Research map / Large language models / AI and society

Path Dependence under Adaptive AI Delegation

NeurIPS 2026 · Large language models › AI and society · approach: Social-science view of AI evaluation · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a two-state ODE model (latent skill and AI-delegation level) with error-driven learning, decay under delegation, and performance-driven reliance, showing bistability with a saddle separatrix, path dependence, and short-run gains masking long-run skill loss.

Problem

How adaptive reliance on AI assistance co-evolves with human skill over repeated tasks, and when small differences in initial skill or reliance lead to different long-run outcomes.

Approach

Couples a natural-gradient skill-learning/decay flow with a delegation flow driven by the performance gap between independent and AI-assisted work, yielding two attracting regimes (high skill/low delegation and low skill/high delegation) separated by the stable manifold of an interior saddle.

Key results

• With adaptive delegation the system has two attracting equilibria (1,0) and (0,1) separated by the stable manifold of an interior saddle at (theta_a, (1-theta_a)/(1-(1-Delta)theta_a)).
• Increasing AI skill theta_a shifts the basin boundary monotonically downward, enlarging the low-skill basin.
• AI assistance can give lower instantaneous loss initially but higher loss than no-AI after a finite final crossing time t_c, with theta(t_c; no-AI) <= theta_a.

Surprising findings

• Sufficiently heavy reliance can place even high-skill learners in the low-skill basin.
• Learning from AI during delegation lifts the low-skill equilibrium but does not eliminate the low-skill, high-delegation regime.

Limitations

• Model is theoretical and not calibrated to human behavior; functional forms chosen for tractability.
• Robustness variants are supported mainly by simulations and local analysis rather than full proofs.
• Stochastic and multi-skill results are simulation-supported.

Experimental setup

Metrics: basin membership performance gap G_ell(t) crossing time t_c

Baselines: fixed delegation no-AI baseline

Builds on: performative prediction item response theory natural gradient

Closest papers by meaning

• A Mathematical Framework for AI-Human Integration in Work (ICML 2025 · similarity 0.85)
• Human-AI Teaming Through the Lens of Calibration (NeurIPS 2026 · similarity 0.84)
• JobBench: Aligning Agent Work With Human Will (NeurIPS 2026 · similarity 0.84)
• When AI Agents Compete for Jobs: Strategic Capabilities and Economic Dynamics of AI Labour Markets (ICML 2026 · similarity 0.84)
• Partner Modelling Emerges in Recurrent Agents (But Only When It Matters) (NeurIPS 2025 · similarity 0.84)
• The Scaling Laws of Skills in LLM Agent Systems (NeurIPS 2026 · similarity 0.84)