Dr. AI Will See You Now: How Answer Engines Frame Medical Controversy

• In Event: 20220 - Section on Communication, Information Technology, and Media Sociology Roundtables
  In Refereed Roundtable (60 minute): Section on Communication, Information Technology, and Media Sociology Roundtable I: Artificial Intelligence I

Abstract

Many people turn to search engines as a valuable source of knowledge they can trust. People also rely on the internet as a source for health information. However, the kind of information search engines return is constantly shifting. Since 2012, search engines have modified returns so that people not only provide a list of hyperlinked websites, but also directly answer queries with text automatically extracted from those websites. The rise of large language models (LLM) accelerated this trend towards question answering rather than source provision, with many search engines now providing detailed, free text answers at the top of users’ search results. Recognizing this shift, some companies have even rebranded themselves from search engines to “answer engines,” eschewing ranked lists of sources altogether. While a great deal of research has sought to understand how search engines influence knowledge production, less is known on how AI-enabled answer engines frame scientific medical information and the extent to which these systems rely on accurate sources of information.

Building on previous research that situates search engines within the sociology of knowledge, this paper captures how answer engines frame controversial topics related to health and analyzes the kinds of sources relied on to build these overviews. These include answers on transgender healthcare for children; abortion healthcare; and weight management - particularly focusing on if a body-mass index is considered healthy. By performing a qualitative audit of Google Search’s algorithmic summaries, we identify the processes of selection, omission, emphasis, and phrasework that these summaries deploy, and how these framing techniques serve to transform how knowledge is represented in the source materials. We conclude with a provocation, comparing the contemporary framing work of the algorithm to the historical framing work of the journalists upon which these algorithms’ predictions are based.

Authors

• Anna Beers, University of North Carolina at Chapel Hill

• Francesca Tripodi, University of North Carolina-Chapel Hill