Generics in science communication: Misaligned interpretations across laypeople, scientists, and large language models

Area of Science:

• Science Communication
• Cognitive Science
• Artificial Intelligence

Background:

• Scientists frequently use generic statements (e.g., "statins reduce cardiovascular events") in research communication.
• Large language models (LLMs) often adopt this generic style when summarizing scientific texts.
• Generics can lead to overgeneralization due to differing interpretations across audiences.

Purpose of the Study:

• To investigate how laypeople, scientists, and LLMs interpret scientific generics.
• To identify potential mismatches in the generalization and credibility of scientific statements.

Main Methods:

• A comparative study involving laypeople, scientists, and two advanced LLMs (ChatGPT-5 and DeepSeek).
• Analysis of systematic differences in the interpretation of generic statements used in scientific contexts.

Main Results:

• Laypeople and LLMs judged scientific generics as more generalizable and credible than most scientists did.
• LLMs rated the generalizability and credibility of scientific generics even higher than laypeople.

Conclusions:

• Mismatches in interpreting generics pose significant risks for science communication.
• Scientists may underestimate the extent to which laypeople overgeneralize their findings.
• LLMs may systematically overgeneralize scientific findings, highlighting the need for careful language choices in AI-mediated communication.