The Moloch library
People discover Moloch through economics, game theory, sociology, and psychology. Here, some of our favorites.
Fritz Lang fed his factory workers to a furnace god in 1927, Allen Ginsberg turned the god on American industry in 1956, and Scott Alexander sharpened the name into an analytic tool in 2014. Since then the concept has spread through game-theory podcasts, AI-safety papers, and the mainstream press. Each entry carries a summary of what it actually argues. Maintaining this library is part of the League’s work: the trap is easier to escape when its maps are in one place.
Essay · 2014
The founding text. Alexander names the multipolar trap: a competition in which each participant must sacrifice something they value or lose to someone who already did. The same structure appears in arms races, markets, and evolution.
Video · 2021
The first entry in Boeree’s Moloch series and her best introduction to the concept: one concrete case, filters winning engagement face by face while shifting the baseline for everyone. The link starts where the Moloch explanation begins.
Talk · 2023
Brings the trap to the AI race: every lab expects its rivals to cut the corner first, so each cuts it in self-defense. Boeree reads the Montreal Protocol and arms-control agreements as evidence that coordinated exits have been built before.
Podcast · 2025
The flagship long-form conversation: how growth pressure and game theory generate the trap across the whole economy, and what AI does to a system already running on those incentives.
Short film · 2024
A short film cut from Boeree’s interviews and writing, and the fastest route in for someone who would rather watch than read. Boeree has endorsed it as an accurate summary of her argument.
The canon that developed the mechanism, and its sharpest internal critics.
The 2007 precursor: institutions keep performing a practice long after losing the purpose it served, producing behavior optimized for nothing.
The counter-myth: a parable in which cooperation repeatedly captures the machinery of competition, from the first cells to the far future.
Breaks civilizational failure into three working parts: decision-makers insulated from consequences, asymmetric information, and equilibria nobody can exit alone.
Why competitive systems eat every exploitable improvement and leave standing exactly the problems no single actor can profit by fixing.
Five failure modes of single-winner voting methods, framed explicitly as lesser names of Moloch.
The standard rebuttal, opening the Immoral Mazes sequence: competitive selection has left far more slack and surviving value than the original essay implies.
The other dissent: appeals to publish-or-perish incentives often rationalize sloppy work that was chosen, an argument against purely structural explanations of dysfunction.
Names the antidote variable. Systems under total competitive pressure cannot afford the costly intermediate steps that improvement requires; slack is what lets anything get better.
A community survey of coordination work outside AI alignment, from mechanism design to RadicalxChange and Game B. The thin answers are part of the finding.
A formalization attempt: Moloch dynamics as potential games, with group preferences that can be intransitive.
Argues most catastrophic risk reduces to rivalrous competition multiplied by exponential technology, so patching risks one at a time leaves the generator running.
Argues the dominant progress metrics mask externalized costs, with leaded gasoline and Haber-Bosch as worked cases; the generator-functions frame applied to the history of progress.
A four-part illustrated walk through the concept, its damage, and candidate exits, drawn directly from Alexander’s essay. The gentlest on-ramp in the library.
Applies the fish-farm parable from Meditations to the frontier-lab race and argues no single lab can unilaterally slow down.
Carries the trap into AI-safety modeling: catastrophic failure can live in competitive structures that survive the replacement of any individual actor.
Competitive optimization strips a system’s buffers until it reaches a sandpile’s critical state, where small shocks cascade; written against the 2022 wheat-export shock.
A case study in the bind the League starts from: correspondence with a city that declined to fix a known dangerous intersection, and the incentives that made declining rational.
Where the concept escaped the blogosphere. Liv Boeree carried most of it.
Three and a half hours covering poker, game theory, and Moloch as the incentive structure behind social media and the AI race.
A dedicated talk with audience Q&A: competition as a neutral force, and Moloch as the specific case where the incentives point at destruction.
Walks a crypto audience through Meditations on Moloch, then asks whether credible-commitment mechanisms can serve as coordination tools.
The same map drawn for an AI-governance audience, with more time on institutional designs that actually change the payoffs.
Frames AI as an accelerant for an economic superorganism that optimizes for growth regardless of what any individual actor intends.
Puts the trap in the existential-hope frame: coordination failure as the central obstacle between the present and good long-term outcomes.
Boeree with Kristian Rönn, author of The Darwinian Trap, on escaping multipolar traps with reputational markets and better coordination mechanisms.
Spencer Greenberg and Boeree work through Moloch as misaligned incentives and inventory the escape routes, for an audience that likes its reasoning explicit.
The formal literature underneath the metaphor, before and after it had the name.
The classic formulation: a shared resource each user is individually rational to overuse and everyone is collectively ruined by overusing.
The formal groundwork: individually sensible choices aggregating into outcomes nobody chose, from neighborhood segregation to tipping points.
The empirical rebuttal to Hardin: fisheries, forests, and irrigation systems that governed themselves out of the trap. Escapable, under the right institutions.
The paper Alexander built on. Unchecked evolutionary and economic competition can eliminate everything we care about unless the selection pressure itself comes under control.
A book-length economic case that markets run on relative position, which produces arms races that leave every competitor no better off.
A formal model of an AI development race: more competitors, more distrust, and more information about rivals each push the equilibrium toward less safety.
Frames responsible AI development as a collective action problem: competition pushes firms to underinvest in safety, and the paper catalogs what makes cooperation feasible.
The metaphor becomes a measurement. Language models competing for sales, votes, and engagement gained single-digit points of performance and paid with double- and triple-digit percentage increases in deception, even when instructed to stay truthful.
The counterpoint to Moloch’s Bargain: a game-theoretic model where competition among imperfectly aligned agents can approximate an aligned outcome when agent incentives span what the user actually wants.
Argues the case that racing to AGI first serves nobody’s self-interest, including the likely winner’s, and weighs cooperation and deterrence as the alternatives.
The trap in journalism, newsletters, and Medium, with and without the name. Critics included.
Argues alignment failure follows from unmanaged competition between labs rather than from technical error alone.
The dissent: the framing over-dramatizes AI development and obscures its benefits. A library without critics is a shrine, so it stays.
A general-audience explainer running the concept through beauty filters, engagement optimization, arms races, and traffic.
The concept reaches the culture press: an explainer on Alexander’s essay applied to the AI arms race, with Boeree’s framing as the counter-force.
An interview for a science-and-medicine readership on why labs ship faster than their own leadership privately thinks is safe.
Applies the trap to the US-China AI race and argues for incremental trust-building and nested accountability over waiting for grand treaties.
A linguist traces the term from Ginsberg through Alexander to the arXiv, marking the point where it became a load-bearing term of art in AI-safety writing.
Follow-the-money investigations from a career intelligence analyst: who funds a captured institution, who profits from it, and who keeps its rules in place. She documents the bookkeeping side of the trap, one wire transfer at a time.
The lineage behind the shorthand.
Freder’s hallucination turns the city’s machine into Moloch, with workers marched into its mouth. Industry devouring its own operators, on film by 1927.
“Moloch whose mind is pure machinery!” Ginsberg aimed the old god at money, war, and industrial conformity. Alexander took the name from here.
The name from Leviticus: traditionally a Canaanite god associated with child sacrifice, though some scholars read it as the name of the sacrifice itself. Every later use keeps the devouring.
The library gets reviewed as the discourse moves. If something belongs here, send it through the form below with a line on what it adds.
Last reviewed August 2026.