The AI risks in very large online platforms and search engines

This policy brief focuses on the integration of AI technologies in Very Large Online Platforms (VLOPs) and Very Large Online Search Engines (VLOSEs), which serve more than 45 million users monthly under the Digital Services Act (DSA). It highlights the key AI technologies, including recommender systems, information retrieval systems, and generative AI, which shape user experiences and impact the dissemination of information. Recommender systems leverage user data to deliver personalised content, while information retrieval systems prioritise relevant documents in response to user queries. Generative AI, a transformative technology, enriches digital content but introduces risks such as hallucinations and the proliferation of misinformation, including deepfakes. Despite the benefits of AI integration, VLOPs and VLOSEs face significant systemic risks. These include privacy and security vulnerabilities, threats to user autonomy, dissemination of harmful content and the addictive nature of these platforms.

The brief discusses the growing concerns about AI-driven rabbit holes that steer users toward extreme content and the potential for generative AI to spread disinformation. Addressing these risks requires a human-centred approach to AI regulation, emphasising ethical design, transparency, user control and human oversight. The policy brief calls for improved training of AI systems with diverse, high-quality data, collaboration among stakeholders, and implementation of explainability techniques to support user control and human oversight. It underscores the need for ongoing policy development to harmonise the DSA with the AI Act, ensuring that platforms and AI systems operate within a legal framework that protects individuals and society while enabling technological innovation.