In July 2003, the public found out that DARPA (the research and dev agency responsible for the internet itself) had been funding a futures market called the Policy Analysis Market. Traders would bet on Middle East political events, including assassinations, coups, terror attacks, and regime changes. The program had been proposed in 2001 by a small San Diego research firm called Net Exchange, with intellectual scaffolding from the economist Robin Hanson at George Mason; and by 2003 it sat inside the Information Awareness Office, whose director was Admiral John Poindexter, Reagan's former National Security Advisor.

Poindexter had been convicted in 1990 on five counts of lying to Congress over Iran-Contra; his convictions were vacated on appeal a year later, on the grounds that trial witnesses had been contaminated by his own immunised Congressional testimony, but their aura never went away.

Senators Ron Wyden and Byron Dorgan went public with the future market on July 28.

The program was killed within 24 hours, and Poindexter resigned two weeks later, because the public reaction - way back in 2003 - was utter revulsion. The idea of betting on whether a head of state would be murdered etc struck almost everyone as obviously gruesome and beyond redemption; editorial writers called it grotesque; and Pentagon officials spent days apologising.

Twenty-two years later, we seem to have drifted a long way from that moral high watermark.

Polymarket ran live contracts in 2024 on whether Vladimir Putin would remain in office, whether Joe Biden would drop out, whether a ceasefire would hold in Gaza by a given date, whether Donald Trump would be assassinated before the November election. Kalshi, the CFTC-regulated American competitor, took hundreds of millions of dollars in volume on the 2024 presidential race. In 2026, folks have been betting on the deaths of Iranian officials and Israeli civilians and nuclear war.

Nobody has resigned, and no senator has been forced to hold a press conference. The markets are covered in the financial press as an actual innovation in retail trading.

We’ve gone from "this is too ghoulish to exist" in 2003 to "this is the new wisdom-of-crowds infrastructure" in 2026. And it's a symptom of how we, all of us, are coming apart.

Prediction markets are, I think, the clearest single sign that our civilisation has entered a late and decadent stage.

The dream and the pitch

The pitch for prediction markets has been the same since Robin Hanson started writing about "idea futures" in 1988 and 1990, and since the Iowa Electronic Markets launched their political futures market that same decade. Markets aggregate dispersed information better than polls, pundits, or committees; if you put money on the line, people stop posturing and start estimating, and prices become a running readout of collective belief.

Hanson's version of this ran deep. He proposed "futarchy," a system where citizens vote on values and markets decide on policy. You'd ask the market whether a given policy would raise GDP, reduce childhood poverty, or cut CO2, and whichever policy the market priced highest would get implemented.

Philip Tetlock's Expert Political Judgment in 2005 and Superforecasting in 2015 supplied the scientific underpinning. Tetlock found that generalist forecasters who updated on evidence, tracked calibration, and competed in open tournaments routinely beat credentialed experts. The Good Judgment Project, funded by IARPA starting in 2011, showed this was repeatable.

Markets do aggregate information. Forecasting tournaments do beat pundits. The humiliation of the 2003 Iraq WMD consensus, and of nearly every major think tank's prediction record in the decade after, gave the prediction-market crowd a genuine argument.

So if the pitch is good, why is the product a sign of rot?

Because the pitch was about epistemics.

The product is about something worse.

What the markets price

Open Polymarket in April 2026. Scroll the trending contracts. You'll find markets on celebrity divorces, CEO firings, troop movements, drone strikes, papal health, celebrity deaths recast as "will X still be alive on December 31," and whether a given pop star will release an album in Q3. The biggest volumes cluster around elections and the personal misfortunes of public figures.

These are bets on whether bad things will happen to specific people, and groups of people, whether institutions will hold, whether the world will feel more or less stable in 90 days.

The prediction-market community will tell you the content of the contracts doesn't matter, because the market's function is to produce accurate probabilities and nothing more - and I don't buy this for a single second. What a society chooses to price reveals what it actually gives a shit about, in the same way that what a society chooses to memorialise reveals what it honours. Tell me which contracts move size and I'll tell you what your civilisation has decided is interesting.

