Stories about AlgorithmWatch
2 related stories
Google's election AI Overviews are opaque, rely on few sources, and sometimes take sides
AI InsightUsing EU DSA access, AlgorithmWatch analyzed 4,480 election queries. Google's AI Overviews proved opaque, relying heavily on YouTube. This shows LLM info-retrieval risks source monoculture and bias, contrasting prior election safeguards.Key TakeawayContrasting prior safeguards, Google AI Overviews were found to rely on single sources and show bias.Why It MattersReveals RAG source concentration and bias risks in sensitive political topics, affecting information neutrality.Who's Affected- RegulatorsProvides a DSA review case for AI overviews, may trigger targeted inquiries and rectifications.
- General UsersSubject to single-source bias and algorithmic bias when retrieving election information.
What's NextWatch if Google responds and adjusts AI Overview's source allocation and transparency for sensitive queries.Importance 78/100AI chatbots regularly link pregnant users to anti-abortion websites without disclosure
AI InsightAI chatbots are linking pregnant users to anti-abortion websites without disclosing their stance, affecting users' access to information; key change: compared to the past, AI chatbots are not explicitly disclosing their stance; significance: it has an important impact on the accuracy of users' information access and the ethical responsibility of AI applications.Key TakeawayAI chatbots are not disclosing their stance.Why It MattersThis event reveals the potential risks of AI chatbots in handling sensitive topics and the importance of user privacy and data security.What's NextIn the future, it is necessary to pay attention to the transparency and ethical standards of AI chatbots in handling sensitive topics.Importance 80/100