. Maybe you disagree with me. Still, as others have suggested, there’s another way to accommodate Friedman’s general idea: admit immigrants as lawful permanent residents but restrict their access to certain government resources. Economists sometimes call this a “keyhole solution”—if the problem is immigrants’ consumption of benefits, then design a policy that narrowly targets that problem rather than restricts their freedom to immigrate entirely.  The main objection to this sort of policy seems to be moral rather than economic. Indeed, Friedman himself was asked about it and he replied that he found the proposal unappealing partly because it’s not “desirable to have two classes of citizens in a society.” That’s a good point. It’s unfair for a government to give some citizens taxpayer-financed benefits but not others. If two people live, work, and pay taxes within a country, government officials should treat them equally, which involves giving them both equal access to government resources.  Notice, though, that a policy of immigration restriction also treats citizens and prospective immigrants differently—it gives citizens, but not immigrants, access to domestic labor markets, private associations, educational opportunities, and more. Consequently, a principle of equal treatment actually seems to imply open borders. Given that Friedman rejects this option, the task becomes that of identifying the second-best solution. (Also, it’s not clear that Friedman can square his objection to keyhole solutions with his endorsement of illegal immigration, which would presumably also create two classes in a society.) Why think that a policy of open immigration with restricted access to benefits is better than outright exclusion? The reason, in brief, is that admission with conditions treats prospective immigrants betterthan exclusion. A policy of open immigration with restricted benefits at least gives people the option to move, and it’s hard to see how giving someone a new option could make them worse off. Here’s an analogy. Suppose John is entering the job market. One employer offers him a job with health insurance and a retirement plan. The next day, he receives another offer—this one comes with no benefits, but a much higher salary. Even if you think he should take the first job, it seems perfectly permissible to offer him the second. John is no worse off for having another option. If he doesn’t want to take it, he can simply decline it. And if he does prefer higher pay without benefits, he’s clearly better off for having the option. John’s case is analogous to the case of a prospective immigrant who expects to earn significantly more by moving to a country where her access to government benefits is limited. If she prefers having access to a wider range of government-provided benefits in her current country to having higher earnings but fewer benefits in a new country, she can decline to move; in this case, she is no worse off for having the option. But if she prefers higher earnings with fewer benefits, the option makes her better off.  Just as it’s permissible—indeed, probably good—to offer John the extra option, so too is it permissible to offer prospective immigrants the extra option.  It’s also worth highlighting another important aspect of restricting immigrants’ access to benefits rather than restricting their movement entirely. Admitting immigrants as lawful permanent residents removes the threat of deportation, among other consequences, that accompanies undocumented entry into a country. Even if you agree with Friedman (as I do) that the keyhole solution of admitting immigrants with reduced access to benefits isn’t totally fair, it’s still more fair than denying prospective immigrants the option of safely moving at all. (0 COMMENTS)

Generative artificial intelligence (AI) is upending professions as diverse as art, cinema, accounting, national defense, and education. Some even argue that AI will render almost all work obsolete. They say its ability to “think” and accomplish tasks previously solely in the realm of human ability will mean that humans will not need to work; the machines will do everything for us. Whether this would be a good thing or a bad thing depends on the story one wants to tell.  Some claim the loss of work to AI will lead to class warfare, as the poor get poorer and the rich get richer. Others think that, if AI ends work, it necessarily means that scarcity ends as well, and thus there will be no need for money. Others claim AI will destroy all human life long before we get to that point and it’ll be moot. Still others say that AI will obey the laws of economics and we’ll never reach that point (on these last two, see Ole Miss economist Henry Thompson’s working paper “Some Economics of Artificial Super Intelligence”). The question of what will happen if AI eliminates work is interesting, but I want to focus on how AI is likely to affect work and jobs. Concerns about the effect of machinery on labor and labor income is hardly new. Many of the concerns about AI echo those of the Luddites, a movement in 19th-century Great Britain against the introduction of weaving machinery into the trade. The Luddites feared that some of the automation entering weaving would lead to cheap, low-quality output and displace skilled workers. They launched a sabotage campaign against the machinery, but ultimately lost the battle. Over the next two centuries, the textile industry became highly automated. Fast forward to the 20th-century, John Maynard Keynes made a similar (albeit less pessimistic) prediction about how automation would affect work. In his 1930 article “The Economic Possibilities for Our Grandchildren,” Keynes argued that automation would allow us to work for just three hours a day. Both predictions ended up being wrong.  Some weavers lost jobs, but the industry was hardly overtaken by slop. Wages actually rose for skilled textile workers. In 1800, the average wage of a textile worker was about 25 shillings a week (£91.68, adjusted for inflation), or approximately £4,767, inflation-adjusted, per year. Currently, the average annual textile wage for a skilled worker is £29,000.  