On 7 April 2020, Émile Servan-Schreiber shared JEDI’s announcement of a scientific challenge against COVID-19. He presented it as a challenge to collective intelligence. The organizer’s post promised awards of up to €2 million for winning scientific teams and described an eight-week effort to analyse billions of molecules that might block SARS-CoV-2.
The announcement brought together high-performance computing, molecular biology, epidemiology and artificial intelligence. Its ambition was to accelerate the route toward a treatment. At that point, the post documented a proposed research effort, rather than a successful medicine or a result from a forecasting market.
A common question, several computational approaches
On 20 May, the Max Planck Institute for Intelligent Systems reported that the challenge had launched at the beginning of the month. Its account explained that teams would screen large molecular libraries and compare results to assemble a list of promising candidates. This moved the initiative from an early announcement toward a defined research process.
The published stage-one rules made the combination more specific. They required three independent computational methods and described comparing different simulation approaches to reduce errors in assumptions. Those requirements show why the effort was more than many laboratories working separately: the design called for their results to be compared and combined before experimental screening.
What the later research found
The research paper A community effort in SARS-CoV-2 drug discovery, first published online on 13 October 2023, reports 130 registered teams, with 31 meeting the submission deadline. The participating teams proposed 639,024 molecules. The organizing effort subsequently synthesized and tested 878 compounds; the paper reports 27 with weak inhibition or binding in experimental assays.
Those are research findings, not evidence of an approved treatment. The paper also identifies a limitation of the competition format: initial communication and sharing between teams were restricted. Its authors suggest that earlier open communication after submissions would have improved the process. The later account therefore adds both a concrete output and a lesson about how incentives can affect collaboration.
Combining predictions requires an external test
The relevance to collective intelligence is methodological: a shared problem, varied approaches, a way to combine their outputs and a test against external evidence. The journal’s coverage of infectious-disease forecasting examines another application of collective judgement. Here the test involved laboratory measurements. A computational shortlist could guide experiments, but it could not substitute for them. The later paper makes that difference concrete.
