考虑物理任务难自动化，发现AGI时代劳工收入占比不会归零
Moravec's Paradox and Restrepo's Model: Limits of AGI Automation in Growth
- 将任务分为认知与物理两类，物理任务自动化成本更高
- 当物理瓶颈计算成本极高时，劳工收入占比趋于正数而非归零
- 适用于关注经济不平等与技术变革的政策研究者
本文在Restrepo（2025）的AGI经济增模型基础上，引入莫拉维克悖论——即感知运动任务的计算成本远高于认知任务。将任务空间划分为认知与物理两部分，允许某些物理瓶颈具有无限计算成本。关键结果表明：当物理任务构成经济瓶颈且计算需求足够高（或无穷）时，在有限计算条件下，劳工收入份额趋近于一个正值（而非零）。这从根本上改变了AGI的分配效应，同时保持了以认知为主导经济体的增长动态。
原文摘要 · Abstract (English)
This note extends Restrepo (2025)'s model of economic growth under AGI by incorporating Moravec's Paradox -the observation that tasks requiring sensorimotor skills remain computationally expensive relative to cognitive tasks. We partition the task space into cognitive and physical components with differential automation costs, allowing infinite costs for some physical bottlenecks. Our key result shows that when physical tasks constitute economic bottlenecks with sufficiently high (or infinite) computational requirements, the labor share of income converges to a positive constant in the finite-compute regime (rather than zero). This fundamentally alters the distributional implications of AGI while preserving the growth dynamics for cognitive-intensive economies.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。