Why did wages rise? Because automation changed the nature of work. Workers who could not do more than recreate what the machines did, lost their jobs. Those who found ways for the machines to complement their work saw their productivity (and thus wages) increase. A recently-accepted paper in the Journal of Labor Economics by Daron Acemoglou, Hans Koster, and Ceren Ozgen finds these results hold even with modern automation. What about Keynes’s prediction? According to the Bureau of Labor Statistics, people in 2024 worked on average 42 hours (full-time) or 34.2 hours (all workers) per week.* While that is down some from Keynes’s day of approximately 47 hours a week,** it’s still a far cry from 15 hours a week.  In a short paper published last year in Industrial and Organization Psychology, my colleague Dr. Anne-Marie Castille and I explore the reasons for these incorrect predictions. We argue that the reason Keynes and the Luddites were incorrect is that they failed to recognize that automation did not resolve the reason we work: resources (not the least of which is time) are scarce.  Initially, we spend our time on high marginal value activities. As automation comes along to reduce the amount of time we need to spend on those activities, it frees up time for lower marginal value activities. That is, activities we were not previously engaging in because the cost was too great. What those activities are differ from person to person. And sometimes, new desires are created or discovered. We write: “Domestic chores used to be very labor-intensive, requiring many labor hours to accomplish. Washing laundry required each item to be hand washed, dried on a clothesline, and brought in when finished. With the invention and proliferation of the washing machine, the time needed to wash laundry dropped precipitously. The labor hours per load of laundry are probably less than 20 mins. Domestic laborers (mainly women) now had much more time on their hands. They could choose to take leisure or choose to spend those hours in other ways. Many women chose to spend these newly liberated hours by joining the workforce. … “As people got wealthier, their basic desires of food security, shelter security and companionship were satisfied using fewer resources. Thus, the trade-off we have discussed came about: Does one use the newfound time for leisure or for work so that we may satisfy other desires?” I argue now that the same pattern will repeat with generative AI. As AI proliferates throughout our society, some jobs will be lost, yes. But new jobs will be discovered. What those jobs are, I do not know. No one knows. Human ingenuity and desire know no bounds. We will discover new ways to satisfy our new desires, and we will continue to work. AI will not solve the underlying motivation for work: scarcity.   *Average weekly hours of all employees, total private, seasonally adjusted, Series ID: CES0500000002. **See table 2. (0 COMMENTS)

To prepare for teaching, I am reading a famous article in AI research: The Bitter Lesson, written by Richard Sutton in 2019. I wondered what would seem prescient and if anything would feel like Sutton had gotten it wrong. At the end, I’ll discuss economic implications.  Sutton draws from decades of AI history to argue that researchers have learned a “bitter” truth. Researchers repeatedly assume that computers will make the next advance in intelligence by relying on specialized human expertise. Recent history shows that methods that scale with computation outperform those reliant on human expertise. For example, in computer chess, brute-force search on specialized hardware triumphed over knowledge-based approaches. Sutton warns that researchers resist learning this lesson because building in knowledge feels satisfying, but true breakthroughs come from computation’s relentless scaling. In AI, scaling means making models larger and training them on more data with more compute.1 The Bitter Lesson is less about any single algorithm than about intellectual humility: progress in AI has come from accepting that general-purpose learning, persistently scaled, outperforms our best attempts to hard-code intelligence. It matters whether Sutton is right or wrong, because we are not at the end of the explosion of AI or the period of time dubbed “The Scaling Era” by Dwarkesh Patel.  EconTalk guests have speculated that AI will save the world or kill us all. See the following:  EconTalk: Eliezer Yudkowsky on the Dangers of AI EconTalk: Erik Hoel on the Threat to Humanity from AI  EconTalk: Marc Andreessen on Why AI Will Save the World Such extreme predictions assume that AI capabilities will advance. Although AI has been improving rapidly since Sutton wrote in 2019, there is no law of nature (that we know of) that insists it must continue to improve. Sometimes people even claim to see AI capabilities leveling off or point out that hallucinations persist even in advanced models.  If scaling is indeed the road to more intelligence, then we can expect AI to continue to exceed expectations if we add more hardware to the system. This hypothesis is being tested: US private AI investment could exceed $100 billion annually, representing one of the largest technological bets ever.  Let’s examine Sutton’s thesis in light of recent performance. We can point to three pieces of evidence that Sutton was correct about scaling. First, game-playing AI provides a clean natural experiment. AlphaZero learned chess and Go through self-play, without human openings or strategy. AlphaZero surpassed earlier systems built on domain expertise. Its success came from scale and computation, just as Sutton predicted. Second, natural language processing (NLP), the branch of AI focused on enabling computers to understand and generate human language, shows the same pattern. Earlier NLP systems emphasized linguistically informed rules and symbolic structure. OpenAI’s GPT-3 and successors rely on generic architectures trained on vast data with enormous compute. Performance gains track scale more reliably than architectural cleverness. The third example is computer vision. Hand-engineered feature pipelines (techniques where programmers manually designed algorithms to detect edges and shapes) were displaced once convolutional neural networks (a type of AI architecture loosely inspired by the visual cortex and designed to automatically learn visual patterns from data) could be trained at scale. Accuracy improved as datasets and compute increased. Sutton’s argument concerns the scalability of methods, but in practice that scalability only becomes visible once capital investment lowers computational constraints.  