Autorite Preview in Academic Search Engines: LLMs Really Evaluate Content?

Artificial intelligence research is increasingly using language models to look for more academic literature, but it has started to question how objective these systems are really. A new study investigated refers to a serious problem in conversational search engines: otorite bias. This prejudice shows that science people make decisions based on factors such as writer reputation and publishing place instead of selecting articles.

The basic findings of the study can be summarized as follows:

• When researchers keep the title and summary the same, they created different scenarios by changing the author prestige, journal quality and number of quotes
• LLM-based search engines showed more prestigious writers and posts tend to prefer articles
• These preferences were often associated with the actual content of the article
• Otorite signals significantly affect models’ recommendations

These findings give significant results. An AI system, which does not call objective information when researching, instead, prefers based on social factors such as prestige and reputation, may miss out on real innovative studies. High quality articles from less recognized researchers may not be preferred to medium-level work from famous names.

This research reveals important questions about how conversational AI systems are used in the academic field and reveals the areas that need to be developed. The researchers emphasize that developing new methods to reduce such bias while training language models are needed.