In 2023, 1,800 volunteers classified simulated neutrino events from Antarctic ice data; 128,000 assessments helped scientists compare human judgments with AI

Beneath two kilometres of Antarctic ice lies one of the most bizarre scientific experiments on our planet. A small city-sized detector is set up to observe some of the most elusive particles in the universe, which can pass through entire planets without interacting. However, until now only physicists and supercomputers had access to the data. But in 2023, the IceCube Neutrino Observatory took a refreshingly traditional approach and asked members of the public to help.

The results have now been published, and they suggest that non-experts can still contribute meaningfully to cutting-edge physics, even alongside advancing machine learning. According to a 2024 peer-reviewed study titled “Citizen science for IceCube: Name that Neutrino,” published in The European Physical Journal Plus, 1,800 participants (most with no background in physics) classified over 128,000 simulated neutrino events between March and September 2023 and produced judgments that lined up closely with those of a specially trained deep neural network. This is a notable result at a time when headlines often suggest AI is replacing human judgment outright.

What did people actually do?


The project, called ‘Name that Neutrino’, was hosted on the Zooniverse website run by Oxford and Chicago. People were shown short clips from simulated IceCube data and asked to classify the light patterns. This was to see if they could identify real neutrino signals. This may seem easy, but the neutrinos are subatomic particles formed when stars explode and black holes form, so detecting them requires identifying faint light signals in frozen water.

Project lead and Physicist Christina Love, Drexel University, described the work, in IceCube’s own project announcement, as having “demonstrated the feasibility of using the citizen science approach to classify IceCube data.” That is, by definition, a successful proof of concept: IceCube’s own summary called the first round a success, not a draw between humans and machines.

The reality of citizen science

Citizen science has something of a marketing problem. It is often portrayed as everyone having a chance to contribute to world-changing science from home, and while that isn’t exactly untrue, there is a lot of structure behind it. The Zooniverse isn’t a small business; it’s backed by the University of Oxford, Adler Planetarium, NASA, and the UK’s National Research Funders amongst others.

The platform itself has now logged over one billion classifications from over three million users, across hundreds of different projects ranging from classifying galaxies to monitoring wildlife. “Name that Neutrino” is modest by comparison. Its real significance was that a working physics collaboration was willing to publish results built on non-expert pattern recognition at all. In physics, it is very important that classifications are correct, as incorrect classifications can produce incorrect conclusions about where in the universe a particle came from.

Why this matters

For young people who have spent the past few years weighing both AI hype and AI anxiety, this is a refreshing story. This was not a competition between humans and machines. The classifications were compared to those of a deep neural network specifically to identify where each was lacking, and the results showed that both needed improving.

This is a more realistic narrative than most tech stories: humans and AI were each used to catch what the other couldn’t, and the scientists still had to decide which source to trust in any given case. Since 2023, the project has been improved and relaunched with cleaner data and better tutorials, a sign the collaboration treated the first round as a successful proof of concept worth building on, rather than a finished product. That continued search for refinement fits a project built around chasing some of the universe’s most elusive particles.