Technical

Recommender Systems

AI systems that suggest content, products, or services to users based on analysis of preferences, behavior, and patterns.

Definition

Recommender Systems are AI-powered systems that analyze user data to suggest relevant content, products, services, or information. They are fundamental to social media, e-commerce, streaming services, and content platforms.

Regulatory Status: Recommender systems face varied regulatory treatment:

  • EU AI Act: Generally not high-risk, but subject to transparency requirements under Article 50 when interacting with users
  • EU DSA: Digital Services Act imposes significant obligations on recommender systems used by very large online platforms (VLOPs)
  • Transparency: Platforms must explain main parameters and provide at least one non-profiling-based option

DSA Requirements for VLOPs (Article 27):

  • Clear terms of service explaining recommender system parameters
  • Disclosure of criteria for content ranking and recommendation
  • User option to modify or disable personalization
  • At least one option not based on profiling

Risk Concerns:

  • Filter bubbles: Users exposed only to reinforcing viewpoints
  • Amplification: Harmful or divisive content may get prioritized for engagement
  • Manipulation: Potential for promoting commercial or political content
  • Addiction: Designs optimizing for engagement over user wellbeing

Related concepts: Profiling, Automated Profiling, Transparency Obligation, AI-Generated Content

Sources

  • EU AI Act Article 50
  • Digital Services Act