Martes, 28 de julio de 2026
CONVERSATIONS
Can a citizen network predict the future of war better than the CIA?
The CIA discovered many years ago that a diverse crowd could match the accuracy of its best analysts with far fewer resources. Today, prediction markets are used to answer questions about the economics of war or the impact of missiles in Ukraine. Can a betting platform predict the next major electoral upset better than the experts? We explore it with Emile Servan-Schreiber, founder of Hypermind, Europe’s leading collective-intelligence tool, in an in-depth conversation in 'Agenda Pública'.
Marc López Plana (left) interviews Hypermind CEO Emile Servan-Schreiber at beBartlet’s headquarters in Madrid.Agenda Pública / Tania Sieira
Prediction markets arrive in Spain. Hypermind, Europe’s leading collective-intelligence tool, and beBartlet, the Spanish public-affairs consultancy, announced yesterday at the Mobile World Congress an alliance to use this tool exclusively in strategic consulting across Europe. This technology, used by European governments to make better decisions, allows organisations, companies, and media outlets to aggregate intelligence to anticipate scenarios and make better decisions.
In this conversation between Marc López Plana, editor and director of Agenda Pública, and Emile Servan-Schreiber, founder of Hypermind, the conclusion is also that the media industry has been making a multi-billion-euro strategic mistake for decades: making its readers smarter and then doing absolutely nothing with that intelligence. Servan-Schreiber, a researcher with a PhD in cognitive science and a member of a historic dynasty of European journalists, creators of L’Express and Les Echos, has a formula to fix it. After 25 years refining collective-forecasting platforms, his diagnosis for the sector is blunt and revealing: "if you create something valuable and don’t use that value, it’s a waste".
The arrival of prediction markets in Spain through beBartlet opens a door to monetising aggregated knowledge. Instead of limiting themselves to static advertising, media outlets can turn their readers into predictive analysts, extracting high-value insights for companies and markets. Servan-Schreiber is clear: "I think the future of journalism will depend on gamifying the news". When users need information in order to bet or compete, engagement soars and the outlet regains its absolute relevance in the value chain.
But this new business model faces an imminent challenge: artificial intelligence is already here. Although it seemed impossible not long ago, technology is advancing at breakneck speed: "a year ago it wasn’t feasible, but now AI can make predictions as well as a diverse crowd of smart humans". If machines are a hundred times faster and cheaper, what role is left for the human reader in the prediction economy? A deep conversation to understand how the fusion of AI, social gamification and the press will define the economic model of the next decade.
You have spent 25 years building Europe’s leading prediction-market project. How did it start?
I started when I was working as a journalist. My family comes from Germany. My father created L’Express, my grandfather created Les Echos, and my uncle created L’Expansion. It’s a family of journalists. But I was never really a journalist: I was a scientist. I studied the brain — cognitive science — and earned a PhD in cognitive science at Carnegie Mellon. I began my career as an artificial intelligence engineer in the 1990s, during what people call the "AI winter". Back then, everybody thought we were clowns.
Eventually, I became a journalist to look for new ideas in a scientific magazine. I stumbled upon an article that suggested something simple: instead of using AI to scour the internet to see what’s happening — thirteen years ago, the technology wasn’t there to do what we do today — instead of using AI agents to go online and find information, maybe we can use the brains connected to the web to generate collective intelligence, especially about the future, which is the hardest problem.
As a scientist, I was interested in that possibility. As a journalist, I was interested in something else: we spend our time providing analysis and news and making people smarter — that’s what your craft is about — but you don’t do anything with it. Maybe you get a letter to the editor or something like that, but not much more. So we give people information through the media, and they give us back, through this prediction platform, a view of tomorrow.
Isn’t it enough that people can become smarter?
No, because if you create something valuable and you don’t use that value, it’s a waste. You make people smarter — and what do you get? A little bit of money, maybe a little bit of influence. But you don’t get to use that intelligence that is at your disposal to make yourself — or your organisation, or your newspaper — more effective.
Imagine you were aware of everything your readers know, and all the analysis they are capable of. How much value would it bring to your media organisation to leverage that? There was a famous CEO of Hewlett-Packard in the 1990s, at the start of the knowledge-management revolution inside organisations. He said: "If only HP knew what HP knows." There’s so much knowledge out there, but you don’t have a way to aggregate it and make it useful.