In Renaissance Florence, the biggest public wagers were on papal elections, the outcomes of condottieri campaigns, and whether the Arno would flood before June; you can reconstruct the city's anxieties from the betting books. Our betting books show a civilisation fixated on the humiliation and removal of a small number of public figures, and on the probability that large systems will crack on a short timescale.

This is an unflattering portrait.

Assassination contracts

Polymarket listed a contract in summer 2024 on whether Donald Trump would be assassinated before the election. The contract was scrubbed after the Butler, Pennsylvania shooting in July, for obvious reasons, but crucially it had traded. There was liquidity. There were people on both sides of the bet.

In 2005, Nick Szabo wrote about the dangers of what a crypto-anarchist named Jim Bell had called "assassination politics" back in 1995. Szabo came close to inventing Bitcoin before Satoshi did, and he knew what he was looking at. Bell's original proposal was a market where anonymous donors could pool money that would pay out to whoever correctly "predicted" the date of a public official's death; and the prediction would, of course, be a contract for the hit.

Every prediction-market platform that goes live has to run a gauntlet around Bell’s ghost. Polymarket's terms of service prohibit contracts that could function as murder contracts, and Kalshi does the same - the lawyers know the argument.

But the argument doesn't depend on intent. Hanson himself has written that you can’t cleanly separate a prediction market on whether X will be killed from an incentive to kill X, because the market is information to a would-be assassin about how much financial upside exists in acting on their impulse; it’s a relatively clean way for a hostile state actor to hedge a covert operation. A sovereign that wants a rival head of state dead can, in principle, acquire a large position on a thinly traded market, wait for someone to commit the act, and pay for the operation with the winnings.

In 2003, this argument was enough to kill a DARPA program and end a career.

In 2026, the same argument is background noise. We've collectively decided that the information value of these markets outweighs the moral cost of treating human lives as tradable securities, and this (to me, at least, and I accept that I may be alone in this) that decision is a bleeding mistake.

The dead pool and the decline

Tudor Londoners wagered on the life expectancy of public figures so routinely that life insurance, as we understand it, grew out of the same market. Geoffrey Clark's Betting on Lives, published in 1999, traces the 18th century English insurance market as a functioning prediction market on the deaths of dukes and royal mistresses. Parliament shut it down in 1774 with the Life Assurance Act, which required insurable interest, because the legislators of the era understood something we've apparently, conveniently and somewhat profitably forgotten. Permitting strangers to bet on whether a named person would live or die produced, in aggregate, darker incentives than the information-gathering benefit could justify. This should be obvious. In fact, to anyone paying attention, this is obvious.

The 18th century London markets at scale were disastrous. Ambassadors were assassinated. Heirs were poisoned. The statute was, by the standards of the 1770s, a moral intervention.

But we repealed that moral intervention, and we repealed it with software. Each new prediction market opens with a standard disclaimer that the platform doesn't allow murder contracts, and then lists contracts on the lives of named public figures, reinventing 18th century betting practices and rebranding them too, as innovations and disruptions.

The Roman Empire late in its decline had booming gambling markets on gladiatorial outcomes. The Byzantines had a full betting economy around chariot racing that produced the Nika riots of 532 CE, which killed tens of thousands. Late Qing China had opium-fueled fan-tan parlors that functioned as quasi-markets on political outcomes. Weimar Germany had the Tauentzienstraße betting shops that took wagers on the next Chancellor and, after 1930, on which faction would be next to be shot in a street brawl.

None of this is to claim that gambling causes decline; that would be a cheap causal argument, and I’m not yet in that business...

My claim is a little narrower, at least.

In each case, a civilisation under strain stopped prosecuting its disputes through argument and institution, and started pricing them; the bettors were reading the decline the way a barometer reads a storm, even if the storm came from somewhere else.

