Board gameEvents

Anonymisation research as an AnoMed board game at the AI Experience Day

November 14, 2023webadm

The amount of medically relevant sensitive data is growing rapidly, and machine learning methods (so-called AI methods) can process this data ever more usefully. However, these learning methods reveal information about the sensitive medical data they were trained on. So how can medical data be processed usefully while at the same time protecting sensitive information about patients? In recent years research has made promising progress on this so-called privacy problem; for most medical applications, however, a lot of research is still needed. This is exactly where AnoMed comes in as one of five competence clusters for anonymisation funded by the BMBF and the EU.

As part of the AnoMed project, the researchers present an interactive board game. In it they playfully demonstrate privacy risks that arise from training AI models. The game is based on cards with personal details about individuals, such as profession, interests and household situation. From these cards a player selects 5 cards and “trains a model”: the attributes directly determine the “model parameters” as the positions of game pieces. The other players, acting as attackers, then try to deduce from the positions of the game pieces what the private attributes of the cards were.

The simple, playful presentation highlights one of the known risks of complex AI models: even if the model parameters may seem incomprehensible to a human at first glance, they can still contain sensitive information about the training data. Sufficiently skilled or well-resourced attackers can read such information from the model. Admittedly, the number of training data points in practice is much larger (millions of data points instead of just 5 out of 15 cards), but the models also have a far greater number of parameters (millions of parameters instead of 4 colours). With the game the researchers hope to open a dialogue that both sparks a general interest in the topic and offers a platform for more detailed discussions.

AnoMed is part of the:Anonymization Research Network Logo
Funded by the Federal Ministry of Education and Research
Funded 2022-2025 by
Financed by the European Union