That was the idea: we give the news, and they give back predictions about what happens next. What happens to the Trump Board of Peace? How many countries are going to sign up? That’s tomorrow’s news — so we can have it today.
Some readers may think prediction markets are similar to polls. What is the difference between prediction markets and surveys or opinion polls?
It’s completely different. You have to realise that surveys and polls were invented less than 100 years ago: in 1936, by Gallup. Before that, do you think people weren’t making predictions about elections?
No…
Betting is probably the second-oldest profession in the world. We have records for fifteen presidential elections before Gallup invented polls, where people were making bets on Wall Street about who would be the next president. They were essentially running prediction markets on the street instead of on the internet.
The scientific record for those elections shows that the favourite of the betting markets won fourteen times out of fifteen. It was already quite accurate. In fact, when polls arrived, people thought: "Finally, something scientific to predict elections." But the track record shows that polls did not do any better, in terms of accuracy, than betting markets before polls existed.
Not only that: since polls were invented, betting markets became less efficient because they were polluted by bad polls. People look at a poll and say: "That’s what’s going to happen, so I’m going to bet this way." Then Trump is elected instead of Hillary Clinton and everybody is surprised.
We also have records from newspapers — the New York Post, for example — from before polls. What did they publish? Not polls: the odds that Roosevelt was going to win. Five-to-one that Roosevelt wins, and then the odds move compared to last week. They treated betting markets exactly like people treat polls as content today.
The second difference is the foundation. Polls rely on a representative sample of the population. If you want to say whether Spaniards prefer this candidate or that candidate, you need to make sure you represent every Spaniard who is likely to vote. Representation is the central principle.
In fact, we see that when people make a bet, the part of the brain linked to emotions is inhibited, while the part associated with rational thinking — the frontal lobe — lights up more. People are asked to give a prediction about what will actually happen on the ground. They have to be in empathy with reality, not ideology.
What is the incentive for people to participate in a prediction market?
There are four kinds of incentives, depending on what you’re trying to achieve.
One: rewards. The money you could make or the prizes you could win.
Two: recognition. For example, inside Google, when there is a prediction market — not about politics, but about the number of bugs in Gmail — people don’t care about money: they are already well-paid. They care about recognition, like a T-shirt that says: "I am the best forecaster in Google." It’s a form of social currency inside the company.
Three: relationships. When you’re betting against each other, you get to know each other. It’s a social activity: you enjoy a community of forecasters competing to see who’s the best. You have to negotiate about the right price for Trump to win the next election, or for Sánchez to remain in power — whatever it is.
Four: relevance. If I read an article in Agenda Pública about the automobile market and electric vehicles in Europe, and I’m genuinely interested in that industry, I already have ideas about what may happen, what should happen, or what is likely to happen. Prediction platforms give me a channel to express those ideas. That relevance — whether it’s work or personal interest — is the ability to express my thoughts there. Otherwise, I could only discuss them with my friends.
What is the real influence that prediction markets could have?
In politics, they can have as much influence as polls — perhaps more — because polls are a weak technology. That’s why you have poll aggregators: because no single poll is reliable enough. Not only that, but it’s not transparent. Every pollster has its own formula, and that formula is as secret as Coca-Cola’s recipe. You can’t fully trust them. We know some polls favour the right, some favour the left — and yet they all still influence people.
There was some manipulation that worked. It gave Trump a psychological advantage that may have made a difference in a very tight election, which he won by very few votes. Prediction markets can influence in that way.
But that’s not the influence I’m most interested in. I’m interested in how prediction markets can quantify the future so that people — especially companies and policymakers — have a better idea of risk. Policymakers often make decisions based on ideology rather than reality. Prediction markets offer a gamified way to consult what people on the ground think will happen. I think it’s a powerful way to choose policies to achieve specific goals.
Prediction markets and artificial intelligence: maybe we no longer need people to make predictions?
