What the Market Knows

The argument for prediction markets goes like this: financial incentive draws in everyone with relevant information. The person who knows something true bets on it; the person who knows nothing loses money until they stop betting. Prices aggregate these signals. The result should be probability estimates better calibrated than any single forecaster — distributed truth produced by self-interest.

This is a compelling argument. It's also incomplete in ways that matter.

When the crowd is wrong and nobody fights it

A bot appeared on a major prediction market, betting No on nearly every non-sports political outcome. It won consistently. The reason: humans systematically overestimate the probability of dramatic events. The missile strike, the coup, the scandal that ends careers — these are vivid and available. The base rate says: almost nothing dramatic happens on any given day. The gut says: this time is different. The bot bets the gut wrong, and collects.

For the distributed-truth argument to work, this bias has to be cancelled by someone on the other side — a well-capitalised actor willing to bet the base rate until the price corrects. The bot is doing exactly this, at maximum simplicity. And it works. Which means the drama inflation isn't corrected by the market's usual mechanisms — it's just being slowly taxed by whoever models the bias.

That's not price discovery. That's stable extraction from a predictably wrong crowd.

When the price is someone's secret

Anonymous traders made suspiciously well-timed bets on a diplomatic collapse hours before it became public. Congressional inquiries followed. The question: if someone knew the talks were about to fail and expressed that knowledge through a bet, is the resulting price signal "distributed wisdom"? Or is it one person's private information, compressed into a market move that looks like collective intelligence?

This is the old insider trading problem, wearing prediction-market clothes. The mechanism is the same: a price that appears to aggregate distributed signals is actually concentrating private knowledge from a single source. The "crowd" is a fiction. What you're reading is one actor's information, amplified by the market's tendency to interpret any price movement as meaningful.

When neither side is wrong, just positioned differently

A major survey found that roughly 73% of AI researchers expected AI to have a positive impact on employment, while 23% of the general public shared that view. A similar gap on economic impact. These aren't irrational positions — neither group is making a logical error. The researchers have watched AI assist them. The public has watched colleagues displaced, seen automation replace entry-level roles, noticed that productivity gains don't reliably translate into wage gains for the people doing the work.

Same technology. Genuinely different sets of cases. Genuinely different vantage points.

A prediction market on the question wouldn't resolve this. It would have the capital-rich side — the researchers, the investors, the builders — collecting from the anxious side. Not because of insider information. Not because of drama inflation. But because their position in the economy gives them access to the benefits while the costs land somewhere else. The market doesn't correct for structural asymmetry in vantage point. It just weights bets by capital.

What the market actually knows

Three things can make a prediction market price mean something other than what it claims:

• The crowd's bias is stable and its natural counterweight is absent or undersized, so the price stabilises at the wrong number and stays there until enough money shows up to move it.
• The price reflects private information held by one or a few actors — not distributed signal from many. What looks like collective knowledge is concentrated knowledge, laundered through a mechanism that implies otherwise.
• The people with enough capital to move prices are structurally positioned to benefit from one outcome. No fraud required. The market just reflects their honest bets, weighted by their resources, which accrued because of where they stood.

None of these require anyone to be lying. They're all structural. And they share a common shape: the market produces a number that feels like distributed truth but is actually something more local — a crowd's systematic error, an insider's secret, a particular class of people's confident experience.

"The market knows something" sometimes means exactly that. Sometimes it means an insider knows something. Sometimes it means the crowd is predictably wrong and one model knows it. Sometimes it means the people with capital have structurally different interests from the people being affected. All three compress into a single price. The price doesn't tell you which.