On 17 July 2026, Émile Servan-Schreiber announced that the Institut Diderot, a Paris think tank, had published a booklet based on a lecture he had given there. He is co-founder and Chief Scientific Advisor of The Forecasting Machine and CEO of Hypermind. Titled Mesures, démesures et modèles de l’intelligence collective (“Measures, excesses and models of collective intelligence”), the booklet is free to download as a PDF. It transcribes the interactive conference of Wednesday 5 November 2025, including the audience’s live poll answers and the questions from the floor. The full lecture is also on YouTube.
A philosopher changes his mind
The foreword is by the philosopher André Comte-Sponville, who calls it the lecture that surprised him most of all those he had attended at the Institute. The reason was not its quality or its format but its content. He writes that he had long been convinced, without having read him, that Gustave Le Bon was right that crowds are stupid, emotional and easy to manipulate. Servan-Schreiber argued almost the opposite. A group can be smarter than the average of its members, and even than its smartest member, but only if it is organised. Comte-Sponville’s summary of the lesson he took away is that collective intelligence is the opposite of collectivism.
What makes a group smart
Servan-Schreiber polled the room before giving each research result. On group IQ, 70% of the audience picked cohesion as the main driver. He then presented the 2010 study in Science by researchers at MIT and Carnegie Mellon that first measured a “collective intelligence factor” in groups. By his account, cohesion and motivation showed no correlation with group performance. The average intelligence of members mattered a little, the presence of one very bright member a little more, and a majority of women more still. The strongest link was with equal sharing of speaking time, along with members’ ability to read others’ emotions.
Leadership got the same treatment. Citing a 2011 study, he said the most effective combination is a participative leader with proactive staff. Most of the room guessed that the least effective would be an authoritarian leader with proactive staff. In the study, it was a participative leader with reactive staff, who wait to be told what to do.
The recipe
After a section on Galton’s ox and the trade-off between expertise and diversity (also discussed in his podcast appearance this summer), Servan-Schreiber set out his central claim:
Il y a une recette à l’intelligence collective. Si on la suit, on produit de l’intelligence.
Émile Servan-Schreiber, Institut Diderot, 5 November 2025
In our translation: “There is a recipe for collective intelligence. If you follow it, you produce intelligence.” Servan-Schreiber contrasted that recipe with the groupthink produced when its conditions are ignored.
The recipe has three ingredients:
- Diversity of opinion, because nobody sees the whole of a complex problem alone.
- Independence, which needs rules, because the urge to conform is, in his words, written into our DNA.
- Objective aggregation: a vote to choose, an average to estimate, a bet to forecast.
Viser le consensus, c’est tuer l’intelligence collective dans l’œuf.
Émile Servan-Schreiber, Institut Diderot, 5 November 2025
In our translation: “Aiming for consensus kills collective intelligence in the egg.” Divergence should instead be valued, then combined objectively.
Forecasting where no one is an expert
His main forecasting example was a 2019–2020 contest run with Johns Hopkins University, later published in BMC Public Health. In his account, several hundred participants, about 70% of them working in health, forecast the severity of infectious-disease outbreaks on about 60 questions over 15 months, on a platform built by Hypermind. Most individuals did no better than chance. The simple average of everyone’s forecasts was beaten by only 1% of participants. A weighted average, which gave more weight to recent forecasts and to people with better track records, was beaten by no one. The study itself is covered in a separate article. He also pointed to Hypermind’s calibration since 2014 and to Glimt, the citizen forecasting project for Ukraine run by Sweden on a Hypermind platform.
AI as “collective intelligence in a box”
In the questions from the floor, Servan-Schreiber explained that since 2024 his team has built a new, fully AI-based product, The Forecasting Machine. It does what the best human forecasters do, but automatically. He described AI as collective intelligence in a box: today’s algorithms are not so different from those of 40 years ago, he said, and what changed is the mass of digitised human knowledge available to train them. He then turned the question of whether humans are still needed on its head. AI has no future without fresh human intelligence, he argued, because models trained on machine-generated content degrade within a couple of generations. AI therefore depends on the diversity of human reasoning, observation and experience.
He closed the lecture with a sentence meant to sum it all up:
L’intelligence du groupe, c’est de demander à chacun, et l’intelligence de chacun, c’est de demander au groupe.
Émile Servan-Schreiber, Institut Diderot, 5 November 2025
In our translation: “The intelligence of the group is to ask each person, and the intelligence of each person is to ask the group.”