Let me explain. We just created a purely artificial forecasting machine based on AI: The Forecasting Machine. It’s very interesting. A year ago it wasn’t feasible, but now AI can make predictions as well as a diverse crowd of smart humans. In six months or a year, AI will be better; in two years, significantly better.
The main advantage is that AI doesn’t need rewards. It doesn’t care whether the prediction is about tomorrow or 100 years from now: it takes all predictions seriously. It’s faster — about 100 times faster than humans — and cheaper — much cheaper.
So why do we still need human forecasters? For different reasons and applications. If you want horizon scanning around your company, AI can do it. But if you want to engage people — because humans still live on the planet and you need to organise them into movements — you need human forecasters.
Right now, we have a project making predictions about the war in Ukraine called Glimt.nu. It’s part of Swedish government aid to Ukraine. The idea is to contribute not tanks or weapons, but intelligence.
Ukrainian analysts pose the questions that worry them, and we created a prediction-market platform with the Swedish Defence Ministry to invite Europeans — right now many Swedes and French, and hopefully some Spanish as well — to answer.
The questions include the economics of the war, Russian assets in Brussels, Russian inflation, oil flows through the shadow fleet. They also include military questions: how many missiles will fall on Kyiv, which town will fall next. And political questions: whether there will be a new US aid package, what will happen to Russian assets, or who will be elected in Poland or Hungary.
The point is that the wisdom of a crowd of European citizens, all interested in Ukraine, can contribute analysis from their perspectives. Those diverse perspectives are combined and delivered to Ukrainian analysts. It’s not about asking AI: it’s about engaging Europeans so they stay informed and engaged.
You mentioned intelligence agencies. How could prediction markets change how they work? It seems like a big shift…
They were the first natural customers. In the United States there’s an agency called IARPA — the Intelligence Advanced Research Projects Agency. About fifteen years ago they started a multi-year project to test whether crowd predictions could be useful for geopolitics.
Instead of Ukrainians, the CIA asked 100 questions per year to anyone interested — including readers of the New York Times, Foreign Policy, and other amateurs.
They compared the forecasts of 10,000 diverse amateurs with the predictions of intelligence analysts: same accuracy, but much cheaper. Within that crowd, about 2% were exceptional — around 30% better than analysts — and they were paid only $200 a year. Incredible.
Is the CIA worried about that?
No, it’s complementary. If it’s cheaper, you give them a badge saying "Recognized Super Forecaster". They put it on LinkedIn and become more valuable. You create forecasting competitions and recognise talent.
These super forecasters have expertise across domains: geopolitics, sports, economics, business. They know how to decompose a problem analytically and recombine it. We know how they think thanks to the Good Judgment Project. We can even teach AI to think like a super forecaster — although humans may still be better for another year or two.
The CIA can use super forecasters. Organisations like RAND Corporation use this community for US government projects. In France we do similar work with governments, and even with regional governments in Spain. Prediction markets can be applied in many ways: it’s not secret.
How do you see the relationship between prediction markets and journalism?
I think the future of journalism will rely on gamifying the news. For example, Polymarket partners with The Wall Street Journal: people need the news to predict market outcomes.
Polymarket’s business model makes money from players — it’s a betting platform. Another model we use allows a publication’s readership to play, creating a community. Companies can then buy access to insights from that community instead of limiting themselves to advertising.
So this is about engagement?
Exactly. The readership is an intelligence asset. If we can aggregate it, we can monetise it. There are many companies that could pay for those insights. Instead of just running ads, you get real predictive intelligence.
Thank you very much.
In this conversation between Marc López Plana, editor and director of Agenda Pública, and Emile Servan-Schreiber, founder of Hypermind, the conclusion is also that the media industry has been making a multi-billion-euro strategic mistake for decades: making its readers smarter and then doing absolutely nothing with that intelligence. Servan-Schreiber, a researcher with a PhD in cognitive science and a member of a historic dynasty of European journalists, creators of L’Express and Les Echos, has a formula to fix it. After 25 years refining collective-forecasting platforms, his diagnosis for the sector is blunt and revealing: "if you create something valuable and don’t use that value, it’s a waste".
