Recent studies show that risks related to security, privacy, and transparency are among the main concerns that could slow down the future development of AI. The project TRUMAN (Trustworthy Human-Centric Artificial Intelligence), in which the subject of Computer Science at Karlstad University participates as a partner, aims to support trustworthy AI solutions that contribute to more human-centric AI.
Within the TRUMAN project, Karlstad University is responsible for developing methods to explain the underlying technologies and risks to users, examining how AI solutions affect people, and involving them in improving the models.
“Manipulating AI can be life-threatening – for example, if machine learning models for self-driving cars are tampered with so that they misinterpret traffic lights. Machine learning models can also leak sensitive personal data they were trained on. Therefore, it is crucial to design AI systems that are robust against security and privacy attacks. Human factors are also essential for making AI systems understandable, transparent, and trustworthy – and for “keeping humans in the loop”. We want to conduct research that makes AI more sustainable for the future by following ethical guidelines for trustworthy AI and the EU’s AI Act,” says Simone Fischer Hübner, Professor of Computer Science and research leader.
Truman, which started in July at the research center Eurecom in France, aims to develop technical methods to strengthen AI systems’ resilience against threats to security, privacy, and fairness, and to increase users’ trust in these systems. This will be achieved by addressing all phases of the AI lifecycle – from data collection to training and deployment.
The project develops customized solutions to protect against existing and emerging types of attacks on AI systems, including those targeting privacy, fairness, and system security. It will also produce comprehensive guidelines and recommendations for conducting risk analyses and building trustworthy and robust AI systems.
Truman is one of three major EU-funded research and innovation projects within the Horizon Europe program that were lunched this summer to support robust, safe and reliable AI solutions to help create human-centric AI, with a total EU contribution of €21.4 million. The project runs for three years, until 2028.
Project website: https://www.truman-horizon.eu/
