Award-Winning AI Literature

This is a list of literary awards granted to works where AI was used (or was suspected) as part of the writing process. There are 30 cases categorized as follows: 4 confirmed AI use, 12 where AI was allowed from the start, 2 where organizers said AI was used, 6 suspected or disputed, 3 nominations and shortlists withdrawn, and 3 near-misses and eligibility cases.

CONFIRMED AI USE

• NOVEL: Sympathy Tower Tokyo by Rie Kudan — 170th Akutagawa Prize, 2024 Kudan initially estimated that about 5% of the novel involved ChatGPT, but later stated it was roughly one page out of 143. She used the AI's output specifically for replies written by an AI character within the novel, and the prize was maintained.
• SHORT STORY: The Land of Machine Memories by Shen Yang, writing as “QuartZen” — 5th Youth Popular Science and Science Fiction Competition, Second Prize, 2023 AI was used to generate the title, prose, illustrations, and pen name over 66 rounds of prompting. The experiment was hidden from the judges by the organizer; while often cited as a national prize, it was a broad association contest where 90 out of approximately 200 entries received awards.
• WEB NOVEL: Modest Skill “Tidying Up” Is the Strongest! — 18th AlphaPolis Fantasy Novel Awards, Grand Prize and Reader’s Choice, 2025 The author admitted that a large majority of the web novel was created using generative AI. Following this, AlphaPolis canceled the print and manga adaptations and updated its rules, though the award labels themselves were not formally revoked.
• MEMOIR: How to Win One Million Dollars and $#!T Glitter! by Luke Stoffel — IBPA Book Award, Benjamin Franklin Silver Medal for Neurodivergent Communities, 2026 Stoffel disclosed his use of ChatGPT for grammar and structural scaffolding in both the book and the award submission. Because IBPA guidelines did not prohibit AI assistance, the book received the Silver Medal.

AI WAS ALLOWED FROM THE START

• SHORT STORY: Are You There? by Kamome Ashizawa — 9th Nikkei Hoshi Shinichi Literary Award, Excellence Award, 2022 Ashizawa used AI to create 101 stories in three weeks, though the winning entry was mostly written personally with minor AI assistance. This marked the first AI-assisted story to win a Hoshi prize.
• SHORT STORY: The Last Painter by Toh Katagiri — 12th Nikkei Hoshi Shinichi Literary Award, Excellence Award, 2025 Organizers classified this as AI-assisted, though Katagiri reportedly told fellow finalists that every sentence was written personally.
• SHORT STORY: Genome Tower by Shamizui — 13th Nikkei Hoshi Shinichi Literary Award, Grand Prix, 2026 The disclosed AI use was minimal, apparently limited to inventing the name of a fictional disease, yet organizers still categorized it as an AI-assisted winner.
• SHORT STORY: Parallel Self-Mediation Co., Ltd. by En Shijima — 13th Nikkei Hoshi Shinichi Literary Award, Excellence Award, 2026 Shijima described the collaboration as feeling more like “choosing” than “writing,” suggesting a more generative process.
• SHORT STORY: Erinnerung by Rino Takinouchi — 13th Nikkei Hoshi Shinichi Literary Award, Excellence Award, 2026 This work involved extensive rewriting of ChatGPT prose, with the author correcting logic, pronouns, and dated phrases while adding original scientific material.
• ARTIST BOOKS: The Library of Nonhuman Books by Karen Ann Donnachie and Andy Simionato — Robert Coover Award, First Place, 2020 An AI system used computer vision and language processing to curate poetic combinations from physical books. One volume, Perception, also won Australia’s Cornish Family Prize.
• POETRY COLLECTION: Výsledky vzniku by Liza Gennart — Zlatá vlna, 2021 This collection, which won a prize for the best original Slovak poetry book, consisted of poems generated by GPT-2 trained on Slovak works and curated by humans.
• POETRY SERIES: ReRites by David Jhave Johnston — Electronic Literature Organization Robert Coover Award, 2022 The project consists of twelve books built from neural-network output that was selected and edited by the author.
• CHAPBOOK: A Black Story May Contain Sensitive Content by Lillian-Yvonne Bertram — DIAGRAM/New Michigan Press Chapbook Contest, Winner, 2023 Bertram used GPT-3 and a specialized model trained on Gwendolyn Brooks to generate machine responses for three sections of the book.
• SHORT STORY: Stargazers by Ruei-Jie Chang — UCSB Mind & Machine Intelligence Writing Contest, First Prize, 2024 Chang used ChatGPT for the initial seed and Claude for organization and translation before final revisions.
• SHORT STORY: 30–70 by Yana Vagner — Valentin Kataev Literary Prize, Winner, 2025 Vagner collaborated with Yandex’s AI assistant, Alice, for recurring climate reports within the story.
• GENERATIVE POEM: WORDS BEYOND WORDS by Sasha Stiles — Lumen Prize Literature & Poetry Award, 2025 The award was granted to a fine-tuned language model that generates a new poem each time the work is refreshed.

ORGANIZERS SAID AI WAS USED

• POEMS: Migrant Writers of Singapore poetry competition — Migrant Literature Festival, 2024 Organizers recalled the entire winners list after identifying three AI-generated entries using detection tools like Scribbr and QuillBot.
• POEMS: Poetry.com November and December contests — Two Monthly First Prizes, 2024 The site redistributed prizes after determining the original winners were created with AI assistance; the accused writers are no longer identified on the results pages.

SUSPECTED OR DISPUTED

• SHORT STORY: The Serpent in the Grove by Jamir Nazir — Commonwealth Short Story Prize, Overall Winner, 2026 Despite a detector labeling it 100% AI-generated, the Foundation accepted Nazir’s evidence of using speech-to-text software due to a medical condition and allowed him to keep the award.
• SHORT STORY: The Bastion’s Shadow by John Edward DeMicoli — Commonwealth Short Story Prize, Canada and Europe Winner, 2026 DeMicoli successfully disputed a 100% AI flag by producing handwritten drafts and research notes.
• SHORT STORY: Mehendi Nights by Sharon Aruparayil — Commonwealth Short Story Prize, Asia Winner, 2026 Aruparayil maintained her award after demonstrating how minor changes to local dialect words significantly altered detector results.
• SHORT STORY: Descend by Chanel Sutherland — Commonwealth Short Story Prize, Overall Winner, 2025 A detector labeled this story 88% AI a year after it won; Sutherland denied the claim, attributing the style to Vincentian oral storytelling.
• SHORT STORY: Back and Forth by Kavyta Kay — Harper’s Bazaar UK Short Story Competition, Winner, 2026 A 100% AI detector result for this story circulated online, though no formal investigation or revocation has been confirmed.
• POEM + UNKNOWN: Peter Doolan and an unnamed winner — Plaza Prizes, 2025 Both winners were disqualified based on detector flags, despite denials and Doolan’s claim that his poem was originally published in 2018.

NOMINATIONS AND SHORTLISTS WITHDRAWN

• COMICS ANTHOLOGY: Stardust the Super Wizard Anthology — Eisner Award for Best Anthology, Nominee, 2026 The anthology was withdrawn after it was discovered one page featured an AI-generated comic by a fictitious creator.
• ILLUSTRATED EDITION: Frankenstein, Clube de Literatura Clássica edition — Prêmio Jabuti Illustration Category, Finalist, 2023 The book was disqualified because it was fully AI-illustrated using Midjourney, which violated rules that had not previously considered generative art.
• ILLUSTRATED BOOK: Unnamed illustrated title — Ink Book Prize, Shortlisted, 2026 A title was removed from the shortlist following complaints regarding AI use in its illustrations.

NEAR-MISSES AND ELIGIBILITY CASES

• NOVELLAS + STORIES: Angel Train and Obligate Carnivore — Ockham New Zealand Book Awards, 2026 These books were briefly barred because of AI-generated covers, but were reinstated because the rule against them was implemented too late.
• SHORT STORY: Cage of Algorithms by Keiji Aono — 12th Nikkei Hoshi Shinichi Literary Award, Finalist, 2025 This story, which was 80% AI-generated, reached the final ten entries but did not win.
• SHORT STORY: The Day a Computer Writes a Novel — 3rd Nikkei Hoshi Shinichi Literary Award, First Round, 2016 In this pre-LLM precursor, software assembled prose from human-supplied components; it passed the first screening under rules that explicitly allowed nonhuman entrants.

China's Emerging AI Landscape and the Digital Marketplace

 Anxious Chinese students are trusting AI to help pick colleges and majors

Alibaba, ByteDance, Baidu, and Tencent guide hundreds of millions of high-school grads to optimize the high-stakes university match system.

By VIOLA ZHOU, 29 JULY 2026

• Chinese high school graduates are turning to free AI chatbots to navigate the college admission process.
• The automated tools are disrupting a costly private counseling market by offering highly utilitarian, job-focused major recommendations.
• Despite the convenience, students must cross-check AI suggestions to avoid database errors that could ruin their matching chances.

After taking part in China’s grueling national college entrance exam, Guo Xinyan has three weeks to make what could be the most important decision of her life: which universities and programs she should apply to. The 18-year-old turned to an AI chatbot for help.

On Alibaba’s chatbot Qwen, Guo entered her score, her rank in her home province of Shandong, her desired majors (law, accounting, or finance), her budget for tuition (moderate), and where she wants to live (coastal cities). The bot generated a report with dozens of choices.

“My family and I have no experience with college applications, so we had to rely on social media and AI,” Guo told Rest of World, adding that her parents had also asked ByteDance’s Doubao to suggest college programs for her. “There’s nothing in particular I want to study. We are just looking for good career prospects.”

Every year, around 10 million Chinese students take the college entrance exam known as the gaokao. Students wait for the results and then have just a few weeks to factor those scores into their application decisions. These grades determine whether a student can attend prestigious universities or pursue popular majors like business and engineering. Filling out these preferences is a high-stakes decision because, unlike in the U.S., Chinese universities have stringent requirements for transferring to a different field of study. Most students have only this one opportunity to decide on their future profession and survival in an increasingly competitive job market.

China’s Biggest AI Players

Many families pay private coaches to guide them through this decision, but AI companies are now offering the service for free to attract consumers to their chatbots. Since 2025, tech giants including Alibaba, ByteDance, Tencent, and Baidu have launched AI tools specializing in researching and recommending college programs.

These are typically agentic AI systems that analyze universities' historical admission cutoff scores. They ask for gaokao scores, career interests, Myers-Briggs Type Indicator (MBTI) personality assessments, tuition fees, and local climate. Eventually, the bots produce a list of suggested programs divided into “reach,” “safety,” and “backup” categories based on the student's admission probability.

