Autonomous driving, chatbots that talk like human beings, robots that care for the elderly – just how much AI is set to change our lives becomes clearer with every day that passes. Blatant security gaps und privacy issues that come with this new technology, however, keep making the headlines. In the ELSA project, partners from industry and science have teamed up to tackle the challenge of making AI applications safer and more secure, fostering the use of modern AI solutions.
A compromise between privacy protection and data utility
One of the core problems in the use of machine learning is its insatiable hunger for data. Researchers from around the world are trying to solve the question how the huge amounts of data used to train AI can be shared securely and analyzed effectively without violating the privacy of the people behind it. And, ultimately, how to guarantee that their privacy cannot be violated. With methods such as differential privacy, it is theoretically possible to guarantee privacy protection in data processing. In practice, however, existing methods are often too compute-intensive, too slow or not yet fully developed. Also, the more effective the methods for anonymizing data and concealing existing relationships, the less meaningful the data becomes.
Defining joint problems and finding solutions
The ELSA project partners are examining, among other things, what options there are for sharing data for different applications using different methods, and for learning from it, without revealing too much about individual people. Antti Honkela, Professor of Data Science at the University of Helsinki, opened this first ELSA workshop. Researchers from INRIA (Institut national de recherche en sciences et technologies du numérique), EPF in Lausanne, IIT (Italian Institute of Technology), Unimi (Università degli studi di Milano), the University of Helsinki and CISPA then introduced the specific research questions they are currently working on. The project partners also discussed the specific privacy challenges that come with the use of AI in the areas of health, robotics, and document security. The ensuing discussions provided new impetus for further research.
Jaakko Lähteenmäki from the VTT Technical Research Centre of Finland contributed to the workshop as a guest speaker. Within the framework of the European Health Data Space (EHDS), he examines how health data can be analyzed and shared, particularly across countries, so as to improve disease diagnosis and treatment.
ELSA coordinator Mario Fritz is very pleased with the kick-off in Helsinki: "We had the opportunity to exchange our ideas on privacy protection, especially regarding the health sector, at a European level. I am grateful for being able to advance the vision of our ELSA – European Lighthouse on Secure and Safe AI together with so many great partners."
