Frontiers in Psychology Research Model

AI Cognitive Offloading Simulator: Dependent vs. Autonomous LLM Use

Operationalizing Zhu Qiuhan et al. & Jordan Gibbs: Trace synaptic agency, counter structural skill decay, and convert passive answer machines into cognitive scaffolding.

Interactive Cognitive Task Pipeline

Route each reasoning phase between human mental effort and LLM coprocessor
PRESETS:
● Blue Synapse: Active Human Cortex (Autonomous) ● Purple Vector: Delegated LLM Coprocessor (Dependent)
COGNITIVE AGENCY SCORE 64% Autonomous Balance
CRITICAL BOTTLENECK Drafting Structural handoff detected
INTELLECTUAL FRICTION 0.54 Socratic resistance factor
VERIFICATION VIGILANCE 80% Hallucination catch buffer
10-Sprint Retention Trajectory Structural: -32% | Conceptual: +12%
💡
Prescribed Scaffolding Intervention Shift Drafting to Socratic outline co-drafting: write topic sentences manually, prompt LLM for counter-theses only.

Prompt Scaffolding Transformation Engine

Decompose passive answer-machine prompts into active cognitive friction prompts
⚠️ Handing over 100% of reasoning yields zero mental friction and rapid skill atrophy.

Grounded Methodology: Zhu Qiuhan (2024) & Jordan Gibbs Framework

Dependent Offloading (Answer Machine) Treating the model as an end-to-end synthesizer removes the neural load of organizing hierarchy, evaluating premise soundness, and grappling with syntax. Over repeated cycles, confidence remains high ("fluency illusion") while real transfer ability collapses.
Autonomous Offloading (Co-Reasoning) Using the model strictly as an adversary, source stress-tester, or divergent explorer preserves the human working memory's role as executive editor. You write the claims; the machine tests the counter-claims.
The 10-Sprint Decay Horizon Our longitudinal equation tracks working memory plasticity $S_t = S_0 \cdot e^{-\lambda t}$ adjusted for task friction $F$. Without minimum critical resistance ($F \ge 0.45$), synthesis capacity decays by ~30% in under six working iterations.