Alibaba’s Qwen recruited 300 college search specialists to improve the system's reasoning, while Baidu promised human expert reviews for its recommendations. By July 8, Qwen had produced 23 million recommendation reports for gaokao takers. Tencent’s Yuanbao bot answered 200 million inquiries related to college admission by July 10, and Baidu reported 15 million users for its AI college advisory feature in late June.

While American families also use AI virtual consultants, Chinese high schools rarely provide career counseling. Before AI, families relied on guidance books, social media influencers, and independent tutors whose coaching sessions could cost thousands of dollars. Consultancy iiMedia estimated this advisory service industry at $160 million. Choosing the right major has become especially critical during a time of high youth unemployment and fears of AI automation. Paradoxically, AI itself has become a popular major choice as universities add programs like “embodied intelligence.”

“Don’t talk about dreams”

The best-known college admission adviser was Zhang Xuefeng, who had over 27 million followers on Douyin. He was known for bluntly telling working-class families which majors (like math and engineering) led to a bright future and which (like journalism and philosophy) did not. His company charged up to $2,700 for assistance.

After Zhang passed away in March, engineers built AI versions of him. One developer created a “Zhang Xuefeng skill” on GitHub, using his books and remarks to improve how AI agents perform tasks. Anyone can use this skill to make an AI agent speak in Zhang’s signature utilitarian manner. For example, the bot might tell an arts-interested student: “If your family has no money, don’t talk about dreams... Pick a major that puts food on the table, like computer science, education, or getting a government job.”

Xindy Lin, a designer in Fujian, built her own chatbot on OpenAI’s Codex to help her sister, combining the “Zhang Xuefeng skill” with current admission data and historical results. The AI suggested majors like supply-chain management and e-commerce marketing. Lin noted that individuals cannot digest the massive amount of information required in such a short time without help.

The Life Consequences of an AI Error

The college coaching industry has long been criticized for profiting from anxiety. Some human coaches, unfamiliar with the vast number of programs, now secretly use AI to generate recommendations. However, economist Ye Xiaoyang pointed out that rural students often lack the computer skills to effectively prompt a chatbot.

Ye launched a free AI coaching platform to help students explore career directions, noting that “a good AI system should be able to guide students to think for themselves.” He warned that AI tools could make mistakes, such as using outdated data or omitting strong options. Furthermore, if AI gives similar recommendations to students with similar scores, it could steer them toward the same programs, leading to match failures that force students into much lower-ranked schools.

Guo, the Shandong graduate, opted to cross-check her chatbot’s suggestions with official books and social networks like RedNote to investigate school dormitory conditions. “I still find real people’s experiences more trustworthy,” she said.

 Based on the provided source material, here is the full text of the article:

The Economics Job Market Since 1974

May 24, 2026 By mattsclancy

The number of positions advertised on the AEA’s Job Openings for Economists (JOE) board has declined in four of the six years since its 2018 peak. The 2025 academic year ended with 1,122 postings — near the COVID-year low of 1,074 recorded in 2020.

Annual postings, 2015–2025

From 2015 to 2018, postings ranged between 1,432 and 1,552, peaking at 1,552 in 2018. The 2020 academic year (the first to show the effects of the pandemic on hiring) dropped sharply to 1,074. The market partially recovered to 1,449–1,497 in 2021–2022 before declining again each year: 1,296 in 2023, 1,221 in 2024, and 1,122 in 2025. The 2025 value sits near the COVID trough and is roughly 28% below the 2018 peak.

Note: 2025 data runs August 2025–May 2026. June and July historically contribute fewer than five postings combined.

The longer view, 1974–2025

Splicing two data sources and indexing both to 2017 = 100 extends the picture back to 1974. The red series (Cawley 2018) shows the market growing from roughly 35 in 1974 — about a third of its 2017 level — to 100 by 2017, with a notable dip around 1980–1983, a plateau in the 1990s, and accelerating growth after 2012. The blue series (Goldsmith-Pinkham) picks up from 2017 and stands at 74 in 2025 — a 26% decline from the 2017 base, and fractionally above the COVID trough of 71 recorded in 2020.

Two counting methods, one index

The two series count jobs differently and cannot be directly compared in levels. The Cawley series draws on the AEA’s annual Report of the Director of JOE, which counted all unique jobs listed in a given calendar year; the 2017 figure is approximately 4,000. The Goldsmith-Pinkham series scrapes JOE Network listings and deduplicates by posting ID within an academic year (August–July); its 2017 count is approximately 1,516.

The roughly 2.6× gap reflects methodological differences — different time windows, deduplication rules, and the 2013 shift from monthly JOE issues to continuous posting — not a discrepancy in the underlying job market. Indexing both series to their respective 2017 values eliminates the level difference and leaves only the relative trend, which is what the chart shows.

Data

This post builds on two sources I am grateful to their authors for making available.

The historical series comes from John Cawley’s “A Guide and Advice for Economists on the U.S. Junior Academic Job Market” (2018 edition), specifically Figure 2, which plots unique JOE listings from 1974 to 2017. Cawley in turn draws on the Report of the Director of Job Openings for Economists compiled by John Siegfried and published annually in the AEA Papers and Proceedings.

The modern series comes from Paul Goldsmith-Pinkham’s JOE tracker, which scrapes JOE Network listings and makes annual Excel files publicly available on GitHub. The tracker covers 2015–2024 in his published files. For the 2025 academic year (August 2025–May 2026), Paul’s most recent export ran only through October 2025, so I downloaded fresh exports directly from the JOE Network website to fill in November 2025–May 2026, then merged and deduplicated the files by posting ID.

Source Coverage

Source	Coverage
Cawley (2018), Figure 2	1974–2017; hand-digitized; original data from Siegfried (2018), AER P&P
Goldsmith-Pinkham JOE tracker	2015–2024; scraped JOE Network listings, deduplicated by jp_id
JOE Network direct exports	2025 (Aug 2025–May 2026); merged with Goldsmith-Pinkham files

Reproducing this analysis

Code and data are in the economics-job-market repository.

Dependencies pip install pandas matplotlib openpyxl

---

mattsclancy Data and statistics work of public interest.

Is the Industrial Revolution a Precedent for Explosive AI Growth?

 

Is the Industrial Revolution a good precedent for explosive economic growth today?

July 27, 2026

One line of evidence that AI might lead to explosive economic growth is the precedent set by the Industrial Revolution. For hundreds of years — 1252 to 1652, to be precise — the compound annual growth rate of per capita real GDP in the UK was around 0.07%. It then began to accelerate, settling into a new compound rate of around 1.02% per year by 1850, which it held until 1913. In other words, growth accelerated by roughly 15 times before; the argument goes that this should make us humble about predicting it can’t happen again, and perhaps we should be open to accelerations of 10 times or more today.

I think this argument is overstated and the analogy between a 10× acceleration today and the acceleration that occurred during the Industrial Revolution is misleading. The goal of the first part of this post is to provide evidence for two claims:

1. At the outset of the Industrial Revolution, annual growth that was 10× the long-run average was relatively common.
2. In the contemporary world, annual growth that is 10× the long-run average for the world is much more rare.

The second part of this post characterizes the acceleration that occurred during the Industrial Revolution in terms of the standard deviation of year-to-year variation in growth rates. Applying the same approach to contemporary growth suggests that an IR-style acceleration would take growth in frontier economies to around 2.8% per year — meaningfully faster than today, but well below the 10× claim that is often advanced.

Why go through this exercise? A common reaction to claims that AI will lead to annual growth rates in excess of 20% per year is skepticism and incredulity — it would be so far outside historical experience. A common retort is that the same incredulity would have been wrong in the 1700s: had someone been told that future growth would be 10× the average and dismissed it, they would have made an error. The goal here is to rescue that initial reaction. A 10x acceleration today is not the same thing as a 10x acceleration in 1700. A person living in the 1700s would have been asked to envision good years becoming much more common — a rate they had already experienced many times. A person today is being asked to envision a qualitatively different kind of economic dynamics, one that falls several standard deviations outside the norm for the world today.

All estimates use the Maddison Project Database 2023, which reports GDP per capita in 2011 USD and population in thousands. The final section considers some objections to the relevancy of this analysis.

How common was 10× faster growth in the pre-IR UK?

We will start by establishing that it was quite common for growth to exceed 10× the long-run average in the UK prior to the Industrial Revolution. Over 1252–1652, the long-run average was 0.30% per year. The compound growth rate over this period — the rate at which wealth actually accumulated across generations — was around 0.07% per year. Roughly 10× this compound rate — around 0.66% per year — was exceeded in about 46% of years. (This figure is not sensitive to the exact window chosen: the compound growth rate ranges from 0.07% to 0.18% across plausible alternative start and end years, and the share of years exceeding 10× that rate ranges from about 40% to 50%.).

Economic statistics from hundreds of years in the past are highly unreliable, so we also consider 20-year compound average growth rates to eliminate year-to-year fluctuations. The mean compound rate across all 20-year windows is 0.06% per year, making the 10x threshold around 0.6%. About 17% of 20-year windows exceeded this threshold, implying that generation-long runs of 10× faster growth were not unheard of prior to the Industrial Revolution.

As a robustness check, we identified eight modern countries (Benin, Burundi, Chad, Haiti, Senegal, Sierra Leone, Togo, and Zimbabwe) with characteristics similar to the pre-IR UK regarding GDP per capita, population, and growth rates. Their average compound growth rate is 0.35%/yr, and annual growth rates exceeding 10× that — around 3.5% per year — were exceeded in about 19% of country-years.

Further analysis of the entire contemporary dataset shows that 10× faster growth is much more common for slow-growing countries. Among economies averaging 0–1%/yr, 94% experienced at least one 10× year between 1986 and 2022. This share falls to 22% for countries averaging 1–2%/yr and to essentially zero for those above 2%/yr.

How common is 10× faster growth today?

In Britain around 1700, a year of 0.66% growth — ten times the compound rate — was not unusual, as about one year in two already exceeded that threshold due to the volatility caused by harvests, wars, and disease. A permanent tenfold acceleration would have felt like a run of good harvests that simply kept coming.

Today, the story is very different. In the contemporary USA, 10× the compound growth rate implies annual growth of roughly 19% per year, a level Americans living today have never experienced. Looking at the whole world from 1950–2022, only 0.76% of country-years experienced growth in excess of 19%.