Sandel's objection, twenty years late

Michael Sandel, the Harvard political philosopher, published What Money Can't Buy in 2012. The core argument of the book is that some goods are corrupted they moment they’re priced. A Nobel Peace Prize that can be bought at auction isn't a Nobel Peace Prize, something that Donald Trump may or may not have grokked; a friendship that's bought and sold cannot possibly qualify as a friendship; a citizenship that has a purchase point, in the Maltese Golden Visa sense, isn't actually any kind of citizenship that actually matters, in any kind of philosophical sense.

Sandel's objection to prediction markets is that certain questions change their nature when you put them in a market frame. Markets don't need to produce bad information for this to go wrong; they do the damage by producing any number at all. Ask "is the Secretary of Defense going to resign by June 1" in a newsroom and you get a political question - you talk about his relationship with the President, the policy disputes inside the cabinet, the institutional pressures from Congress etc. The question is embedded in a set of relationships and public obligations.

Ask the same question on a prediction market and you get a probability between 0 and 1. The market has no view on whether he should resign, whether the policy fight is worth winning, whether the institutional damage is worth the political cost and so on - It only has a price, because it only needs a price.

Prediction markets route around normative argument without destroying it; they provide a parallel answer, priced and continuous, that makes the unpriced conversation feel slow and unserious by comparison. Why listen to a journalist reason about whether the ceasefire will hold when you can see that it's trading at 34 cents?

The laziness dividend

Cass Sunstein and Richard Thaler wrote Nudge in 2008 with a section on prediction markets that reads, now, as a period piece. They praised the markets as a way to get past groupthink and expert capture and perhaps they were right about the epistemic problem, but I think it’s easy to see that they were wrong about where the pressure would move. The pressure has moved toward laziness - once a price exists, a journalist stops reporting and an analyst stops analysing and a decision-maker stops deciding. Everyone's waiting for Polymarket to update.

During the 2024 US election, major news outlets, including the Financial Times and the Washington Post, quoted Polymarket's implied probabilities in running coverage. The number was treated as a live readout of election reality, and when the numbers moved, articles were written about the movement. The question of what was driving the movement, which is the actual journalism, came second. In 2024, Nate Silver shifted to publishing both his own forecast and the Polymarket number, and spent much of October explaining why they diverged. His model at 538 had dominated election coverage for 12 years before that. The work of explanation became a reaction to the price.

Silver is one of the more honest figures here. He's said in print that prediction markets are his competitors, that they force him to sharpen his reasoning, and that he thinks the aggregated number contains signal his model misses. And fair enough - I can accept that as a good faith position. But the broader effect, across the field, has been that journalism about uncertain future events has collapsed into price commentary, and the markets have become the story, and the story about the markets has replaced the story about the world.

Replacing argument with price

Alasdair MacIntyre argued in After Virtue, published in 1981, that modern moral discourse is a ruin. We use the vocabulary of older ethical traditions, Aristotelian virtue, Christian duty, Kantian rights, without the shared community of practice that gave those words their meaning, and so we shout past each other, trading fragments that no longer cohere. His example was the debate over abortion - but you could use almost any political question from 1981 forward.

The prediction market is the ultimate post-MacIntyre moral technology, asking only what will happen. Questions about what we owe each other, what justice requires, what a good outcome would be, what a morally defensible position would represent - the market has no machinery for. Values drop out of the picture, because the price is the only fact.

he defenders rarely argue that the markets produce better outcomes in any thick sense of "better." They argue that the markets produce more accurate probabilities, as though accuracy is the only remaining virtue; but it's the virtue you keep when you've stopped believing in any of the others.

When a civilisation loses its ability to answer "what should we do" it retreats to answering "what will happen?" The late Romans did it, and late medieval astrologers did it, and late 19th century social Darwinists did it too. Each of these movements felt, to its practitioners, like a rigorous clarification, and each, in retrospect, is closer to a surrender.

Prediction markets are the 21st century version of that surrender: a technology for converting questions of value into questions of fact, and then trading the facts.

