Abstract
Large language models (LLMs) like ChatGPT are increasingly treated by the public as if theypossess subjective experience, with multiple studies probing the popular opinion on the matter. Thepresent study provides the first large-scale measurement of this phenomenon in Poland. Anationally distributed online survey panel (N = 1,736) rated ChatGPT’s capacity for consciousexperience on a 1–100 scale. "Most respondents (74%) attributed ChatGPT at least minimalconsciousness on the continuous rating scale, closely mirroring prior U.S. findings; because thisfigure is sensitive to question format, however, we conservatively estimate that more than 10% ofrespondents would affirm ChatGPT's status as an experiencer under a stricter, categorical framing.Small exploratory associations suggested that older and more highly educated participants wereslightly more skeptical, whereas right-leaning respondents attributed slightly more consciousness.A multivariate regression confirmed that these associations were not simply artifacts ofconfounding among demographic variables, though together they explained less than 3% of thevariance in consciousness attributions. These effects, though modest, indicate theoretically relevantvariation. Interpreting these results through Epley’s three-factor anthropomorphism model, thework argues that limited mechanistic understanding of LLMs, highly human-like behavior, andunmet social needs likely contribute to widespread phenomenal consciousness attribution. Givenethical concerns about over-trust, parasocial attachment, and responsibility misperception, thesebeliefs matter for AI deployment and governance. Our findings highlight the need for evidence-based educational strategies that improve public understanding of how LLMs work while allowingbeneficial use of the technology