Of the 89 country-years that exceeded 20% growth, most fall into two categories: oil and resource windfalls (e.g., Kuwait, Libya, Equatorial Guinea) or post-conflict recovery (e.g., Lebanon, Bosnia, Rwanda, Iraq). These cases represent GDP returning to previous levels rather than an economy operating in a new gear. A third, smaller category involves early industrialization (e.g., South Korea in 1953, Botswana in the early 1970s), which might be a partial analogue for AI if it unlocks a fundamentally new production frontier.

In sum, a 10× growth acceleration today would be a qualitatively new way for the economy to operate, whereas, during the Industrial Revolution, it would have felt like an ordinary good year becoming more frequent.

The standard deviation approach

If we understand the Industrial Revolution as “more of the good years, fewer of the bad years,” we can use the standard deviation (SD) of annual growth rates as a ruler. The annual SD for various samples are:

• Pre-IR UK (1252–1652): 6.9%
• Modern analogue countries: 5.5%
• USA (1950–2022): 2.3%
• All countries (1950–2022): 6.2%

The Industrial Revolution represented a 15× acceleration (0.07% to 1.02%), which is an increase of about 0.17 standard deviations when measured against the modern analogue distribution. Applying this same 0.17 SD increase to the US frontier baseline of 1.9% would take growth to about 2.8% per year. In contrast, a genuine 10× increase today (to 18.9%) would require an increase of 3.1 to 7.5 standard deviations, depending on the sample used.

Some potential objections and replies

Objection 1: Pre-industrial and industrial growth have different mechanisms. One might argue that pre-industrial fluctuations (weather, war) are irrelevant to technological progress. However, the speed of economic change may be governed by factors beyond technology, such as property rights transfer, labor preferences, savings rates, and market size. These factors may limit growth speed in both eras.

Objection 2: Sensitivity to the definition of interval. Some argue that measuring growth over one-year intervals is arbitrary. However, shorter intervals are useful because failing to hit a growth rate on a short interval makes it less likely to be hit on a longer one, much like a runner's 400m time informs their potential for a mile race. The Industrial Revolution did not require unusually fast short-run rates by its own standards, but explosive growth today would require hitting short-run rates that are unusually fast by contemporary standards.

A closing observation

The agricultural revolution likely followed a similar pattern, where "good years" simply became more common relative to "bad years". In neither the Industrial nor the agricultural revolution did short-run growth rates likely surprise contemporary observers by dramatically accelerating relative to existing precedents.

Reproducing this analysis The full code and data are in the growth-acceleration repository. Data is sourced from the Maddison Project Database 2023. Dependencies include pandas, numpy, matplotlib, and openpyxl. The analysis is generated by running python3 analysis.py.

The Incalculable Scale of Life and the Macroeconomy

 The article titled "The world is bigger than you can imagine: Why it is difficult to evaluate an economy" by Scott Sumner explores the theme that the vastness and complexity of both human life and the macroeconomy make them nearly impossible to accurately evaluate.

The Scale of Human Life and Memory

The author begins with the philosophical claim that a human life is so vast that individuals cannot reasonably evaluate their own. He compares this to a visual blind spot where the brain fills in gaps, giving the false impression that a person is seeing their entire life when they are actually only seeing tiny fragments. For example, he recalls that at age nine, his life felt much richer and more significant in "utility" terms than at age seventy, yet he can only remember a few dozen events from that entire year.

Sumner argues that life is composed of a diverse "iceberg" of events—work, school, travel, illness, and committee meetings—most of which are forgotten or inaccessible until a sudden memory triggers the feeling of "sonder," the realization that life is much bigger than what one can recall. He also emphasizes the role of the narrative arts, suggesting that films can make life feel three times as long by providing experiences more engrossing than "real life". He questions whether common life evaluations focus too much on career and family while underrating the importance of hobbies, music, and "trivial" pursuits that may feel more real than actual acquaintances. Ultimately, he concludes that his own life evaluation changes based on his current mood, making any objective appraisal difficult.

Part 1: There’s a Great Deal of Ruin in a Nation

Turning to the macroeconomy, Sumner argues that bad analysis often stems from underestimating the economy's size and complexity. He cites an admission from The Economist regarding Donald Trump’s policies: while observers predicted that tariffs and immigration stops would be "unambiguously negative," the U.S. economy continued to grow faster than other G7 countries.

Sumner explains that while policies like the "MAGA tax" might have reduced growth by $300 billion, that amount is only about one percent of GDP and is easily obscured by offsetting factors like an AI boom or monetary policy changes. He notes that people often overestimate the impact of shocks to a single sector—such as the 2006 homebuilding collapse, which was only 6% of GDP—while underestimating nominal monetary shocks that affect all markets simultaneously. This complexity explains why doomsday scenarios regarding resource depletion or sanctions often fail to materialize.

Part 2: Policy Regimes are More Complex Than They Appear

Policy regimes are similarly vast and difficult to categorize. For instance, Singapore is ranked as one of the world's freest economies despite having highly interventionist elements. The U.S. federal government alone has over 1.1 million distinct regulatory restrictions filling nearly 200,000 pages, not counting state and local rules. This complexity allows people to engage in "motivated reasoning," finding specific examples to support their preferred policy positions.

To make sense of this, Sumner relies on natural experiments where economies "took off" after major reforms, including:

• West Germany after ending price controls in 1948.
• South Korea after removing trade barriers in 1964-65.
• China after improving rural property rights in 1979.
• Poland after privatization in 1990.
• India after deregulation in 1991.

He notes that while economic theory generally predicts that freer markets encourage wealth creation through competitive equilibrium and secure property rights, reality is multidimensional; for instance, Denmark combines free markets with a large welfare state.

Conclusion: The Universe of the Unknown

The author concludes with several other examples of things being larger than imagined:

• Physical Size: A safari covers only a tiny, two-dimensional thread of Tanzania, which is just one of Africa's 54 countries.
• Culture and Life: Cultures are often contradictory, and the diversity of biological ecosystems is beyond comprehension.
• History and Arts: History is constantly revised with new evidence, and there are far more "top" films and great writers than any one person can discover.

Sumner ends by reflecting on a quote from Scott Alexander regarding whether the increased quantity and variety of modern music compensates for a decreased profundity of experience. He concludes that we only know what we know and "vastly underestimate the universe of things that we don’t know".

I Am From Bosnia, Take Me to America

 The article "I Am From Bosnia, Take Me to America: Notes on the Western Balkans" was published by NOTCOMPETING on July 20, 2026. It chronicles a three-week journey through Slovenia, Croatia, Bosnia and Herzegovina, Serbia, Montenegro, and Albania, a region characterized by competing international influences via investment, migration, and real estate.

Minimum Viable History

After World War II, the region was united under Yugoslavia, comprising Slovenia, Croatia, Bosnia, Serbia (with Kosovo), Montenegro, and North Macedonia. National identities were primarily defined by religion: Slovenes and Croats were mostly Catholic, Serbs and Montenegrins mostly Orthodox, and Bosniaks Muslim. Following the fall of communism in 1991, the federation broke apart through a series of wars. While Slovenia gained independence quickly, the conflicts in Croatia and Bosnia were more protracted, with the Yugoslav army—dominated by Serbs—aiming to create a "Greater Serbia".

The 1995 Dayton Accords ended the fighting in Bosnia but divided the country, leaving roughly half as Republika Srpska, a Serb ethnostate that remains largely isolated from the rest of the nation. The final violent split occurred in 1999 in Kosovo, leading to NATO bombing and Kosovo's unilateral independence in 2008. Since then, Slovenia and Croatia have joined the EU and prospered, while Serbia has aligned more with Russia and Bosnia remains dysfunctional.

Bosnia

Bosnia is the most ethnically diverse of the former republics. During the author's visit, the 15-year-old song "I am from Bosnia, take me to America" saw a resurgence. While the U.S. is viewed as a symbol of wealth, there is resentment regarding the Dayton Accords, which established a tripartite government that many locals blame for state corruption and stagnation. In Sarajevo, the author observed the American Corner in the library and noted that while "heavy-handed" book displays do little, exchange programs (defunded in 2025) successfully generated pro-American sentiment.

The author highlighted the following observations in Bosnia:

• Migration: Young Bosniaks, including engineering and CS students, often plan to migrate to Germany or Dubai for better opportunities.
• Tourism: A wave of Arab and Turkish tourism in Sarajevo has brought investment but also sparked backlash regarding tourists seeking "white muslim wives".
• Ethnic Segregation: The city of Mostar remains deeply segregated. In 1993, Croatian militia destroyed the historic 16th-century bridge, and the Croatian side now displays militia emblems and a large cross on the ridge where artillery was once placed.
• Yugoslavia Nostalgia: Older generations often view the Yugoslav era as superior, citing higher rates of interethnic marriage, though studies suggest these claims may be exaggerated.

Serbia

Belgrade is described as the only "modern-looking global city" in the region. Serbia benefits from Chinese, Russian, and Turkish capital. Following the 2022 influx of Russians, some locals have reportedly refused to speak Russian, preferring English. Chinese influence is prominent, including a cultural center and hotel built on the site of the U.S. embassy bombed in 1999.

The Serbian rail network is noted for its "Chinese Train Station Design Language," but it has suffered from tragedy. In 2024, an awning collapse at the Novi Sad station killed 16 people, sparking massive anti-government protests.

Montenegro and Albania

Montenegro is on track to join the EU in two years and has already adopted the Euro. The capital, Podgorica, is described as boring but possessing good urbanism. The author noted a prevalence of cheap Chinese consumer goods and a lack of functioning railway crossing gates, leading to common collisions.

Albania is characterized by a dual currency regime (Lek and Euro) and a complete lack of nostalgia for its communist past, which was uniquely isolated. In Tirana, the "attract-tourists plan" involves building "super-weird skyscrapers," such as one shaped like the head of national hero Skanderbeg. The author also witnessed the "Flamingo Revolution" protests, which were sparked by a luxury development deal involving Jared Kushner and Ivanka Trump.

Conclusion

The author concludes that the Balkans, despite recent ethnic violence and high gun ownership, is one of the safest regions for walking. However, regional transit is unreliable, and digital tools like Google Maps are often inaccurate regarding local schedules and station closures. Ultimately, for many young people in the region, the primary "democratic" goal of EU accession is the ability to leave for higher-income countries.

The Hayekian Triangle: Why Liberty Is Not Conservatism

 Based on the provided source, here is the full text of the article "On Being A Conservative: Lessons from Friedrich Hayek" by Cass Sunstein:

---

On Being A Conservative

Lessons from Friedrich Hayek

CASS SUNSTEIN JUL 20, 2026

1 Friedrich Hayek is a great hero to conservatives, and rightly so. He revered traditions, despised socialism, and favored free markets. Hayek helps to define modern conservatism. And yet one of his most interesting and vivid essays has this title: Why I Am Not A Conservative.

