AI · PREDICTION MARKETS · PROBABILITY · RISK
Kalshi and Polymarket turn uncertainty into prices as AI sharpens the odds
Prediction markets are becoming a serious financial business just as AI
systems grow better at forecasting everything from elections and sports
to markets and weather. Kalshi and Polymarket can turn collective
expectations into probabilities, while machines can process more signals
than any individual bettor. Yet neither removes the thing every wager
ultimately depends on: an event that has not happened yet.
A strange thing happens when prediction becomes extremely good.
The bet does not disappear.
It moves.
The obvious opportunity disappears first.
The price changes.
Everyone receives more information.
And the wager survives somewhere inside what remains unknown.
That is becoming increasingly important as artificial intelligence enters
one of humanity's oldest activities:
trying to predict what happens next.
Kalshi and Polymarket are putting prices on the future
Prediction markets such as Kalshi and Polymarket allow people to trade
contracts linked to future events.
Will something happen?
Will a company reach a particular valuation?
Will a team win?
Will a political event occur before a certain date?
Will temperatures cross a threshold?
Participants buy and sell contracts whose prices can be interpreted as
market-implied probabilities.
A contract trading around 70 cents on a dollar payout can broadly be read
as the market assigning something close to a 70% chance to that outcome,
although liquidity, fees, market structure and trading behaviour can make
the interpretation less exact.
The market is effectively asking:
What is uncertainty worth right now?
AI changes how quickly that question can be answered
Human forecasters have always searched for signals.
Form.
Weather.
Injury reports.
Polling.
Economic data.
Historical performance.
News.
Behaviour.
Rumour.
Price movement.
AI can process these categories at a scale and speed far beyond any
individual person.
A model can analyse thousands of previous matches.
It can compare weather conditions against historical performance.
It can read breaking news.
It can detect statistical patterns hidden inside enormous datasets.
It can update a forecast repeatedly as conditions change.
That makes prediction sharper.
It does not make the future certain.
The difference between probability and prophecy
This distinction is easy to lose.
If an AI system says there is an 82% chance that something happens,
humans tend to hear:
It is going to happen.
That is not what 82% means.
Properly interpreted, it means that across a sufficiently large number
of comparable predictions made with similar calibration, events assigned
an 82% probability should occur about 82% of the time.
The remaining probability did not disappear.
It is the reason the bet still exists.
A favourite can still lose
Sport makes the distinction obvious.
A football team may be superior by almost every measurable indicator.
Better players.
Better recent form.
Better expected-goals data.
Home advantage.
Fewer injuries.
Better historical performance against the opponent.
A sophisticated model might therefore give it a very high probability
of winning.
Then its goalkeeper is sent off after seven minutes.
A defender slips.
A shot deflects.
It rains harder than expected.
The underdog scores.
Prediction never promised that the unlikely outcome was impossible.
It merely told us that it was unlikely.
The better AI becomes, the more valuable the remaining mystery becomes
This sounds paradoxical.
If AI improves prediction, surely mystery should become less valuable.
In one sense, yes.
Poor information becomes easier to exploit.
Obvious mispricing may disappear faster.
Markets can incorporate new information more quickly.
But when every participant has access to increasingly capable
forecasting tools, the advantage shifts toward the things models still
struggle to know.
New information.
Hidden information.
Human behaviour.
Sudden shocks.
Rare events.
Events without enough historical precedent.
The unknown does not vanish.
It becomes more concentrated.
Prediction markets already combine humans and machines
A prediction market should not be imagined as one giant forecasting
computer.
It is a collection of participants holding different information,
models, beliefs and incentives.
One trader may use an AI model.
Another may know an industry extremely well.
Another may follow local reporting unavailable to most participants.
Another may simply believe the market has overreacted.
Their disagreement creates trading.
Their trading creates a price.
The price becomes a compressed representation of competing beliefs.
That is what makes prediction markets interesting.
They do not need everybody to agree.
They need people to disagree enough to trade.
Perfect prediction would destroy the bet
Imagine a machine that genuinely knew tomorrow's result with absolute
certainty.
Not 80%.
Not 99%.
100%.
Once everyone trusted that machine, there would be almost nothing left to
trade.
The outcome would no longer be uncertain.
A contract paying £1 if the event occurs would simply be worth £1 if the
machine said yes and almost nothing if it said no.
The meaningful wager disappears when uncertainty reaches zero.
Betting does not require ignorance. It requires uncertainty.
That is why mystery makes the market
People sometimes describe gambling as a contest between knowledge and
luck.
Prediction markets reveal a more interesting relationship.
Knowledge determines the price.
Uncertainty determines whether the contract still has something to
resolve.
If everyone knew exactly what would happen, there would be no disagreement
worth trading.
Mystery is not an accidental weakness in the market.
It is part of the product.
AI can also create false confidence
There is another danger.
Sophisticated models can make probability look more authoritative than
it really is.
A forecast of 63.7% looks scientific.
The decimal point suggests precision.
But the result is only as useful as the assumptions, data and model
underneath it.
If the model has poor data, the number can be precisely wrong.
If the world changes, historical relationships can break.
If important information is missing, processing more of the wrong
information does not solve the problem.
The model cannot learn tomorrow from yesterday when tomorrow is different
Most forecasting systems learn from patterns in previous information.
That works extremely well when the future behaves enough like the past.
