Submission 115
Source Memory for AI- vs. Human-Generated Online Content
SymposiumTalk-01
Presented by: Luise Metzger
The increased capabilities and availability of large language models have changed the online information landscape: Apart from traditional, human-curated sources—such as forums, news websites, or encyclopedias—, AI-generated information—such as chatbot responses or automated search summaries—is now also readily available. This research project examines whether people spontaneously categorize and recall web content as human- vs. AI-generated.
In two online studies, we adapted the “Who said what?” paradigm: Participants were shown trivia statements randomly allocated to informational websites, half of which were human- and half AI-generated. After a filler task, they then completed a source monitoring test where they were presented with both the original statements along with new distractor statements and were asked for each if it was new, or, if they assumed they had seen it previously, which website it had appeared on. We further assessed individual attitude measures and expected negative attitudes towards AI to enhance participants’ monitoring of AI-generated sources.
We analyzed the data using hierarchical multinomial processing trees with hierarchical ML estimation (Nestler & Erdfelder, 2023) based on Klauer and Wegener’s (1998) two-high-threshold model of social categorization. Initial results indicate that categorization occurs only partially.