Well, why wasn’t he? And what might we learn from his clear desire to distinguish himself from conservatives?

2 In the relevant essay, written in the late 1950s and published in 1960, Hayek rejected the idea that socialists are on the left, conservatives on the right, and liberals in the middle. He didn’t like the idea of some kind of line from left to right. He wrote: “Nothing could be more misleading. If we want a diagram, it would be more appropriate to arrange them in a triangle with the conservatives occupying one corner, with the socialists pulling toward the second and the liberals toward the third.” Three corners, not a straight line.

3 Hayek had plenty of nice things to say about conservatives and conservatism. His essay is written in sadness, not in anger. (It was apparently started as a paper delivered to the Mount Pelerin Society in 1957, sharply critical of the traditionalism associated with Russell Kirk, who apparently delivered an extemporaneous response at the time. That must have been quite an occasion.)

In tribute to conservatism, Hayek said this: “To their loving and reverential study of the value of grown institutions we owe (at least outside the field of economics) some profound insights which are real contributions to our understanding of a free society.” Hayek shared with conservatives that reverence for grown institutions, the beneficial products of social processes, not of any individual designer.

With that justified reverence, conservatives “did show an understanding of the meaning of spontaneously grown institutions such as language, law, morals, and conventions that anticipated modern scientific approaches and from which the liberals might have profited.” This understanding was central to Hayek’s own thinking; it was even defining. Liberalism, as Hayek understood and championed it, shares that reverence, and hence is deferential to traditions. It never sneers at them. This is a close connection here between Hayek and Burke (and on Hayek’s account, Adam Smith as well).

Divergent attitudes toward traditions help to explain Hayek’s split from John Stuart Mill (and from other liberals). In Hayek’s telling, Mill was a rationalist who thought that human reason could be used to challenge traditions. Hayek did not like that at all. He thought that it was a “fatal conceit.” Hayek well knew that many contemporary liberals follow Mill, and he lamented that. Hayek’s preferred form of liberalism, which he believed to be the truest form, treasures not arrogant reason, but invisible hands, undesigned orders, the common law, the rule of law, and customs.

Designs without designers—they are fundamental to Hayekian liberalism and to conservatism alike. Hayek did not want to freeze any status quo, but like conservatives, he tended to favor incrementalism rather than large-scale change. He greatly admired Burke. “There would not be much to object to if the conservatives merely disliked too rapid change in institutions and public policy; here the case for caution and slow process is indeed strong. “

4 In 1955, William F. Buckley, Jr., a defining American conservative, wrote that his new magazine, and the conservative movement as he would like it to be, “stands athwart history, yelling Stop.” That’s a great line. It’s exactly what Hayek rejected. I don’t know if Hayek knew of the line, but I suspect so. You can see his essay as a direct response to it.

5 Hayek identified the following as “the decisive objection to any conservatism which deserves to be called such”: “by its very nature it cannot offer an alternative to the direction in which we are moving.”

Sure, “It may succeed by its resistance to current tendencies in slowing down undesirable developments, but, since it does not indicate another direction, it cannot prevent their continuance. It has, for this reason, invariably been the fate of conservatism to be dragged along a path not of its own choosing.”

For Hayek: Ugh. We might say that in Hayek’s view, conservatism lacks a theory. It’s mostly a red light. By contrast: Focused on freedom, spontaneous orders, traditions, and free markets, Hayekian liberalism has a theory. It tells us where to go. It’s a path.

6 Hayek complained that conservatives want to freeze things. They do not want to allow free markets to do their work, or to permit freedom to have its way, or to see what surprises the future, governed by liberal principles, has in store for us.

In particular, Hayek lamented that conservatives “are inclined to use the powers of government to prevent change or to limit its rate to whatever appeals to the more timid mind. In looking forward, they lack the faith in the spontaneous forces of adjustment which makes the liberal accept changes without apprehension, even though he does not know how the necessary adaptations will be brought about. It is, indeed, part of the liberal attitude to assume that, especially in the economic field, the self-regulating forces of the market will somehow bring about the required adjustments to new conditions, although no one can foretell how they will do this in a particular instance.”

Hayek much liked entrepreneurship and innovation. He thought that the future could not be predicted—and that we should be open to and delighted by surprise, so long as it came from exercises of freedom.

Here is where Hayek got tougher on conservatives. In his account, they do not like freedom nearly enough. They are open to the exercise of arbitrary power. They might even welcome it. They are fine with coercion, so long as they are the ones who are behind it.

He objected in particular to “the characteristic complacency of the conservative toward the action of established authority and his prime concern that this authority be not weakened rather than that its power be kept within bounds. This is difficult to reconcile with the preservation of liberty. In general, it can probably be said that the conservative does not object to coercion or arbitrary power so long as it is used for what he regards as the right purposes” (emphasis added).

7 Hayek urged that in this respect, the conservative was “[l]ike the socialist.” (For Hayek, those are exceptionally strong words.) The reason is that “he is less concerned with the problem of how the powers of government should be limited than with that of who wields them; and, like the socialist, he regards himself as entitled to force the value he holds on other people.”

Coercion, then, was one of Hayek’s key concerns. “[T]o the liberal neither moral nor religious ideals are proper objects of coercion, while both conservatives and socialists recognize no such limits. I sometimes feel that the most conspicuous attribute of liberalism that distinguishes it as much from conservatism as from socialism is the view that moral beliefs concerning matters of conduct which do not directly interfere with the protected sphere of other persons do not justify coercion. This may also explain why it seems to be so much easier for the repentant socialist to find a new spiritual home in the conservative fold than in the liberal.”

That is especially interesting (and the last sentence is shrewd). Hayek did not endorse Mill’s Harm Principle, which he thought potentially destructive of liberty. But he came pretty close to that principle, certainly insofar as he wanted to limit coercion.

8 There is a further point here about tolerance—not a great word, but a word of Hayek’s time. A better term might be: mutual forbearance.

Hayek objected: “The typical conservative is indeed usually a man of very strong moral convictions. What I mean is that he has no political principles which enable him to work with people whose moral values differ from his own for a political order in which both can obey their convictions.”

Hayek tended to agree with conservatives on those particular values. He shared many of them. (How many? Good question.) But still: “The acceptance of such principles means that we agree to tolerate much that we dislike. There are many values of the conservative which appeal to me more than those of the socialists; yet for a liberal the importance he personally attaches to specific goals is no sufficient justification for forcing others to serve them.”

Here is Hayek’s central claim: “To live and work successfully with others requires more than faithfulness to one’s concrete aims. It requires an intellectual commitment to a type of order in which, even on issues which to one are fundamental, others are allowed to pursue different ends.”

You can see this as a pragmatic point about what is necessary to enable different people to live together (a modus vivendi). Hayek did make that pragmatic point. But you can also see it as a point about liberty, and for Hayek, that was the more fundamental one.

9 Hayek also offered a vigorous claim about novelty, and the right attitude toward it. Hayek urged “Though the liberal certainly does not regard all change as progress, he does regard the advance of knowledge as one of the chief aims of human effort and expects from it the gradual solution of such problems and difficulties as we can hope to solve. Without preferring the new merely because it is new, the liberal is aware that it is of the essence of human achievement that it produces something new; and he is prepared to come to terms with new knowledge, whether he likes its immediate effects or not.”

Thus "the most objectionable feature of the conservative attitude is its propensity to reject well-substantiated new knowledge because it dislikes some of the consequences which seem to follow from it—or, to put it bluntly, its obscurantism. I will not deny that scientists as much as others are given to fads and fashions and that we have much reason to be cautious in accepting the conclusions that they draw from their latest theories. But the reasons for our reluctance must themselves be rational and must be kept separate from our regret that the new theories upset our cherished beliefs” (emphasis added).

This is an argument against motivated reasoning, and a suggestion that conservatives tend to succumb to it. (Not incidentally, Hayek also had some things to say against nationalism, and the conservative tendency to favor it.)

10 Still: In important respects, and notwithstanding his title, Hayek really was a conservative. He emphasized throughout his life how much we do not know, and how much we benefit from traditions, practices, and institutions that no human being designed. On that count, consider this crucial passage:

”What I have described as the liberal position shares with conservatism a distrust of reason to the extent that the liberal is very much aware that we do not know all the answers and that he is not sure that the answers he has are certainly the right ones or even that we can find all the answers. He also does not disdain to seek assistance from whatever non-rational institutions or habits have proved their worth.”

Hayekian liberals and conservatives are humble; they know how much we do not know. Like the conservative, the Hayekian liberal welcomes institutions and habits that have proved their worth, even if we do not really understand them.

Hayek had some things to say about religion here:

“Unlike the rationalism of the French Revolution, true liberalism has no quarrel with religion, and I can only deplore the militant and essentially illiberal antireligionism which animated so much of nineteenth-century Continental liberalism. That this is not essential to liberalism is clearly shown by its English ancestors, the Old Whigs, who, if anything, were much too closely allied with a particular religious belief. What distinguishes the liberal from the conservative here is that, however profound his own spiritual beliefs, he will never regard himself as entitled to impose them on others and that for him the spiritual and the temporal are different spheres which ought not to be confused.”

Okay then.

11 We can easily see why Hayek has inspired, and continues to inspire, conservatives and conservatism all over the world. But he thought it important to say that he was not a conservative. He really did not like the idea of standing athwart history, yelling stop.

He had an account of liberty, associated with his understanding of markets, traditions, and dispersed knowledge. Sure, he agreed that we need a “brake on the vehicle of progress.” But what he added, with some passion, was this: “I personally cannot be content with simply helping to apply the brake. What the liberal must ask, first of all, is not how fast or how far we should move, but where we should move.”

---

Cass’s Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

 

Nineteen thoughts on AI and Europe

Protectionism and industrial policy won't fix our problems

PIETER GARICANO AND SIMON GRIMM JUN 15, 2026

After the events of last weekend, here are some thoughts on where Europe is with AI. This post is coauthored with my colleague Simon Grimm.

1. The reporting so far seems to indicate that foreign citizens were banned from using Anthropic’s model Fable because the Trump administration needed a way to shut down access to the model for everyone – including American citizens – and export controls were the easiest way to do so. As American model regulations become more sophisticated, we should expect more narrowly targeted restrictions.