The arrival of prediction markets in Spain through beBartlet opens a door to monetising aggregated knowledge. Instead of limiting themselves to static advertising, media outlets can turn their readers into predictive analysts, extracting high-value insights for companies and markets. Servan-Schreiber is clear: "I think the future of journalism will depend on gamifying the news". When users need information in order to bet or compete, engagement soars and the outlet regains its absolute relevance in the value chain.
But this new business model faces an imminent challenge: artificial intelligence is already here. Although it seemed impossible not long ago, technology is advancing at breakneck speed: "a year ago it wasn’t feasible, but now AI can make predictions as well as a diverse crowd of smart humans". If machines are a hundred times faster and cheaper, what role is left for the human reader in the prediction economy? A deep conversation to understand how the fusion of AI, social gamification and the press will define the economic model of the next decade.
Servan-Schreiber has a long track record in artificial intelligence. Photo: Agenda Pública / Tania Sieira
You have spent 25 years building Europe’s leading prediction-market project. How did it start?
I started when I was working as a journalist. My family comes from Germany. My father created L’Express, my grandfather created Les Echos, and my uncle created L’Expansion. It’s a family of journalists. But I was never really a journalist: I was a scientist. I studied the brain — cognitive science — and earned a PhD in cognitive science at Carnegie Mellon. I began my career as an artificial intelligence engineer in the 1990s, during what people call the "AI winter". Back then, everybody thought we were clowns.
Eventually, I became a journalist to look for new ideas in a scientific magazine. I stumbled upon an article that suggested something simple: instead of using AI to scour the internet to see what’s happening — thirteen years ago, the technology wasn’t there to do what we do today — instead of using AI agents to go online and find information, maybe we can use the brains connected to the web to generate collective intelligence, especially about the future, which is the hardest problem.
"We can create platforms where people bet against each other to make predictions about the future, and the collective intelligence will be greater than any individual"It was the first time you could massively connect a lot of human brains and see what you could do with that network. Before the web, it was impossible. As a scientist — and as someone who has studied the brain — I knew that all intelligence is based on collective interactions of many smaller agents that are not so smart on their own. Neurons, for example: there are 80 billion neurons in your brain. When it became possible to leverage the network of human brains created by the web, I thought: this is fantastic.
As a scientist, I was interested in that possibility. As a journalist, I was interested in something else: we spend our time providing analysis and news and making people smarter — that’s what your craft is about — but you don’t do anything with it. Maybe you get a letter to the editor or something like that, but not much more. So we give people information through the media, and they give us back, through this prediction platform, a view of tomorrow.
Isn’t it enough that people can become smarter?
No, because if you create something valuable and you don’t use that value, it’s a waste. You make people smarter — and what do you get? A little bit of money, maybe a little bit of influence. But you don’t get to use that intelligence that is at your disposal to make yourself — or your organisation, or your newspaper — more effective.
Imagine you were aware of everything your readers know, and all the analysis they are capable of. How much value would it bring to your media organisation to leverage that? There was a famous CEO of Hewlett-Packard in the 1990s, at the start of the knowledge-management revolution inside organisations. He said: "If only HP knew what HP knows." There’s so much knowledge out there, but you don’t have a way to aggregate it and make it useful.
That was the idea: we give the news, and they give back predictions about what happens next. What happens to the Trump Board of Peace? How many countries are going to sign up? That’s tomorrow’s news — so we can have it today.
López Plana focuses on the comparison between prediction markets and polls. Photo: Agenda Pública / Tania Sieira
Some readers may think prediction markets are similar to polls. What is the difference between prediction markets and surveys or opinion polls?
It’s completely different. You have to realise that surveys and polls were invented less than 100 years ago: in 1936, by Gallup. Before that, do you think people weren’t making predictions about elections?
No…
Betting is probably the second-oldest profession in the world. We have records for fifteen presidential elections before Gallup invented polls, where people were making bets on Wall Street about who would be the next president. They were essentially running prediction markets on the street instead of on the internet.
The scientific record for those elections shows that the favourite of the betting markets won fourteen times out of fifteen. It was already quite accurate. In fact, when polls arrived, people thought: "Finally, something scientific to predict elections." But the track record shows that polls did not do any better, in terms of accuracy, than betting markets before polls existed.
