September 30, 2026
For decades, predicting the future was considered one of the areas where human judgement would remain particularly difficult to replace.
Markets are uncertain.
Politics is unpredictable.
Companies make unexpected decisions.
Wars and crises can change assumptions overnight.
Yet a London AI startup is now building a business around the idea that artificial intelligence can become significantly better at forecasting complex events.
Mantic, founded by former Google DeepMind researcher Toby Shevlane and Ben Day, has raised $25 million in seed funding after its AI system outperformed human participants in the 2026 Metaculus Cup, a forecasting competition involving hundreds of human forecasters.
The development points toward an unusual new category of AI: systems whose primary purpose is not generating text or images, but estimating what is likely to happen next.
Forecasting is different from ordinary prediction
The distinction is important.
An AI model answering a question about history is retrieving or reasoning about existing information.
A forecasting system has to operate under uncertainty.
It may have to estimate the probability of an election outcome, a market event, a company’s product launch or a geopolitical development.
The system cannot simply know the answer.
It has to assign probabilities.
That makes forecasting closer to decision science than conventional chatbot interaction.
Mantic says its systems specialise in adapting frontier AI models for forecasting and repeatedly improving their performance using historical data and feedback.
The competition that attracted attention
The company’s breakthrough came during the 2026 Metaculus Cup.
According to Mantic, its system outperformed all 676 human participants in the competition. Reuters independently reported that the AI entrants collectively marked a first in which technology dominated the competition.
That does not mean AI can reliably predict the future.
It does not.
Forecasting remains inherently uncertain, and the outcome of one competition cannot establish that an AI system will outperform humans across every forecasting domain.
But the result demonstrates something important: AI systems can now compete seriously in tasks that require probabilistic judgement about future events.
That is a different capability from simply generating plausible text.
Why companies are interested
Forecasting has obvious commercial applications.
A retailer wants to know what demand will look like next quarter.
A manufacturer wants to anticipate supply-chain disruptions.
A financial institution wants to estimate market-moving events.
A company considering an acquisition wants to understand how competitors might respond.
A government agency may want to analyse potential geopolitical scenarios.
The better those forecasts become, the more valuable they can be.
Mantic says companies and financial firms are already using its systems for areas including geopolitics, macroeconomics, business decisions and market-related events.
That helps explain why investors are willing to fund a company operating in what is still an emerging category.
AI does not need to know the future to be useful
The commercial value of forecasting does not require perfect accuracy.
A human decision-maker rarely needs certainty.
They need better information about probabilities.
Suppose a company is considering entering a new market.
A forecasting system could estimate the probability of different regulatory outcomes, competitor responses and market developments.
The executive still makes the decision.
But instead of relying exclusively on intuition or internal debate, the company has another analytical input.
This is one reason forecasting AI could become particularly influential in corporate planning.
The problem with forecasting AI
There is also a fundamental limitation.
Forecasting systems can become overconfident.
Historical patterns do not always survive major disruptions.
A model trained on previous economic crises may not correctly predict an entirely new crisis.
Political systems can change.
Technologies can emerge unexpectedly.
And people can react to predictions themselves.
If a sufficiently influential forecasting system predicts that an event is likely, its users may change their behaviour, potentially affecting the probability of the event occurring.
That creates a feedback problem.
The forecast is no longer simply observing the future.
It can become part of the environment it is attempting to predict.
The rise of machine-generated probabilities
This could eventually change how organisations make decisions.
Today, forecasts are often produced by economists, analysts, consultants, intelligence specialists and internal strategy teams.
AI systems could increasingly become another layer in that process.
Instead of asking an AI:
“What do you think will happen?”
organisations may ask:
“Give me probability distributions for the possible outcomes, explain the evidence, identify the variables that would change your forecast, and update the forecast as new information arrives.”
That is a much more sophisticated use of AI.
It also makes the system easier to audit.
A probabilistic forecast can be measured against reality.
If a system repeatedly assigns a 70% probability to events that happen only 20% of the time, its calibration is poor.
That makes forecasting AI unusually testable compared with many generative AI applications.
The London startup’s new funding
Mantic’s $25 million round was led by Radical Ventures, with participation from investors including Microsoft’s M12, Thinking Machines Lab and Balderton Capital.
The company says it will use the funding to expand its team and increase the compute and data available for its forecasting systems.
That last point is particularly revealing.
The future of forecasting AI may depend not only on better algorithms but also on access to large quantities of high-quality information and enough computing power to continually update predictions.
Europe has an interesting position in this market
Mantic is headquartered in London, placing the company inside one of Europe’s largest financial and technology ecosystems.
That is strategically useful.
Forecasting has immediate applications in finance, insurance, corporate strategy and government.
London also provides access to large financial institutions that are natural customers for probabilistic decision-support systems.
If forecasting AI becomes a significant commercial category, Europe could therefore develop an important niche even without producing the world’s largest general-purpose language models.
The technology does not necessarily require a trillion-dollar consumer platform.
It requires reliable models, good data and organisations willing to use probability as part of decision-making.
The real breakthrough may be calibration
The most interesting development may not be that AI can sometimes beat humans.
Humans are already bad at many probabilistic tasks.
The bigger breakthrough would be an AI system that can consistently recognise when it is uncertain.
That means distinguishing between:
- situations where historical evidence is strong;
- situations where several outcomes remain plausible;
- situations where the available information is insufficient;
- and situations where a completely new event could invalidate existing assumptions.
Such systems could become valuable precisely because they know the limits of their predictions.
A new AI market is emerging
The first generation of generative AI companies built systems that could create.
The next generation may increasingly build systems that estimate.
Estimate demand.
Estimate risk.
Estimate market movements.
Estimate geopolitical developments.
Estimate the probability of business outcomes.
That does not mean machines have learned to see the future.
They have not.
But they are becoming increasingly capable of turning enormous quantities of information into explicit probability estimates.
And once businesses start making decisions based on those estimates, forecasting could become one of the most commercially important applications of AI.
The future, ironically, may become another dataset for artificial intelligence.
Why it matters: Mantic’s progress suggests that AI is moving beyond generation and into probabilistic decision support. The technology remains uncertain and should not be treated as an oracle, but forecasting could become a major AI application across finance, business, government and supply chains.
Laat een reactie achter