Nicosia: The Cyprus Center for Trustworthy AI (CyCAT) at the Open University of Cyprus has announced the successful conclusion of its PINNACLE project, aimed at enhancing public engagement in AI evaluation for education and system benchmarking. Supported by a 150,000 Euro fund from the Cyprus Research and Innovation Foundation (BRIDGE2HORIZON/0823E/0203), the project spanned 18 months and focused on initial scientific research, culminating in a proposal for further work under Horizon Europe.
According to Cyprus News Agency, the rise of Generative AI and Foundation Models has led to an influx of AI-enabled products and services, such as Google's AI-generated search answers and recommendation systems on platforms like Netflix and Spotify. These systems are becoming an integral part of daily life, yet many users lack the tools to assess their fairness, privacy safeguards, and overall behavior. The PINNACLE project addressed this by involving users directly in the evaluation process of AI systems.
The project developed mechanisms for public participation in AI evaluation, aligning with the EU's vision for Trustworthy AI, which is outlined in the EU AI Act. This vision mandates that AI systems must be technically robust, fair, non-discriminatory, and respectful of user privacy. The methodology introduced by PINNACLE allows public users to systematically evaluate the trustworthiness of AI-enabled systems they interact with.
As part of the project, CyCAT offered an online course titled "AI in Everyday Life" on two occasions, which tested different approaches to public AI evaluation. The first course encouraged participants to use a bottom-up approach, evaluating AI applications weekly based on ethical principles. The second course used a top-down approach, where participants evaluated specific chatbots like ChatGPT, Gemini, or Copilot through guided tasks. Over 500 participants from Cyprus and Greece contributed user-sourced evaluations of AI systems during these courses.
In addition to the course outcomes, the PINNACLE project established a framework for user-centered AI evaluation, resulted in nine peer-reviewed publications, and coordinated a proposal for Horizon Europe that exceeded the required threshold, setting the stage for future resubmission. These accomplishments have bolstered CyCAT's involvement in Horizon Europe and expanded its collaborative network throughout Europe.