2. American control over models is no greater than American control over consumer and enterprise software. That arrangement has not worked so badly for Europe: Google earned a lot of money over the last twenty years, but European consumers and business enjoyed Maps, Drive, Gmail, YouTube, and Search. Artificial intelligence may be quantitatively much more important to the economy, but qualitatively the dependence is no different.

3. Economic convergence across history has always depended on the importation and adoption of foreign technology, not its rejection. Mexico, Argentina, and Brazil lost much of the twentieth century pursuing import substitution; Singapore, Taiwan, and South Korea have become rich by integrating themselves into global manufacturing. Even the Chinese have gritted their teeth for decades and accepted dependence on the Americans.

4. Dependence on American technology does not imply permanent vassalage. It seems unlikely that AI will be the only technology, or Anthropic the only company. Superhuman intelligence may make progress in plasma physics, but that will create a need for companies that build fusion reactors. Those could be German as much as they could be American.

5. Becoming major users of American models makes restrictions less likely because of the cost to American companies of losing European customers. Nvidia is lobbying the Trump administration to allow it to sell chips to China because of its importance as a market.

6. Champions of a European AI model should ask themselves if a European effort would be more effective than Meta, which this year will spend more on chips ($125 billion) than Germany spends on defense ($114 billion) and offer salaries of over $100 million to attract the best researchers, and is still failing to catch up. Elon Musk tried and failed to build a good AI model.

7. European governments tried to build their own Google in 2005, a Franco-German alternative called Quaero. After Franco-German disagreements, the Germans spun off Theseus. Both Quaero and Theseus were discontinued by 2013.

8. Fable teaches us something general about the ‘intelligence commoditization’ thesis and the ability of worse models (whether Chinese or European) to substitute for the frontier ones. Having experienced a better model, Opus 4.8 is painful to use even on tasks where it previously felt helpful.

9. New models can do things that old models could not do: no amount of GPT-4 prompts could solve an Erdos problem. But the preference for better models is also because returns in life are non-linear: competition between firms, applications for jobs, financial trading, and legal work share the characteristic that a tiny increase in performance can deliver enormous increases in payoffs.

10. In a world of highly skewed returns to capabilities, the second mover earns nothing. The revenue that Mistral reported earlier this year is less than one percent of Anthropic’s today. Any European sovereign AI effort would have to invest billions to catch up without making any money along the way.

11. The response to the revenue problem should not be the Buy European Tech Act, if ‘Buy European’ means buying worse models, baking in security vulnerabilities and less useful responses.

12. The above is all true if we are in a world of increasing returns to model capabilities, where companies with better models can use them internally – and the revenue they earn from them – to pull away from weaker labs. For now, this seems true: the gap between Google DeepMind on the one hand and Anthropic and OpenAI on the other has increased over the last year. In the increasing returns world, it is enormously hard for any company, whether European, American, or Chinese, to catch up with the current leaders.

13. This world of increasing returns is not guaranteed. Many smart people don’t believe in it: Google is selling millions of chips to Anthropic rather than giving them to DeepMind, because Anthropic’s willingness to pay for the chips — their belief in near-term progress — is currently greater than Google’s. If progress did slow down, other providers, including European ones, would be able to catch up.

14. We interviewed a number of European data centers over the last two months; all said that European data centers would not be built at scale without subsidies. Part of this is due to Europe’s high power prices and population density. In the areas where these problems are less pressing – Scandinavia and Iberia – smaller data centers are being built by commercial providers.

15. A reason why American companies don’t want to build their biggest data centers in Europe is political risk. The companies worry that the European Union will change its rules on copyright, fine them, and come up with extractive taxes once the investments are committed.

16. It may seem implausible for a democracy to commit economic self-harm on that scale. But, through the energy transition, Europe has signaled that governments will tolerate destroying industries as important as the German car and chemical industries. It is not clear how you rebuild this trust.

17. If it does come down to a transatlantic stand-off for model access, the upside of combining forces in the European Union may be lower than the downside of increased coordination costs. There are many members of the EU with conflicted interests who might yield to pressure: Eastern Europe needs the American military presence in Europe to deter Russia. Having countries like these involved in decisions may diminish total European leverage.

18. It would be helpful for national leaders in Europe to think about AI from the perspective of their national interest. The member states are the ones that control the factors that matter: labor markets, taxes, permitting, and energy. The Netherlands could choose to loosen the planning laws that are slowing the expansion of ASML; Germany could lower its electricity costs enormously with a nuclear restart. They have far more control over their individual destinies than does the European Commission.

 INNOVATION

The Great AI Divide: Navigating U.S. and Chinese dominance

At a Rest of World event during New York Tech Week, we explored the challenges and possible solutions to the dominance of American and Chinese AI companies.

By RINA CHANDRAN 9 JUNE 2026

Since the launch of ChatGPT at the end of 2022, AI has quickly changed how we work, how we learn, how we love, how we heal. It has also made a handful of companies such as Nvidia, Anthropic and OpenAI very powerful, and put the U.S. and China far ahead of every other country. What does this mean for everyone else? Is there a way to ensure a more equitable future?

Last week, Rest of World hosted an event titled “The Great AI Divide,” as part of New York Tech Week. We asked three experts to weigh in on these questions: Sam Winter-Levy, a fellow at the Carnegie Endowment for International Peace; Aditya Vashistha, an assistant professor at Cornell University, where he leads the Cornell Global AI Initiative; and Peter Micek, general counsel and United Nations policy manager at digital rights group Access Now.

Here is a summary of the conversation, edited for brevity and clarity.

Sam, is the AI race all but over for everyone? What happens when countries are subject to the whims of Washington and Beijing?

It’s fair to say that AI is primarily a two-horse race. The U.S. and China control 90% of global computing power, attract between 70% and 80% of global investment in AI, and have huge concentrations of talented AI researchers. That creates a world in which other countries are dependent on the U.S. and China for access to AI systems. And both the U.S. and China have shown a willingness to use that access for leverage. That puts most of the Global South in an uncomfortable position. And the middle powers remain exposed to the disruptions that AI could cause, even if they don’t necessarily capture the benefit. So they’re still exposed to job-related disruption, the social effects of AI systems. That’s the pessimistic story about how the rest of the world could be left behind.

Three trends make this pessimistic story particularly likely right now.

“The middle powers remain exposed to the disruptions that AI could cause, even if they don’t necessarily capture the benefit.” — Sam Winter-Levy, fellow, Carnegie Endowment for International Peace

The first is that a lot of the frontier developers in the U.S. are switching to a more managed-access approach to their technology. Anthropic’s Mythos model is a good example of this. They’re rolling the model out to small groups of companies that they select, at least initially. That puts other countries in an uncomfortable position where U.S. companies pick who gets access to these systems. Second, we’re now in a world where there are quite severe compute constraints. Demand for these models is outstripping the compute that the companies have. Again, companies are rationing who can use those systems. And finally, we’re starting to see the U.S. government and the Chinese government playing a more assertive role in who can have access to these systems. So those trends together could lead to a situation where a small handful of very wealthy companies, very wealthy countries have control over who has access to this technology.

Aditya, you design, build, and evaluate AI for marginalized communities. What are the risks of having only American and Chinese AI systems?

AI technologies today are designed by and for WEIRD societies — Western, educated, industrialized, rich, and democratic — which represent only 14% or 15% of the world’s population. What about the 85% and how they are represented in the AI systems? Back when ChatGPT was launched, and you would ask, “What does a Muslim man look like?” there was a homogenized view of that — someone who is wearing a headscarf and has a white beard — while a Hindu man would be someone who’s wearing saffron. There were all sorts of problematic issues for any kind of marginalized population, like people with disabilities. Many of these AI models could not even show what these looked like.

There have been many advancements, more safety integrated into these models. But these models continue to have many of these biases — religious, linguistic, identity, and so on. These biases are not what they were a couple of years ago, but they are still deeply rooted in these models. So as we think about AI futures for the rest of the world, we need to think about whose values, whose voices, whose languages, and whose cultures are represented in these models. The layers which we have now work for only a minority of the world’s population, and many of the benchmarks for safety do not even account for ableism, when we have 1 billion people in the world with disability. So some of these biases are not just towards people living in the Global South, but about marginalized voices. If you look at preference data sets, many of these do not exist for the majority of countries in the Global South. Many are in English, and not in other languages. If we do not take into account these biases, these languages and cultures, then we are continuing to design our AI technologies to work efficiently only for a sliver of the population.

---

90%

The amount of compute that the U.S. and China control together.

---

Peter, the first U.N. General Assembly resolution on AI was about the need for safe, secure and trustworthy AI systems, and the application of human rights to AI. There was also a resolution on international cooperation in AI capacity-building. Where do we stand on these now?

The first two resolutions on AI were led by the U.S. and China, respectively. And unlike most resolutions, they each went it alone. There was some good language in the Biden era U.S.-led resolution, applying human rights to the entire life cycle of AI. It talked through development, model building, to feedback and implementation, and ensuring that the entire bevy of human rights — the last 60–70 years of progress — is affirmed and does apply to the AI space. That’s a really positive assertion, along with the assertion that certain applications of AI are impossible to reconcile with the human rights framework, and that leaves a lot of questions as to what those applications are.

“We have seen the application of AI to military contexts as well as humanitarian, and that’s putting machine learning systems in direct control over life or death decisions.” — Peter Micek, general counsel, Access Now

In the last year or two, we have seen the application of AI to military contexts as well as humanitarian, and that’s putting machine learning systems in direct control over life or death decisions. The idea that a human ultimately pushes the button, but depends on libraries of targets and scaling and selection that the machine gives — even with a human in a loop, that’s simply not enough.

Sam, in his speech at Davos, Mark Carney said, “Great powers have begun using economic integration as weapons, tariffs as leverage, financial infrastructure as coercion, supply chains as vulnerabilities to be exploited.” He called for countries in between to combine to create a third path. Is there a way to do that for AI?

The problem with the Carney vision is that … the correct analysis of the geopolitical situation runs up against this technological moment where these two countries, the U.S. and China, really are dominant. So for middle powers, there are a few options. One, you form some sort of middle-power coalition, you run French models or Canadian models on data centers in Australia, and you pool resources to try and catch up.

A second approach is trying to build your own sovereign models. The UAE, India talk about this a lot. I think it’s very difficult because of the scale and the amount of money needed. They’re still using U.S. chips designed by Nvidia in data centers that are serviced by U.S. companies, so it doesn’t get rid of all these vulnerabilities.

The third approach is to essentially bandwagon, or sidle up to one of the great powers, whether it’s the U.S. or China, and build a very close relationship to make sure to have access to the technology. But that does expose you to the U.S. or China. If they don’t like your policies or something, they have a lot of leverage over you.