The rate of AI advancement reflects not just technological possibility but the unprecedented mobilization of financial resources. The typical person using ChatGPT to make grocery lists might not know what the word “scaling” means. A possible reason for underestimating the rate of progress is not just a misunderstanding of the technology but a missed estimate of how much money would be poured into it.  I compare this to the Manhattan Project. People doubted the Manhattan Project not because it violated physics, but because it seemed too expensive. Niels Bohr reportedly said it would require “turning the whole country into a factory.” But we did it. We are doing it again. We are turning the country into a factory for AI. Without all that investment, the progress would be slower. However, neither the doomers nor the utopians will turn out to be right if we are near a limit to either the power of scaling or our ability to physically continue to scale. Is the bitter lesson useful for seeing us through 2026 and beyond? This matters for unemployment today and existential threat tomorrow.  Recent economic research offers a nuanced view. In a January 2026 paper, economist Joshua Gans develops a model of “artificial jagged intelligence”. Gans observes that generative AI systems display uneven performance across tasks that appear “nearby”: they can be excellent on one prompt and confidently wrong on another with only small changes in wording or context. Anyone who has used ChatGPT to help with a work task and then watched it hallucinate a plausible-sounding falsehood has experienced this jaggedness firsthand. What makes Gans’s analysis economically interesting is his treatment of scaling laws. In his model, increasing scale (represented by the density of known points in a knowledge landscape) shrinks average gaps and improves mean quality in a roughly linear fashion. This is good news for Sutton’s thesis: more compute does mean better average performance. However, jaggedness persists and errors remain. Scaling raises average performance without eliminating surprises or long-tail failures. Gans frames AI adoption as an information problem: users care about local reliability (will the AI help me with my task?), but typically observe only coarse, global quality signals (benchmark scores). This mismatch creates real economic frictions. A legal assistant might trust an AI that performs brilliantly on 95% of contract reviews, only to be blindsided by a confidently wrong answer on a seemingly routine clause. The experienced errors, Gans shows, are amplified by what statisticians call the “inspection paradox”. Users encounter errors precisely in the gaps where they most need help. Gans’s 2026 paper does not directly cite or refute Sutton, but it can be read as exploring a structural limitation that persists even when following the Bitter Lesson path. Scaling works, but the economic benefits of scaling may be partially offset by the persistent unpredictability that scaling does not cure.  This limitation has practical implications for how businesses adopt AI: they cannot simply trust benchmark performance but must invest in human oversight and domain-specific testing. This also means that AI will not spell the end of human jobs.  Sutton was right about the direction, but we shouldn’t take his insight out of context. Scaling alone is not enough, and simply adding more scaling is unlikely to get us to superintelligence. Models still need human insight and structure to be maximally useful to companies. RLHF (Reinforcement Learning from Human Feedback), a training technique where human evaluators rate AI outputs to help the model learn which responses are helpful and safe, is an ingredient that injects human values into models. Earlier architectures didn’t become GPT-4 only by adding more data. Also, we cannot just “scale more” forever. Energy costs and data limits are real-world constraints. Thus, if AI is going to get much better it will need efficiency and algorithmic cleverness, not just brute force. Human insight has not faded into irrelevance yet. It has shifted from encoding intelligence directly to shaping, constraining, and steering scaled learning systems. Overall, let’s give Sutton due credit. Scaling works. But the efficiency of that scaling depends on human insight about how to structure and deploy these systems. Economists will recognize this as a familiar pattern: capital and labor remain complements, even when the capital is measured in GPUs and the labor involves designing loss functions. Gans’s work adds an important economic footnote: even as scaling improves average AI performance, the jagged, unpredictable nature of that performance creates real costs for adopters. Businesses and individuals must navigate a landscape where AI is simultaneously more capable and persistently unreliable in ways that are hard to anticipate. The economic returns to AI investment depend not just on raw capability but on developing institutions and complementary human expertise to manage jaggedness. The bitter lesson may be that pure scaling is powerful, but the sweet corollary is that human ingenuity is still a vital ingredient for progress in the future. [1] Compute, in AI research, is the total amount of computational power (typically measured in floating-point operations (FLOPs)) used to train or run a model. (0 COMMENTS)

How did an industry survive a technology that should have made it obsolete? Aled Maclean-Jones explains to EconTalk’s Russ Roberts how Japanese quartz watches nearly wiped out Swiss watchmaking with cheaper, more accurate alternatives–and how the Swiss redefined the value of a watch to recover market dominance. Maclean-Jones discusses the Japanese innovations that led to […] The post Seiko, Swatch, and the Swiss Watch Industry (with Aled Maclean-Jones) appeared first on Econlib.