On 9 January 2025, Émile Servan-Schreiber argued that social networks had failed an early promise: decentralised collective intelligence. In his view, platforms such as Facebook and Twitter had come to reflect the decisions of their owners rather than the collective judgement of their users. His criticism concerned who controls the process through which many contributions become a public account of the world.
On 27 September 2025, he made a related argument about access to forecasts. Responding critically to the US administration, he proposed The Forecasting Machine as a tool anyone could consult about future threats. His post went further, asserting advantages in speed, cost and intelligence over US intelligence estimates, but supplied no comparative evaluation. The defensible core of his proposal was broader access to a way of asking questions about the future.
Access is a distinct claim from accuracy
The two posts pose a useful question: does bringing many people or models together make intelligence available to the public, or merely concentrate influence in another institution? Servan-Schreiber’s answer was to advocate public access to forecasting. The September 2025 launch is covered separately; the issue here is the governance argument that accompanied it.
The governance question begins before a probability is calculated. Someone chooses which uncertainty to examine, which outcomes to distinguish and what evidence would count as resolution. Readers assessing a public forecast can ask how those choices affect the answer. Opening that process to scrutiny gives the proposal a practical meaning: people need to understand the question as well as see the final number.
A public forecast still needs scrutiny
Public access allows readers to examine an estimate, but accuracy still requires comparison with observed outcomes. The assessment should preserve the question and the information available when the forecast was made, so that later knowledge does not obscure the original uncertainty. Recording successive estimates also makes it possible to ask whether changes followed new evidence. These are practical ways to hold a public forecasting process accountable.
The thread connecting January and September is therefore a proposal about who can ask questions and inspect answers. Servan-Schreiber saw collective intelligence as something institutions should organise for wider use. The forecasting implication is concrete: a probability becomes more accountable when its question, evidence date, reasoning and eventual outcome can be checked by people beyond the organisation that produced it.
