Artificial intelligence sycophancy, epistemic calibration, and verification intentions in first-year aerospace engineering students’ evaluation of research proposals
Year: 2026
Type: article
Abstract: AI-generated responses may shape students’ judgements through both their informational content and their interpersonal style. This study examined whether conversational praise and epistemic stance were associated with differences in first-year students’ evaluations of research proposals and their self-reported need to verify claims of novelty. Participants were 120 first-year Aerospace Engineering students at a Spanish university, randomly assigned to one of four conditions in a 2 × 2 between-participants design crossing conversational tone (praising vs. neutral) with epistemic stance (unsupported validation vs. evidence-oriented calibrated assessment). Each participant reviewed the same four aerospace topics under one assigned condition and then completed one global evaluation. Factorial analyses showed significant effects of tone and epistemic stance on perceived novelty and perceived need to verify; the largest effect was the stance effect on need to verify, 𝐹 ( 1 1 1 6 ) = 6 7 . 9 0 , 𝑝 < . 0 0 1 , 𝜂 2 𝑝 = . 3 6 9 . Epistemic stance was also associated with differences in choice intention, perceived publication probability, and detection of sycophancy, while tone showed the clearest independent effect on trust in AI. Final decisions differed across conditions: direct acceptance was 63.3% under praising + validation and 10.0% under neutral + calibrated assessment, whereas choosing to search the literature was 10.0% and 46.7%, respectively. A directed content analysis of the open-ended responses was descriptively consistent with this pattern: systematic verification strategies and requests for evidential support were more frequent under calibrated assessment than under validation. Because calibrated responses explicitly introduced prior-work and verification cues, these effects should be interpreted as consequences of the broader contrast between unsupported validation and evidence-oriented calibration, not as a pure effect of disagreement. The study measures verification intention rather than observed verification behaviour or general critical-thinking ability.
Source: Thinking Skills and Creativity
Institution: Universitat Politècnica de València
Language: English
Subfield: Developmental and Educational Psychology
Field: Psychology
Domain: Social Sciences
SDG: Quality education
Open Access status: hybrid
Estimated APC paid: $3,730
OpenAlex ID: W7225255432