The best option is some version of the third approach, where you get access to U.S. technology or Chinese technology, and you are well aware of the vulnerabilities. You need to bargain very hard with the great powers over getting durable guarantees. It would mean finding sources of leverage — whether that’s in the AI supply chain, in the semiconductor supply chain — and using that as leverage to say, we’ll give you access to our raw materials, our critical minerals, our robotics capabilities, our chip design capabilities. And in exchange, we want guarantees that you’re going to keep giving us access to the best models.

Sam, what leverage do middle powers have? India has been a massive source of data and talent for American companies, for example.

To be competitive in AI, you need access to tons of researchers, you need access to energy, you need access to chips, and access to data. So countries that have those things and can offer them to the U.S. — not as a gift, but in exchange for things in return — have a relatively strong hand to play. The Netherlands, Taiwan, Japan, and South Korea play key roles in the chip supply chain. That gives them quite a lot of leverage. India obviously has huge amounts of data, even Ukraine has huge amounts of data from the battlefield that American companies are desperate for. Talent becomes harder to use as leverage. Energy is another thing that a lot of countries can use as leverage. The key thing they need to do is figure out what they have that the U.S. or China really need, and use that as leverage — whether it’s upstream in the supply chain or downstream in actual deployments — to bargain harder for access to the frontier models.

Peter, RightsCon this year was cancelled because of pressure from China on the Zambian government. Is it far-fetched to imagine China or the U.S. dictating terms to a country that it sells AI systems to?

“As we think about AI futures for the rest of the world, we need to think about whose values, whose voices, whose languages, and whose cultures are represented in these models.” — Aditya Vashistha, assistant professor, Cornell University

We have direct experience with being used as a ping-pong ball between great powers, and experiencing the pressure that one of them can bring on one of these smaller states. We were supposed to be in Zambia last month for the conference with around 2,600 folks in person, thousands more online — it’s the world’s biggest conference on the future of the internet, going deep into human rights and into labor and the environment.

Unfortunately, about five-six days before the conference was supposed to start, we were told that the Chinese government had got wind that there would be Taiwanese participants, and that more time was needed for security clearances and other diplomatic overtures. Ultimately, we learned that they needed full moderation of online and offline panels, for mention of LGBTQ issues, denial of access for Taiwanese participants, and adherence to the One China policy. Zambian authorities gave in to this pressure, but it is not the only example that we’ve seen. The U.S. is also exerting pressure on civil society right now in the digital rights space, canceling visas, sanctioning individuals, researchers, and activists.

Aditya, at the India AI summit in February, there were high expectations that it could offer a third way to challenge how AI power is distributed. Did that happen?

It’s too soon to say. What the summit accomplished was getting a lot of people in the room — government officials, heads of state, CEOs and CTOs of big tech companies, small tech companies, nonprofits, civil society organizations — to talk about the Global South. It is something we should have been doing for a very, very long time, and this was the big achievement of the summit. Just having all these people in the room talking about AI safety, security, fairness, governance, and other challenges which come with designing, building and evaluating AI technologies, forming partnerships and collaborations was a great success.

Sam, for countries that don’t want to align with the U.S. or China, what’s the outlook?

There’s one conversation that’s taking place in Silicon Valley, in Washington, and in some capitals around the world, where there is this belief that you need access to frontier models. You need access to the best large language models that are coming out of Anthropic, OpenAI, and that these models are going to be absolutely critical to national security, and the economy. There’s also this conversation in the middle powers, where there’s a little bit more optimism that you can use small language models. That you can focus on use cases, your own open-source, and you don’t need access to the very best models, you don’t need access to a huge infrastructure, or investments.

We don’t know exactly which of those technological paradigms is right. For a lot of critical national security use cases, critical economic use cases, some cybersecurity, if you have a model that is okay but not great, and you’re going up against a state that has access to the best models, you’re just going to be totally outcompeted. In parts of finance, scientific R&D, if you’re not able to do research as well as the other countries, you are going to lose a lot of advantages. But it might be the case that good enough open-source models that aren’t quite at the front end will work for most use cases. And if that’s true, then the world is much less bleak for a lot of middle powers.

Access to the best models and entrepreneurial culture will unlock huge possibilities for populations around the world to create businesses, to build products that they would not have been able to in the past. With just a handful of people in India or the Philippines, you could build things that in the past you needed vastly more resources to do. So there will be a hugely democratizing effect if you have access to systems in a safe and responsible way, and you can get tremendous economic benefits. But that’s contingent on having access. And that access is not guaranteed.

Rina Chandran is a deputy editor at Rest of World, based in San Francisco.

Gulf Money and the SpaceX IPO: Financing the AI Boom

 GLOBAL DISPATCH

Saudi Arabia and the UAE are funding America’s AI boom — and getting data centers in return.

What the SpaceX IPO reveals about Gulf money in AI

By INDRANIL GHOSH Indranil Ghosh is the Middle East and Africa Editor at Rest of World, based in Abu Dhabi.

12 JUNE 2026

GLOBAL DISPATCH This essay was first published in our Global Dispatch newsletter. Sign up here to get it straight to your inbox.

Poring over the SpaceX IPO prospectus, a recurring theme surfaces: the massive, quiet influence of Middle Eastern finance in the most ambitious IPO in history. Sovereign wealth funds of Saudi Arabia and the United Arab Emirates, their AI subsidiaries, and the technology companies building data centers as part of these deals were all in the document.

SpaceX lists on Nasdaq June 12 at a $1.75 trillion valuation. The S-1, as the IPO filing is known, shows Elon Musk’s rocket and satellite company is looking to sell up to $75 billion in shares. Saudi Arabia’s Public Investment Fund alone is in talks to put in $5 billion.

ChatGPT, Claude, and Grok, three of the most widely used AI tools in the U.S., are all partly funded by Middle Eastern governments. For the millions of U.S. professionals who open these tools at work, the source of that money matters.

Unlike venture capital, sovereign wealth comes with conditions, and those conditions almost always involve building AI infrastructure on the investing country’s own soil. Those deals are putting AI data centers in the Middle East, not in the U.S.

Where the deals lead: Deal by deal, capital is flowing from the Middle East to Silicon Valley, and computing power is getting built at the other end, on sovereign soil, under the watch of the governments writing the checks. Data center jobs, tax revenue, and the economic activity that comes with building AI infrastructure are going to the Middle East instead of to communities in the U.S.

The prospectus also shows how strong the ties between individual Gulf investors and Musk’s empire have grown over the past 15 years.

• MGX (UAE) has a stake in OpenAI, Anthropic, and xAI/SpaceX.
• G42 is now building a data center campus in Abu Dhabi.
• Humain (Saudi) put $3 billion into xAI earlier this year. A joint AI data center in Saudi Arabia came with the deal.
• Microsoft committed $15.2 billion for data centers in the UAE through G42 subsidiary Khazna.

In 2011, Prince Alwaleed bin Talal, a Saudi billionaire, put $300 million into X (then Twitter). When Musk bought the company in 2022, Alwaleed rolled his stake in rather than selling. When Musk folded X into xAI and merged xAI with SpaceX, that stake became shares in the rocket company.

Kingdom Holding, Alwaleed’s investment firm, now values the position at $10.6 billion at the expected IPO price. A social media bet placed 15 years ago has multiplied many times over, landing in a spacecraft business.

Until this month, most of these arrangements were private. The SpaceX prospectus is the first public filing to put them on the record.

World-First Clinical Trial for Cellular Reprogramming and Rejuvenation

 The provided source outlines several key objectives for the world-first clinical trial of cellular reprogramming, ranging from immediate medical goals to long-term scientific aspirations.

Immediate Clinical and Therapeutic Objectives

The primary clinical objective of this landmark trial is to treat specific diseases of the eye, specifically a form of glaucoma that can cause blindness. The trial aims to:

• Regenerate neurons in the optic nerve: These neurons, which connect the eye to the brain, do not normally regenerate in adults.
• Restore vision loss: By coaxing aged cells to take on a "younger identity," researchers hope to reverse vision damage, a result previously seen in animal studies.
• Expand to other conditions: The trial eventually plans to include participants with NAION, a severe, acute condition that also causes nerve damage in the eye.

Technical and Scientific Objectives

The trial is designed to test a novel gene therapy approach known as partial reprogramming. This involves:

• Activating three specific genes: These genes are used to nudge adult cells "back in time" to restore youthful features.
• Maintaining cell identity: A critical objective is to ensure cells behave as if they are young without pushing them so far back that they lose their specialized function or identity entirely.
• Precise control: The system is designed for high control, allowing researchers to switch the genes on or off using an antibiotic (doxycycline) to ensure expression does not last longer than necessary to rejuvenate the cells.

Safety Objectives

Given that this is a world-first human trial, testing safety is a paramount objective. The stakes are high because:

• Cancer prevention: There are significant concerns that reprogramming could tip cells into a cancerous state.
• Minimizing risk: The eye was chosen as the initial site because the potential for "life-threatening" or "catastrophic" side effects is lower than in other organs.

The Larger Context and Long-Term Goals

While the current trial is localized to the eye, it sits within a much larger vision for the future of medicine:

• Disease-by-disease approach: The sponsoring company, Life Biosciences, aims to tackle "one age-related disease at a time," having already studied the approach in animal models of liver disease.
• Organ rejuvenation: Some scientists argue that successful partial reprogramming could eventually be used to rejuvenate entire old organs.
• Whole-body rejuvenation: Although not the current focus, the ultimate "someday" goal mentioned is the potential for whole-body rejuvenation and enhanced longevity.

The logistics of the world's first cellular reprogramming clinical trial involve specific delivery methods, a controlled patient group, and unique safety-management systems designed to mitigate the risks of this novel technology.

Trial Sponsorship and Status

The clinical trial is sponsored by Life Biosciences, a company based in Boston, Massachusetts. As of June 9, 2026, the company announced that the first participant has been treated, marking the official commencement of human testing for this approach.

Participant Selection and Scope

The trial is initially focused on a small, specific group of patients to test the safety and efficacy of the therapy:

• Initial Group: The company aims to treat as many as 12 people suffering from a form of glaucoma.
• Expansion: The trial intends to eventually include participants with NAION, a severe and acute condition that causes nerve damage in the eye.
• Strategic Location: The eye was chosen as the initial site for the trial because it offers a higher degree of safety; researchers believe the risk of "life-threatening" side effects is lower when targeting the eye compared to other internal organs.

