In the agent-based model, task difficulty and AI quality set the baseline level of overreliance, with difficulty the dominant global factor¶

Claim · All claims
Evidence · 1 study (2 entries) · theoretical r3 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Overreliance climbs with task difficulty, and at hard tasks higher AI quality lowers it. → Ahana Biswas 2026 q2 i? Global Morris sensitivity analysis ranks difficulty and AI quality as the leading drivers of behavioral overreliance. → Ahana Biswas 2026 (2)

Evidence¶

Ahana Biswas 2026¶

Ahana Biswas. (2026). Modeling AI Overreliance as a Complex Adaptive System. arXiv. https://arxiv.org/abs/2608.19616 doi:10.48550/arxiv.2608.19616

q2 · i? · theoretical · r3

Simulation sweep over task difficulty crossed with continuous AI quality (Fig. 2), averaged over the final 10 steps across seeds with 95% CIs. Overreliance rises with difficulty from "≈0.02→0.38", and at hard tasks quality moves it "from 0.38 (poor AI) to 0.16 (good AI)".

"The phase diagram (Fig. 2) shows overreliance climbing with difficulty (≈0.02→0.38); at hard tasks AI quality moves it from 0.38 (poor AI) to 0.16 (good AI)."

Ahana Biswas 2026 (2)¶

Ahana Biswas. (2026). Modeling AI Overreliance as a Complex Adaptive System. arXiv. https://arxiv.org/abs/2608.19616

q2 · i? · theoretical · r3

Global Morris elementary-effects screening of behavioral overreliance across the swept parameter ranges (Fig. 6a). The article reports that "difficulty (µ ∗ = 0.29) and AI quality (0.12) lead"; these µ* values are screening measures, not effect sizes.

"A global Morris analysis (Fig. 6a) confirms difficulty (µ ∗ = 0.29) and AI quality (0.12) lead."

Discussion¶

Related Claims¶

• In the model, high-quality AI on hard tasks produces the highest regret despite moderate overreliance, because agents over-defer and rarely self-rely — related
• In the model, interventions that alter the feedback channel (verification visibility, social-proof damping) reduce overreliance and regret, while lowering verification cost alone does not lower regret — related
• In the model, topology shifts aggregate reliance only under opinion dynamics, when influential hubs transmit correlated beliefs that move agents across the verify/use margin — related
• In the model, visible unverified peer use (social proof) suppresses verification and tips the population into collective overreliance, via a smooth crossover with no hysteresis — related
• Theoretical quality of both theory-based and non-theory agents remains stable across Easy, Medium, and Hard scenario difficulty levels — related
• In one quasi-experimental VR shooting study (n = 50), reported in a review, rule-based dynamic difficulty outperformed progressive difficulty — related
• Classroom verification norm is predicted to show threshold behavior and hysteresis rather than a smooth response to teacher-set requirements
• Task category is the dominant factor in agent performance: cross-category spread (26.3 points) far exceeds cross-model spread (8.8 points) — related