LLM digital shadows reproduce the human pattern of positive science perceptions anchored in negative mathematics associations¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Across 14 LLMs, behavioural forma mentis networks show positive associations around science alongside negative associations around mathematics, mirroring the human bias reported by Stella and colleagues. → Esposito 2026

Evidence¶

Esposito 2026¶

Esposito, N., Tricarico, A., Porzio, L., Aghazadeh Ardebili, A., & Stella, M. (2026). Math Education Digital Shadows for Investigating Learning with GenAI: Mathematics Performance, Anxiety, and Confidence in LLMs. arXiv preprint. https://arxiv.org/abs/2604.27618 doi:10.48550/arxiv.2604.27618

q2 · i? · associational · r2

Task 3 behavioural forma mentis network analysis of semantic associations and valence ratings for 50 cue words. The article reports negative associations anchoring mathematics in methodological terms such as "algorithm", "computation", and "equation", and states the divergence is a persistent pattern across 14 different LLMs.

"The LLM-generated network in Fig. 4a reproduces the human bias/cognitive dissonance observed by Stella and colleagues38: like human students, also digital shadows cognitively associate positive perceptions of science with negative perceptions of its building blocks, such as mathematics."

Discussion¶

Related Claims¶

• LLMs simulating survey respondents under AI-assistant instructions report higher mathematics self-efficacy and lower anxiety than under human-shadow conditioning — related
• Censoring appears to affect negative-valence ratings of mathematics-related cues between model variants — related
• Priming human shadows as math lovers or haters shifts LLM emotional tone about mathematics — related
• Human-shadow prompting generates broad psychometric variance while AI-assistant responses are uniform — related