Delivery and Technical Execution

The logistics of delivering the gene therapy require precise biological tools:

• Viral Vector: The therapy uses a common virus to act as a "shuttle," delivering the three reprogramming genes directly into the retinal ganglion cells.
• Target Area: Specifically, the treatment targets the long fibers that make up the optic nerve.

Safety and Control Logistics

A defining logistical feature of this trial is the "switch" mechanism used to manage the activation of the genes:

• The Doxycycline Toggle: To provide "a lot of control," the genes are designed to only switch on when the participant takes the antibiotic doxycycline.
• Activation Control: If the antibiotic is withdrawn, the genes switch off. This allows researchers to ensure that gene expression does not last "longer than is necessary to rejuvenate the cells" and can be stopped if adverse effects occur.

The methodology behind the world's first cellular reprogramming clinical trial relies on a technique called partial reprogramming, which aims to rejuvenate aged cells by restoring youthful features without causing them to lose their specialized functions.

The core components of this methodology include:

Genetic Intervention

The trial utilizes a novel gene therapy approach that involves turning on three specific genes. These are selected from a group of four genes typically used in laboratories to revert adult cells back into a stem-cell-like state. By using only three of these genes, researchers aim to "nudge" the cells back in time just enough to restore youthful behavior without pushing them so far that they lose their identity as specialized retinal cells.

Delivery Mechanism

To get these genes into the target area, the methodology employs a viral vector. Specifically, a virus commonly used in gene therapy acts as a "shuttle" to deliver the reprogramming genes directly into the retinal ganglion cells, the long fibers of which form the optic nerve.

Precision Control System

A critical and unique part of the methodology is the use of a chemical "switch" to manage gene expression:

• The Doxycycline Toggle: The system is designed so that the three genes are only activated when the patient takes the antibiotic doxycycline.
• Reversibility: If the antibiotic is stopped, the genes switch off. This provides researchers with significant control, allowing them to ensure the cells are not exposed to the reprogramming proteins for longer than necessary to achieve rejuvenation.

Scientific Foundation and Targets

The methodology is based on research from David Sinclair’s lab at Harvard Medical School, which demonstrated in 2020 that this approach could promote neuron regeneration and reverse vision loss in mice with glaucoma and aged mice. Before moving to humans, the sponsoring company, Life Biosciences, validated this method in rodents and monkeys, reporting no serious adverse effects.

The current human trial focuses on retinal nerve damage because the eye provides a contained environment where the risk of catastrophic or life-threatening side effects—such as the potential for cells to become cancerous—is lower than in other organs.

The scientific foundation of the cellular reprogramming clinical trial is built upon a decade of research into epigenetics and cell biology, specifically focusing on the concept of partial reprogramming to reverse the effects of aging.

Core Biological Concept

The trial is based on the principle that adult cells can be "nudged" back in time to restore youthful features. Unlike full reprogramming, which turns adult cells into stem cells, partial reprogramming aims to restore a cell's youthful function without forcing it to lose its specialized identity—ensuring, for example, that a retinal cell remains a retinal cell.

The Sinclair Lab and Initial Breakthroughs

The scientific impetus for this human trial stems largely from research conducted in David Sinclair’s lab at Harvard Medical School.

• 2020 Mouse Study: Researchers demonstrated that activating three specific genes in mice with damaged optic nerves could promote neuron regeneration and reverse vision loss in both aged mice and those with glaucoma.
• Gene Selection: The therapy utilizes three of the four "Yamanaka factors"—genes typically used in laboratories to revert adult cells to a stem-cell-like state. By using only three, researchers hope to achieve rejuvenation while avoiding the risk of cells becoming undifferentiated or cancerous.

Preclinical Validation

Before moving to human trials, the sponsoring company, Life Biosciences, conducted extensive animal testing to validate the safety and efficacy of the approach:

• Rodent and Monkey Studies: The company has studied partial reprogramming in rodents and monkeys, reporting no serious adverse effects from the treatment.
• Disease Modeling: Beyond the eye, the company has also tested this approach in animal models of liver disease, suggesting the underlying science may be applicable to various age-related conditions.

Ongoing Scientific Debates

While the animal data is promising, the scientific community remains cautious about translating these foundations to humans:

• The Cancer Risk: A major lingering concern is that the reprogramming process could inadvertently tip cells into a cancerous state.
• "True" Youth: Some scientists, such as translational neurobiologist Pete Williams, question whether modified cells are truly becoming "younger" in a biological sense or if they are simply being reprogrammed to behave differently.
• Early-Stage Technology: Experts emphasize that the technology is still in its infancy, and the potential for "catastrophic side effects" remains a significant scientific hurdle.

The cellular reprogramming clinical trial, while promising, is accompanied by significant risks and concerns ranging from immediate biological dangers to long-term implications for the scientific field.

Biological and Safety Risks

The most pressing safety concerns mentioned in the sources involve the unpredictable nature of reprogramming adult cells:

• Cancer Risk: A primary and "lingering concern" for researchers is that the reprogramming process—which involves turning on specific genes to revert cells to a more youthful state—could inadvertently tip those cells into a cancerous state.
• Potential for Catastrophic Side Effects: Experts note that because the technology is in its "really early" stages, there is a high potential for "catastrophic side effects". This is why the eye was specifically chosen for the first trial; researchers believe the risk of life-threatening outcomes is lower when targeting the eye compared to major internal organs.
• Irreversibility vs. Control: While the trial uses a chemical "switch" (doxycycline) to turn the genes off if something goes wrong, the fundamental risk remains that the cellular changes might lead to unforeseen permanent damage.

Scientific and Conceptual Concerns

Beyond physical safety, there are several scientific uncertainties regarding the efficacy of the treatment:

• "True" Rejuvenation: Some scientists question whether the therapy is actually making cells biologically "younger" or simply reprogramming them to behave differently without reversing the underlying aging process.
• Translation from Animal Models: Although animal studies in rodents and monkeys have not shown serious adverse effects, the move to human trials represents a major leap, and safety remains a paramount concern that animal data cannot fully resolve.

Reputational Risk to the Field

There is also a significant concern regarding the public and professional perception of the trial:

• The Impact of Failure: Because this trial has received a "bright public spotlight" and significant hype, some experts worry about the fallout if it fails. Pete Williams, a translational neurobiologist, warned that if the trial goes "catastrophically wrong," it could negatively impact the future of all rejuvenation research, potentially "screwing" the field for years to come.

The Behavioral Economics Guide 2026

 The concept of Homo Experiens is presented in the Behavioral Economics Guide 2026 as the foundation for "Behavioral Economics 3.0," or the "third wave" of the field. Proposed by Ulrike Malmendier, this model argues that economics must evolve beyond treating humans as "robotic" decision-makers and instead view them as living organisms whose minds and bodies are durably marked by their life histories.

The Evolution Toward Homo Experiens

The sources contextualize Homo Experiens by contrasting it with previous models of human behavior:

• Neoclassical Economics (Homo Economicus): Portrayed people as perfectly rational utility maximizers.
• Behavioral Economics 1.0 & 2.0: Improved realism by identifying systematic "bugs" or biases (like loss aversion) in human reasoning. However, Malmendier argues these models remain "mechanical" because they assume the same "program" runs for everyone, regardless of their unique life history.
• Behavioral Economics 3.0 (Homo Experiens): Focuses on "experience effects"—the idea that personally lived experiences shape beliefs and actions in ways that theoretically learned information cannot.

The Biological Basis: A "Rewiring" Issue

A central argument for Homo Experiens is that life experiences leave physical traces in the brain and body, drawing heavily on the life sciences.

• Neuroplasticity: The brain is not a static processor; it physically reorganizes itself in response to experience.
• Long-Term Potentiation (LTP): Repeated exposure to conditions (like years of high inflation) strengthens specific neural pathways, making those experiences leave "deeper traces".
• Emotional Tagging: Memories accompanied by strong emotions (like fear during a market crash) are encoded more deeply and retrieved more readily.
• Hardwired Beliefs: Because these effects are biological, they often cannot be "lectured" away; knowledge has limited power to neutralize a physical synaptic trace.

Empirical Evidence

The sources provide several examples of how this model explains real-world behavior that traditional models miss:

• "Depression Babies": People who grew up during the Great Depression remained risk-averse for their entire lives, long after the economy recovered.
• Expert Immunity: Even elite decision-makers are affected. For instance, Henry Wallich, a former Federal Reserve Governor who witnessed German hyperinflation as a child, became the most "hawkish" inflation-fighter in Fed history, despite having access to the same data as his colleagues.
• The "Inflation Scar": The Baby Boomer generation overpaid approximately $22 billion for fixed-rate mortgages in the late 1980s and 1990s because their lived experience with high inflation made them irrationally fearful of adjustable rates.

Implications for Policy

The shift toward Homo Experiens fundamentally changes the "remedy" for suboptimal behavior. While previous models suggested providing more information or financial literacy training, the Homo Experiens perspective suggests the targeted design of experiences.

An example provided is the Early Start Pension in Germany. Rather than just teaching children about the stock market, the policy gives them small monthly contributions to invest automatically starting at age six. The goal is to allow them to "live through" market cycles, building a personal experience history that rewires their perception of risk and return over time.

The Behavioral Economics Guide 2026 characterizes biological foundations as the catalyst for "Behavioral Economics 3.0," a shift away from modeling humans as "robotic" processors toward viewing them as living organisms whose decision-making systems are physically altered by their environments.

The Brain as a Dynamic System (Neuroplasticity)

A central theme is that the human brain is not a static computer; it is an organ that physically reorganizes itself in response to lived history, a property known as neuroplasticity.

• Long-Term Potentiation (LTP): Repeated or prolonged exposure to specific conditions, such as a multi-year recession or high inflation, strengthens neural pathways through LTP. This is a measurable biological process where synaptic connections become physically stronger with repeated activation, meaning the longer an economic episode lasts, the deeper the "traces" it leaves.
• Emotional Tagging: Biological foundations explain why personal experiences override textbook knowledge. Experiences accompanied by intense emotions like fear or anxiety (e.g., a stock market crash) receive "emotional tagging," which causes them to be encoded more deeply and retrieved more readily than purely learned information.
• Physical Connectivity: Beliefs are described as a "rewiring issue, not a firing issue". This means that once a life experience has physically reshaped the brain's structural connectivity, it often cannot be neutralized simply by providing better data or education.

Interacting Biological Systems vs. Isolated Preferences

Isabelle Brocas argues that traditional economics incorrectly partitions behavior into distinct "preference modules" like risk, patience, or self-control. Instead, a biologically informed view sees behavior as the output of interacting biological processes—including perception, attention, valuation, affect, and regulation—that are recruited in different combinations depending on the task.