Not only that: since polls were invented, betting markets became less efficient because they were polluted by bad polls. People look at a poll and say: "That’s what’s going to happen, so I’m going to bet this way." Then Trump is elected instead of Hillary Clinton and everybody is surprised.
We also have records from newspapers — the New York Post, for example — from before polls. What did they publish? Not polls: the odds that Roosevelt was going to win. Five-to-one that Roosevelt wins, and then the odds move compared to last week. They treated betting markets exactly like people treat polls as content today.
The second difference is the foundation. Polls rely on a representative sample of the population. If you want to say whether Spaniards prefer this candidate or that candidate, you need to make sure you represent every Spaniard who is likely to vote. Representation is the central principle.
"Everybody has biases. But our job is not to express your preference; your job is to express your prediction"In betting markets, representation counts for nothing. What counts is knowledge. That’s why we recruit people who are knowledgeable about politics, who are interested, and who absorb new information all the time and act on it. Nobody is perfect. Everybody has biases: if you’re on the left or the right, you won’t necessarily have the same perspective or consume the same information. But here you’re not asked to express preference; you’re asked to express a prediction. We’re not calling on your emotions.
In fact, we see that when people make a bet, the part of the brain linked to emotions is inhibited, while the part associated with rational thinking — the frontal lobe — lights up more. People are asked to give a prediction about what will actually happen on the ground. They have to be in empathy with reality, not ideology.
Hypermind’s CEO explains how prediction markets work. Photo: Agenda Pública / Tania Sieira
What is the incentive for people to participate in a prediction market?
There are four kinds of incentives, depending on what you’re trying to achieve.
One: rewards. The money you could make or the prizes you could win.
Two: recognition. For example, inside Google, when there is a prediction market — not about politics, but about the number of bugs in Gmail — people don’t care about money: they are already well-paid. They care about recognition, like a T-shirt that says: "I am the best forecaster in Google." It’s a form of social currency inside the company.
Three: relationships. When you’re betting against each other, you get to know each other. It’s a social activity: you enjoy a community of forecasters competing to see who’s the best. You have to negotiate about the right price for Trump to win the next election, or for Sánchez to remain in power — whatever it is.
Four: relevance. If I read an article in Agenda Pública about the automobile market and electric vehicles in Europe, and I’m genuinely interested in that industry, I already have ideas about what may happen, what should happen, or what is likely to happen. Prediction platforms give me a channel to express those ideas. That relevance — whether it’s work or personal interest — is the ability to express my thoughts there. Otherwise, I could only discuss them with my friends.
What is the real influence that prediction markets could have?
In politics, they can have as much influence as polls — perhaps more — because polls are a weak technology. That’s why you have poll aggregators: because no single poll is reliable enough. Not only that, but it’s not transparent. Every pollster has its own formula, and that formula is as secret as Coca-Cola’s recipe. You can’t fully trust them. We know some polls favour the right, some favour the left — and yet they all still influence people.
"Every pollster has its own formula, and that formula is as secret as Coca-Cola’s recipe. You can’t fully trust them"For example, the New York Times reported that most Americans didn’t like what Trump did in his first year. Would you trust that? It wouldn’t be the same if it came from Fox News. In the last election, the prediction market — two weeks before — began favouring Trump, surprisingly, because the polls were showing a 50/50 split for three weeks. Nobody could differentiate.
There was some manipulation that worked. It gave Trump a psychological advantage that may have made a difference in a very tight election, which he won by very few votes. Prediction markets can influence in that way.
But that’s not the influence I’m most interested in. I’m interested in how prediction markets can quantify the future so that people — especially companies and policymakers — have a better idea of risk. Policymakers often make decisions based on ideology rather than reality. Prediction markets offer a gamified way to consult what people on the ground think will happen. I think it’s a powerful way to choose policies to achieve specific goals.
Servan-Schreiber backs open and reliable technologies. Photo: Agenda Pública / Tania Sieira
Prediction markets and artificial intelligence: maybe we no longer need people to make predictions?
Let me explain. We just created a purely artificial forecasting machine based on AI: The Forecasting Machine. It’s very interesting. A year ago it wasn’t feasible, but now AI can make predictions as well as a diverse crowd of smart humans. In six months or a year, AI will be better; in two years, significantly better.