The Scott Alexander problem

Scott Alexander Siskind, writing as Scott Alexander at Astral Codex Ten, is the most thoughtful public defender of prediction markets working today. His argument, refined across a dozen essays from 2012 to 2025, is about this: prediction markets are useful tools for aggregating information and forcing experts to put money where their mouths are. They have costs, yes, but the costs are manageable, and so we should want more of them, not fewer.

But the question that matters isn't "do prediction markets produce accurate probabilities." They do, sometimes, on questions where they have enough liquidity and no manipulation incentive. I think the question is whether a civilisation that routes more and more of its public life through these markets is one in good health or coming apart at the seams.

The rationalist position = that better epistemics is always a good; knowing what's true is the first step to making things better, and you can't improve what you can't measure. But some things, some parts of our existence are degraded by measurement. Marriage quality, and artistic achievement, and the sanctity of a deliberative process. When you take a thing that was embedded in relational or political context and reduce it to a number, you may have made the thing more easy to understand; but you've also changed what the thing is.

This was James Scott's argument in Seeing Like a State, published in 1998, about forestry and city planning. The state, in order to manage a forest, has to render it as timber volume; once rendered, the forest is managed as timber; and so the ecological complexity, the cultural meaning, the local knowledge of which stands of trees matter for which villagers, all of it disappears into the measurement...it’s all timber, all the way down. Which of course, is not so different from defining an entire population as so many corpses.

Prediction markets render deliberation as probability, and once rendered, public questions are managed as probability, and the deliberation that produced the question vanishes - the argument for why the question matters vanishes too.

What's left?

The price.

Who benefits

The money on Polymarket and Kalshi comes from some identifiable sources. Crypto-native traders looking for a new volatility surface after the 2022 collapse of the lending markets, and Quant firms running information-arbitrage strategies, and political operatives testing narratives, and Journalists and hobbyists putting down small stakes for entertainment.

In the 2024 US election, Polymarket data showed a single French trader, named in a Wall Street Journal piece by Alexander Osipovich and Shane Shifflett in October 2024, putting down about $30 million across several accounts to bet on Trump. The bet moved the implied probability for weeks. And he won about $85 million when the results came in.

Set aside whether he had inside information. The point is that the "market consensus" on the most important political question of the year was shaped by the convictions of one rich person willing to take a large position. The market aggregated information, yes, but the information it aggregated was dominated by the bankroll of one participant.

This is the manipulation problem in miniature. In any market with thin liquidity and high civic importance, the price is going to reflect the beliefs of whoever's willing to put the most money in. The people who gain from this arrangement are the same people who gain from any financialisation of public life. Traders, platform operators, and a small cohort of well-capitalised political actors who can now move the apparent consensus on a question by buying it. The people who lose are everyone else. The citizen who reads the Polymarket number as a fact is consuming a number produced in part by someone's willingness to spend. The journalist who quotes the number is laundering that person's money into public knowledge. The policy-maker who uses the number to justify a decision is delegating to the bankroll. This was always the critique of modern financial markets at scale, from Hyman Minsky in the 1980s through to Adair Turner and Mariana Mazzucato in the 2010s. Prediction markets inherit it, and the civic stakes make it worse.

Why this feels like decay

Joseph Tainter argued in The Collapse of Complex Societies, published in 1988, that collapses share a signature. The society develops expensive institutions to manage complexity, and the returns on those institutions decline. The society can't afford them, and so the institutions fail or are abandoned. Reading Tainter alongside the current prediction-market boom is a strange experience. The pattern fits, but it fits sideways. The expensive institutions are things like the professional press, the civil service, the academy, the peer-reviewed journal, the national statistical agency - these were built through the 19th and 20th centuries to produce reliable public knowledge. They're all, right now, in various states of crisis // collapse…

Prediction markets are cheap. They don't need credentialed staff, or editorial judgment, or institutional memory. They produce a continuous stream of apparently reliable numbers, on any question you can phrase, for the cost of a small trading fee. From a pure cost-benefit standpoint, they look like a massive improvement on the old institutions. From a civic standpoint, they're a replacement, like a drive-through replacing a dinner table. The drive-through feeds you faster and cheaper; but the thing the dinner table was for, a slow shared practice of attention, isn't among the things the drive-through provides. Tainter's argument is that societies rarely notice they're replacing the dinner table with the drive-through until the dinner tables are gone. The cost savings look real and the institutional loss is invisible until a crisis demands the capabilities that only the old institutions had.