But some of the most consequential events are consequential precisely
because they break those patterns.
A sudden war.
A political assassination.
A previously unknown disease.
A market crash.
An unexpected technological breakthrough.
A rule change.
A human decision nobody anticipated.
Historical data can prepare a system for categories of surprise.
It cannot guarantee advance knowledge of every surprise.
Humans have another advantage: sometimes they know something the model does not
A trader may know that a chief executive sounded unusually uncertain in
an industry meeting.
A football analyst may know that a player technically listed as fit has
looked uncomfortable in training.
A local political reporter may understand voter sentiment before national
polling reflects it.
This is not mystical intuition.
It is information that has not yet entered the machine's usable dataset.
Of course, humans are also exceptionally good at imagining signals that
do not exist.
That is why neither human intuition nor AI prediction should automatically
be treated as truth.
AI may ultimately make markets harder to beat
If large numbers of participants use increasingly capable forecasting
models, easily exploitable mistakes should theoretically be corrected
faster.
Suppose AI discovers that a market is underpricing rain probability for a
particular weather contract.
Traders buy.
The price moves.
The opportunity begins disappearing because people acted on it.
Intelligence changes the market it is trying to exploit.
This is one of the central differences between prediction and ordinary
factual knowledge.
Knowing yesterday's temperature does not change yesterday's temperature.
Identifying a profitable probability error can change the price almost
immediately.
The machine can become part of the thing it predicts
This becomes even stranger in financial markets.
Imagine thousands of AI systems independently conclude that an asset is
likely to rise.
They begin buying.
Their purchases contribute to the price rising.
The prediction has influenced its own outcome.
That feedback loop makes forecasting social systems fundamentally
different from predicting an eclipse.
Humans and machines react to forecasts.
Those reactions change the environment being forecast.
Kalshi's growth shows how valuable uncertainty is becoming
Prediction markets are moving beyond their academic origins.
Kalshi has expanded rapidly across event-based trading, while Polymarket
has become one of the most recognisable global platforms for market-based
forecasts.
Their growth reflects something larger than gambling enthusiasm.
There is commercial value in continuously pricing uncertainty.
Markets want to know what may happen next.
Businesses want to know.
Governments want to know.
Investors want to know.
Insurers want to know.
Ordinary people want to know.
Increasingly, they can trade on the answer.
Prediction is expanding beyond sport and politics
Event markets can now cover subjects ranging from company valuations and
economic releases to technology, weather and entertainment.
That changes the meaning of the word bet.
A wager can increasingly resemble a financial instrument for expressing
a view about a real-world event.
The distinction is legally important and remains contested in some
jurisdictions.
But conceptually the attraction is straightforward:
uncertainty becomes tradable.
What happens when everybody has the same AI?
This may eventually become the more interesting question.
Imagine every trader has access to a model of roughly equivalent quality.
Everyone receives the same weather forecasts.
The same polling analysis.
The same statistical breakdown.
The same injury information.
The same news summaries.
AI then stops being the advantage.
It becomes infrastructure.
The edge moves elsewhere.
Better data.
Earlier information.
Better judgement about model weakness.
Better understanding of rare events.
Or simply willingness to take a position everyone else's model dislikes.
The contrarian becomes more interesting in a machine-consensus world
If every sophisticated system reaches roughly the same conclusion, anyone
betting against that conclusion will appear irrational.
Usually they may be.
But occasionally they will possess something important that the consensus
missed.
Markets need those people.
Without disagreement, there is no trade.
A world of perfect machine consensus would therefore create an unusual
tension.
The better prediction becomes, the more attention may eventually shift
toward the remaining minority probability.
Reputation will matter too
If anyone can produce an AI forecast, the identity and history behind a
forecast may become more important.
Did this forecaster predict similar events accurately before?
Were failures preserved as well as successes?
Was the prediction made before the outcome became obvious?
Has the methodology changed?
Does the model have a documented record?
Can that record be independently verified?
A screenshot saying “I predicted this” after the event is almost
worthless.
A timestamped history of forecasts is something different.
In a world overflowing with prediction, provenance becomes part of
credibility.
AI will not kill the bet by becoming smarter
It may change what people bet on.
It may change the speed at which odds adjust.
It may eliminate weak strategies.
It may make professional forecasting vastly more sophisticated.
It may even make certain outcomes so predictable that meaningful betting
around them becomes unattractive.
But as long as the future remains genuinely unresolved, another
probability remains available.
The final bet is always against uncertainty
Humans have spent centuries trying to beat uncertainty.
Mathematics helped.
Statistics helped.
Computers helped.
Big data helped.
Artificial intelligence will help enormously.
But prediction and certainty remain different things.
AI can tell us which outcome appears more likely.
Markets can turn that expectation into a price.
Traders can search for mistakes in that price.
Then reality arrives.
And reality does not care how sophisticated the forecast was.
AI sharpens the odds. Mystery makes the bet.
Perhaps that is what betting has always been.
Not an attempt to know the future.
An attempt to decide how much confidence we should place in the part of
it we think we understand.
Editorial note: Prediction-market prices are market
prices, not guarantees or necessarily precise objective probabilities.
They can be affected by liquidity, participant composition, market rules,
trading behaviour and available information. Artificial intelligence can
improve forecasting but does not eliminate model error or uncertainty.
Gambling and event-contract regulation differs significantly between
jurisdictions.