The main advantage is that AI doesn’t need rewards. It doesn’t care whether the prediction is about tomorrow or 100 years from now: it takes all predictions seriously. It’s faster — about 100 times faster than humans — and cheaper — much cheaper.
So why do we still need human forecasters? For different reasons and applications. If you want horizon scanning around your company, AI can do it. But if you want to engage people — because humans still live on the planet and you need to organise them into movements — you need human forecasters.
"Ukrainian analysts pose questions they worry about, and we built a prediction-market platform with the Swedish Defence Ministry to invite Europeans to answer"Making people think about the future is a way to make them smarter today. You can’t be smart about the present if you don’t think about tomorrow — the environment, world peace, inequality, social organisation. The best way to do that is to gamify the effort, because thinking about the future is hard. Most people don’t spend much time thinking about the future, and that’s a problem for them and for the community.
Right now, we have a project making predictions about the war in Ukraine called Glimt.nu. It’s part of Swedish government aid to Ukraine. The idea is to contribute not tanks or weapons, but intelligence.
Ukrainian analysts pose the questions that worry them, and we created a prediction-market platform with the Swedish Defence Ministry to invite Europeans — right now many Swedes and French, and hopefully some Spanish as well — to answer.
The questions include the economics of the war, Russian assets in Brussels, Russian inflation, oil flows through the shadow fleet. They also include military questions: how many missiles will fall on Kyiv, which town will fall next. And political questions: whether there will be a new US aid package, what will happen to Russian assets, or who will be elected in Poland or Hungary.
The point is that the wisdom of a crowd of European citizens, all interested in Ukraine, can contribute analysis from their perspectives. Those diverse perspectives are combined and delivered to Ukrainian analysts. It’s not about asking AI: it’s about engaging Europeans so they stay informed and engaged.
You mentioned intelligence agencies. How could prediction markets change how they work? It seems like a big shift…
They were the first natural customers. In the United States there’s an agency called IARPA — the Intelligence Advanced Research Projects Agency. About fifteen years ago they started a multi-year project to test whether crowd predictions could be useful for geopolitics.
Instead of Ukrainians, the CIA asked 100 questions per year to anyone interested — including readers of the New York Times, Foreign Policy, and other amateurs.
They compared the forecasts of 10,000 diverse amateurs with the predictions of intelligence analysts: same accuracy, but much cheaper. Within that crowd, about 2% were exceptional — around 30% better than analysts — and they were paid only $200 a year. Incredible.
Hypermind has already worked with international intelligence agencies and now lands in Spain with beBartlet. Photo: Agenda Pública / Tania Sieira
Is the CIA worried about that?
No, it’s complementary. If it’s cheaper, you give them a badge saying "Recognized Super Forecaster". They put it on LinkedIn and become more valuable. You create forecasting competitions and recognise talent.
These super forecasters have expertise across domains: geopolitics, sports, economics, business. They know how to decompose a problem analytically and recombine it. We know how they think thanks to the Good Judgment Project. We can even teach AI to think like a super forecaster — although humans may still be better for another year or two.
The CIA can use super forecasters. Organisations like RAND Corporation use this community for US government projects. In France we do similar work with governments, and even with regional governments in Spain. Prediction markets can be applied in many ways: it’s not secret.
How do you see the relationship between prediction markets and journalism?
"Nobody would read a financial newspaper if there were no stock market: it would be boring. If you want to play the market, you need the paper to stay informed"The market exists, so the need for the paper exists.
I think the future of journalism will rely on gamifying the news. For example, Polymarket partners with The Wall Street Journal: people need the news to predict market outcomes.
Polymarket’s business model makes money from players — it’s a betting platform. Another model we use allows a publication’s readership to play, creating a community. Companies can then buy access to insights from that community instead of limiting themselves to advertising.
So this is about engagement?
Exactly. The readership is an intelligence asset. If we can aggregate it, we can monetise it. There are many companies that could pay for those insights. Instead of just running ads, you get real predictive intelligence.
Thank you very much.
Ahora en portada
27 de julio de 2026
27 de julio de 2026
Lo más leído