• Internal Resource Scarcity: The brain itself faces economic constraints, such as limited cognitive resources that must be allocated across tasks. Performance limitations are not "errors" but endogenous responses to these internal resource constraints.
• Self-Control as Optimization: Rather than a "moral struggle," self-control is modeled as a constrained optimization problem where the brain weighs rewards against costs subject to internal computational and regulatory limits.

Biological Embedding of Environment and Development

The sources emphasize that social and economic environments become "biologically embedded" over time.

• Life-Cycle Heterogeneity: Biological architecture is not constant; it changes significantly from adolescence to old age, impacting how systems involved in reward processing and cognitive control function at different stages.
• Environmental Impact: Factors like poverty, chronic stress, poor nutrition, and sleep disruption can physically shape the architecture of the systems used for decision-making. For instance, a child in an unstable environment may biologically adapt to prioritize immediate rewards because the future is physically perceived as unreliable.
• Physical Markers: High-stress economic roles can leave permanent biological marks, such as accelerated aging and reduced life expectancy.

Implications for Policy Design

Taking biological foundations seriously fundamentally changes policy interventions.

• New Policy Levers: Instead of just adjusting prices or incentives, the guide suggests policy should target biological factors like stress, nutrition, sleep quality, and cognitive load.
• Experience-Based Remidies: Because knowledge has limited power to "undo" a synaptic trace, the guide advocates for the targeted design of experiences. An example is the German "Early Start Pension," which allows children to "live through" market cycles to biologically rewire their perception of risk and return over time.

The Behavioral Economics Guide 2026 defines neurobiological mechanisms as the structural and functional foundations that transform human beings from "robotic" processors of information into living, breathing organisms whose decision-making systems are physically altered by their environments.

Structural Adaptation: Rewiring vs. Firing

A core argument in the guide is that economic beliefs are a "rewiring issue, not a firing issue". This means that life experiences do not just cause neurons to activate (fire), but physically reorganize the brain's structural connectivity through several key mechanisms:

• Neuroplasticity: The brain is not a static processor but an organ that physically reorganizes itself in response to every significant lived experience.
• Long-Term Potentiation (LTP): This is a measurable biological process where synaptic connections become physically stronger with repeated or prolonged activation. In an economic context, this explains why lasting episodes—such as years of high inflation or a long recession—leave deeper biological traces than brief shocks.
• Emotional Tagging: Memories accompanied by intense emotions like fear (e.g., during a market crash) or anxiety (e.g., job loss) are encoded more deeply and retrieved more readily. These "tagged" memories have a disproportionate power to influence behavior compared to textbook knowledge because they are tied to physical responses like a racing heart or sleepless nights.

Interacting Biological Systems

The guide moves away from treating behavior as a set of isolated "preference modules" (like risk or patience) and instead views choice as the output of interacting biological processes.

• Process Deconstruction: Decision-making is deconstructed into a sequence of core operations: representing the problem, valuing actions, selecting among them, evaluating outcomes, and learning.
• Modulation of Valuation: Mechanisms like self-control are not a "moral struggle" but a constrained optimization problem. Neurobiological evidence suggests successful self-control involves the modulation of the valuation system itself, where the brain weighs immediate rewards against future costs subject to internal computational limits.
• Internal Resource Scarcity: The brain faces its own economic constraints, such as a scarcity of cognitive resources. This means performance limitations are often endogenous responses to internal scarcity rather than simple "errors".

The Developmental Trajectory of the Brain

Neurobiological mechanisms are not constant throughout life; they evolve across the full life cycle.

• Adolescence: This period is marked by major changes in reward sensitivity, peer orientation, and control systems, explaining why behavior during this stage is more exploratory and volatile.
• Aging: As the brain ages, changes in affective and motivational circuits reshape how individuals evaluate gains and losses, respond to uncertainty, and balance immediate vs. delayed outcomes.
• Disorders: Conditions like ADHD and autism are described as having different biological architectures in systems governing attention, reinforcement learning, and social inference, resulting in behavioral patterns that are internally coherent even if they differ from the norm.

Policy and Biological Embedding

The guide emphasizes that social and economic environments, such as poverty or chronic stress, become "biologically embedded" over time. These conditions can physically alter gene expression and executive functioning, meaning that adult behavior (like high impatience) may be an adaptive expression of a developmental history marked by instability.

Consequently, the guide advocates for policies designed to work with our biology. Because knowledge has limited power to "lecture someone out of a synaptic trace," the guide suggests the targeted design of experiences—such as Germany's "Early Start Pension"—to biologically rewire perceptions of risk and return through long-term, "dosed" exposure to market cycles.

In the Behavioral Economics Guide 2026, the relationship between stress and memory is presented as a fundamental pillar of "Behavioral Economics 3.0," shifting the focus from how people process data to how their lived history physically reshapes their decision-making systems. The guide argues that memory is not a neutral recording of facts but a dynamic process filtered through emotional and physiological stress.

Emotional Tagging: Why Lived Experience Overrides Data

A central concept in the guide is "emotional tagging," a neurobiological process where memories associated with intense emotions—such as fear during a market crash or anxiety during unemployment—are encoded more deeply and retrieved more readily.

• The "Racing Heart" Effect: Lived experiences have a disproportionate power over textbook knowledge because, as the sources note, "the textbook does not come with the racing heart and the sleepless nights".
• Rewiring vs. Firing: Because stress triggers structural changes in the brain (like Long-Term Potentiation or LTP), economic beliefs become a "rewiring issue," where stressful memories leave physical synaptic traces that cannot be easily neutralized by providing new information or education.

Stress as a Filter for Memory and Narratives

Ulrike Malmendier provides empirical evidence that individual stress responses, or "stress elasticity," are powerful predictors of future economic beliefs.

• Memory Distortion: Whether a person recalls a past period as one of "high inflation" is more closely linked to the physical stress symptoms (e.g., stomach problems, body tension) they felt during that time than to the objective financial strain they endured.
• Causal Narratives: People with high stress elasticity are significantly more likely to generalize a specific stressful episode (like the "energy crisis") into a permanent global narrative about how the economy works, effectively "scarring" their lifelong economic worldview.
• Controllability: A key mitigating factor is perceived control. Individuals who felt they or the government had agency during a crisis were significantly less likely to be "scarred" by the memory of that period.

Biological Embedding of Environmental Stress

The guide highlights that environments characterized by chronic stress, such as poverty, become "biologically embedded".

• Structural Impact: Prolonged exposure to stress and unpredictability can physically alter the architecture of the brain's systems for memory and executive functioning.
• Adaptive Impatience: What appears to be "impatience" in adults may actually be the expression of a developmental history where stress taught the individual that the future is unreliable and immediate rewards are safer.

Policy and Intervention Design

Recognizing that you "cannot lecture someone out of a synaptic trace," the guide advocates for policies that work with, rather than against, our biology:

• Targeted Experience Design: Instead of just teaching financial literacy, policy should focus on designing positive experiences. For example, Germany's Early Start Pension provides children with "dosed" exposure to market cycles to build a non-traumatic memory history, biologically rewiring their perception of risk over time.
• New Policy Levers: Effective policy should aim to directly lower stress, improve sleep, and stabilize expectations, as these factors determine how the biological systems for memory and regulation are taxed.
• Tailored Support: In education, recognizing that high-stakes incentives can increase stress and reduce performance for anxious students allows for interventions that provide structure and movement rather than just academic tutoring.

The Behavioral Economics Guide 2026 advocates for a fundamental shift in policy and intervention design, moving from traditional information-based remedies toward an approach that recognizes humans as living organisms shaped by their unique life histories and biological constraints. This "Behavioral Economics 3.0" framework suggests that the next generation of policy must be more tailored, diagnostic-focused, and experiential.

1. From Education to the Targeted Design of Experiences

A central theme is that traditional remedies, such as providing more information or financial literacy training, are often insufficient because you "cannot lecture someone out of a synaptic trace". Because lived experiences physically rewire the brain (neuroplasticity), policy should focus on shaping the experiences through which information is encoded.

• The Early Start Pension (Germany): Instead of just teaching children about stocks, this policy gives them small monthly contributions to invest automatically starting at age six. The goal is to build a personal history of "living through" market cycles to biologically rewire their perception of risk and return over time.

2. Expanding the "Policy Levers": Targeting Biology and Environment

Isabelle Brocas argues that taking biology seriously changes what economists treat as a policy lever. Interventions should not only target prices or incentives but also address internal and environmental factors that "tax" our regulatory systems:

• Biological Levers: Effective policy might involve efforts to lower stress, improve sleep, stabilize expectations, or improve nutrition, as these factors determine how the brain processes information and regulates impulses.
• Identifying Binding Constraints: Before designing a "nudge," policymakers must diagnose which specific process is binding—is it a lack of information, emotional overload, cognitive depletion, or a lack of perceived control?.

3. Context, Localization, and Development

The sources highlight that interventions often fail because foundational assumptions are built on narrow "WEIRD" (Western, Educated, Industrialized, Rich, and Democratic) samples that do not travel well.

• Diagnose Before You Design: In international development, the binding constraint is often structural or institutional rather than cognitive. The guide advocates for mapping social and structural factors before intervention.
• Cultural Adaptation: A case study on localizing digital health in Costa Rica shows how US-centric "individualist" framing (focusing on personal goals) had to be shifted toward collectivistic framing (emphasizing family and community benefit) and uncertainty avoidance (providing clear, stepwise instructions) to be effective.

4. System-Level Changes vs. Individual Motivation

In high-engagement environments like social media, individual-level literacy training often fails because the digital environment continues to reinforce "low-effort, high-engagement" behaviors over accuracy.

• Modifying Contingencies: Effective interventions must alter the environment by adding friction before sharing, making verification easier, and increasing the visibility of credibility cues.
• Tax Compliance (Norway): The Norwegian Tax Administration uses a "whole-of-community" approach, combining credible enforcement with trust-building initiatives to make compliance the "social and economic default" rather than just relying on audits.

5. AI as a Policy Pre-Testing Tool

Large Language Models (LLMs) are introduced as a new way to pre-test policies through scalable behavioral simulations.

• Synthetic Personas: By equipping LLMs with specific personas (e.g., a low-income single parent), researchers can simulate how different groups might respond to a cash transfer or a new regulation, identifying potential behavioral channels (like "threshold bunching") before launching in the real world.
• Choice Engine: Tools like the "Choice Engine" use LLMs and theoretical frameworks (like COM-B) to predict context-specific decisions and provide a causal narrative for why an intervention might succeed or fail.