AI编程的多重困境:认知不可达、博弈错位与制度失效的分类分析(The Multiple Dilemmas of AI Programming: A Typological Analysis of Cognitive Unreachability, Game Misalignment, and Institutional Failure)

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

摘要 AI编程工具在提升短期交付速度的同时,引发了代码质量退化、安全漏洞累积、开发者技能萎缩与责任真空等系统性问题。既有研究常将这些现象混为一谈,导致对策错位。本文首先对AI编程中的核心现象进行性质分类,识别出六类不同性质的机制:认知类(心智模拟失效)、博弈类(囚徒困境)、信息类(质量信号失真)、合同类(显式/隐式指标分裂)、组织类(审查失效)与系统类(同质化与能力再生产中断)。在此基础上,本文论证:囚徒困境是独立存在的博弈结构,但其“合作不可行”的原因在不同场景下性质不同——组织内部是认知不可达,外包场景是合同无义务,行业层面是协调失败。三者不可混同,需分别分析。本文最后按机制类别提出分层对策,并讨论其理论贡献与实践启示。 关键词 AI编程;心智模拟失效;囚徒困境;显式/隐式指标分裂;认知不可达;制度失效 Abstract While accelerating short-term delivery, AI programming tools have given rise to systemic problems: degradation of code quality, accumulation of security vulnerabilities, atrophy of developer skills, and a vacuum of accountability. Existing studies often conflate these phenomena, which produces misdirected remedies. This paper first classifies the core phenomena of AI programming by their nature and identifies six mechanisms of distinct kinds: cognitive (failure of mind simulation), game-theoretic (the prisoner's dilemma), informational (distorted quality signals), contractual (the split between explicit and implicit metrics), organizational (review failure), and systemic (homogenization and interrupted capability reproduction). On this basis, the paper argues that the prisoner's dilemma is an independently existing game structure, yet the reasons why "cooperation is infeasible" differ in nature across scenarios: within organizations it is cognitive unreachability, in outsourcing it is the contractual absence of obligation, and at the industry level it is coordination failure. The three must not be conflated and require separate analysis. The paper closes by proposing layered remedies organized by mechanism category and by discussing their theoretical contributions and practical implications. Keywords AI programming; failure of mind simulation; prisoner's dilemma; explicit/implicit metric split; cognitive unreachability; institutional failure