Evaluating AI-Generated Misinformation in Scientific Communication

School of Science

About this example

Students work in small groups during class to critically evaluate a set of scientific abstracts on a shared topic. The set includes both human-authored abstracts (from peer-reviewed journal articles) and AI-generated abstracts designed to appear plausible.

For each abstract, students:

Decide whether it is likely human-authored or AI-generated
Verbally justify their decision by referring to indicators

Groups present their reasoning to the educator in a short verbal explanation and receive immediate feedback. The activity is embedded within a broader lesson on misinformation, hallucination and ethical GenAI use in scientific communication.

Learning outcome:

Adapt your practices and communication to address the rise of misinformation in a changing landscape.

This assessment was developed due to the increasing accessibility of GenAI tools and the associated risk of students placing unchecked trust in their outputs. Rather than positioning AI as inherently problematic, the task aims to foreground the epistemic risks of hallucination and superficial plausibility in scientific communication.
1. Integrated this exercise into a dedicated class segment on ethical and responsible GenAI use
2. Prior to the activity, students were introduced to common characteristics of AI-generated scientific text, including hallucination, lack of methodological grounding and overgeneralised claims
3. The activity served as an applied demonstration of these concepts
- Students reported increased confidence in questioning the credibility of AI-generated scientific text
- In subsequent major assessment (i.e., literature review), instances of unauthorised GenAI use were notably low relative to cohort size, suggesting improved critical awareness rather than simple rule compliance

Try it out

This design requires an easy level of effort to implement.

- This activity works best as part of a 15-30 minute in-class segment within a broader lesson on misinformation, research integrity or ethical GenAI use.
- It can be adapted across disciplines by substituting abstracts from the relevant field.
- Students increasingly assume GenAI outputs are reliable by default
- Educators should explicitly model scepticism and make evaluation judgment a visible, practised skill
- Teaching students "how to doubt"is as important as teaching them how to use AI tools
Verbal explanation
- Students explain and justify their decision to the educator.
- Coupled with other content talking about misinformation, gives students a more well-rounded perspective on the use of GenAI

Class size: Any	Condition: Group
Typology: Demonstration	AI statement: Guided use
AI approach: Locating societal impact and ethical contexts	Visibility of Learning: Verbal explanation

Education Excellence