We've seen previews of this. During the 2020 pandemic, prediction markets priced the course of the disease with about the same accuracy as public health agencies, sometimes better. Many technologists used this as evidence that the CDC and WHO should be replaced in part by forecasting infrastructure. But the CDC was built to coordinate the response, distribute vaccines, run surveillance, and train the next generation of epidemiologists so the country would have them when the next crisis came - and forecasting was a small piece of a much larger public-health organism.

When you propose to replace the organism with a market, you're trading a capability for a number. The number is cheaper. When the next crisis comes, the number won't help. This is what civilisational decay looks like in detail. Expensive institutions are eaten by cheap substitutes, who are capable of doing only one thing the institutions did; the other things, the work of being a polity that can act, drop out. And by the time the polity needs to act, the infrastructure is long, long gone.

Late moves, short histories

Late-period civilisations discover elegant, efficient-looking technologies right before they're unable to use them...

The late Roman Empire had glass blowing, sophisticated concrete, hydraulic engineering, and long-distance banking on a level the West wouldn't rebuild for 800 years. The late Song dynasty had printing, gunpowder, movable type, and a paper currency system a millennium before European equivalents. Each civilisation's late period looks, to the historian, like a technological peak right before a collapse. This is partly survival bias, I’ll admit: we notice the technologies because they survived the collapse in the written record. But civilisations under pressure do accelerate innovation as a substitute for institutional repair - because the efficient new tool is cheaper than the expensive old practice, and the new tool gets adopted fast. The old practice atrophies, and when the new tool runs up against a problem it can't solve, the civilisation has no fallback.

Prediction markets fit this. They're elegant, they're efficient, they're sold to all of us as a modern replacement for expensive institutional practice, and they do solve one real problem, which is that credentialed experts are overconfident in their forecasts. But that problem is a tiny piece of the problem that the old institutions were built to handle. If you read Polybius on late Rome, or Ibn Khaldun on the Maghreb dynasties of the 14th century, or Gibbon on the Antonine age, you’ll highlight the same shit: clever technical solutions proliferate, and civic competence declines. People blame the institutions for their inefficiency, without noticing that the efficiency of the replacements is achieved by discarding the functions that the institutions existed to provide. Khaldun called this stage haḍara, the settled, luxurious phase of a civilisation where the original virtues have been hollowed out by comfort and specialisation.

What's missing from the price

Take a contract on "will Israel and Hamas reach a ceasefire by December 31, 2026." The price, on any given day, is some number between 0 and 1.0 Say it's 0.22.

What the price doesn't contain: an account of why a ceasefire would be a moral good, an account of who bears responsibility for the failure of previous ceasefires, a theory of what international pressure could shift the outcome, a map of which hostages are still alive, a record of what the killed journalists were writing before they died, or any sense of what it would mean, for the children now growing up in the Middle East, for the war to end in November instead of January.

None of this can be priced - the market cannot hold it.

The market can only hold the collapsed summary.

The market's defenders will say that all of this exists elsewhere, in the journalism, the NGO reports, the academic analysis, the long-form commentary. But "elsewhere" is losing its funding, losing its audience and losing its status, while the market is gaining all three. The price is becoming the authoritative output, while the elsewhere is becoming the decorative commentary around the price. And when you invert the relationship between the deliberation and the summary, you change what the summary means. In a healthy system, the probability number is a shorthand for a rich debate; in a decaying system, and we are in a decaying system, the debate is a shorthand for the probability number.

From orbit…

If I were diagnosing a civilisation from orbit, I'd look at what it bets on and what it refuses to bet on. A healthy civilisation bets on games, on contests, on horses, on private entertainments; it draws a line around the sacred or the civic, and refuses to price what's inside that line, and the line might move around, but at least it exists. A civilisation in decay erases the line: everything becomes a contract, from the death of a public figure, to the course of a war, to the outcome of an election, the next pandemic, the marriage of a celebrity, the survival of a pope. Nothing is held out of the market, because nothing and no one is sacred.

The first prediction markets, the Iowa markets in the late 1980s, confined themselves to electoral outcomes. Intrade, which launched in 2001 and collapsed in 2013, pushed the envelope into celebrity deaths and ran into legal and reputational trouble; Polymarket, since 2020, has been willing to list almost anything that generates volume. Each platform that pushed the boundary might not have gotten away with it, the boundary still moved all the same. The social response got weaker, and so did the legal response. I don’t think the boundary actually exists anymore, not in any meaningful sense. You can bet, right now, on the death of almost any named public figure, on the outcome of active military operations, on whether specific children of specific celebrities will be arrested. I don't think this happened because we decided as a society that it was fine. I think it happened because we stopped having a mechanism for deciding anything as a collective bunch of normal bloody people.

There's no golden age of public deliberation to return to. The 18th century betting books I described earlier coexisted with slavery, wife-selling, and press-ganging, and the 20th century public sphere excluded roughly half the population. My claim isn't that we've fallen from some prior height of mora superiority. I’m no fool. But civilisations - ours, specifically - can build institutions that hold certain questions out of the market, treat them as scred, and handle them through deliberation instead. When those institutions are healthy, the society can argue and act together; when they rot, the market floods in and prices what the institutions held out.

A version of us that wasn't decaying would have, in 2003, rejected the Policy Analysis Market, built better public forecasting inside the civil service, and kept the private prediction markets confined to commerce and entertainment. A version of us that wasn't decaying would treat Polymarket contracts on assassinations the way we treat snuff films, as something the market can technically produce and that the society refuses to consume. But we don't treat them that way, do we?

We cite them in the Financial fucking Times.

A small hopeful note

A few people are still holding the line. The UK's Government Office for Science has experimented with internal prediction markets limited to scientific and technical questions, while keeping political questions out. Singapore's civil service uses forecasting tournaments of the Tetlock kind, carefully scoped. The Metaculus platform, non-monetary and governed by a research norm, has tried to build forecasting infrastructure with stronger civic guardrails than the commercial markets.

These are small efforts, fighting against a much larger tide, but they suggest that the choice between "no forecasting at all" and "price everything" isn't the only choice available. You can have institutions that use prediction-market techniques on some questions, under constraints, while defending a line that keeps other questions civic.

I think Polymarket and Kalshi are early rather than final. The infrastructure is cheap, the regulatory fights are mostly won, and the cultural objection has collapsed. Over the next 10 years, you'll see prediction markets embedded in news apps (they’re already live in Substack), used as the primary data feed for political coverage, integrated into corporate decision-making, and deployed inside political campaigns as both polling infrastructure and voter-suppression tools.

You'll see a second wave of markets on things that now seem unthinkable: markets on the outcomes of specific criminal trials, markets on marriages and divorces of named ordinary people who become briefly famous, markets on child custody outcomes, markets on refugee-camp mortality. There's no principled line that stops the expansion once the line against "civic questions" is gone, and that line fell in the early 2020s.

The prediction markets are the clearest sign of decay because they're the case where the pitch is most defensible, the technology works, the outputs are useful, and the long-term effect is corrosive anyway. You can't argue against them on their own terms; the terms are already the problem.

What you can do is keep asking the questions that the market can't price. What do we owe each other? What should we refuse to sell, even if someone wants to buy it? What are the things we used to know and have started forgetting? Those questions produce arguments and if we’re lucky, sometimes the arguments produce institutions, and if we’re luckier still, the institutions are the load-bearing walls of a civilisation that's still alive.

Our civilisation can still produce them. It mostly doesn't.