法律的可预测性有多高？

背景与摘要： 本文通过人工智能模型对过去50年间通过的数百部联邦法律进行分析，探讨了法律在实际运行中偏离立法者初衷的程度。研究发现，绝大多数（约89%）的联邦法律基本能够达到预期效果，但仍有约11%的法律在实际实施中出现了严重偏离，其影响要么被极度放大，要么大打折扣，甚至走向了完全相反的方向。文章指出，正如新兴技术一样，法律在现实社会中往往会被以其创造者从未设想过的方式被利用、扩展或无视。

Executive Summary

Laws are often treated as precise instruments, but how often do they actually achieve what their authors intended? To answer this question, an AI-driven analysis was conducted on hundreds of federal laws passed over the last 50 years.

The study found that the vast majority of federal laws (around 89%) perform broadly as expected. However, roughly 11% experience substantial divergences, resulting in impacts that are either vastly larger, drastically smaller, or entirely different than originally anticipated. Much like emerging technologies, laws create new legal and institutional capabilities that often get co-opted, expanded, or ignored in ways their creators never imagined.

立法的意想不到的后果

历史上充满了这样的法律：它们在现实世界中产生的影响让它们的创造者始料未及：

The Unintended Consequences of Legislation

History is full of laws whose real-world impacts completely blindsided their creators:

国家环境政策法 (NEPA): 1969 年构想时，它仅被视为一份旨在保护环境的指导性政策声明，其在最后一刻加入的环境影响声明（EIS）要求几乎没有经过国会辩论。然而，它后来却成为环保组织叫停或拖延重大基础设施项目的主要武器。
401(k) 退休账户: 最初作为 1978 年《税收法》中一个模糊且“微不足道”的条款，由一位 28 岁的初级律师引入，后来却演变成为数千万美国人最主要的退休储蓄工具。
加州的住房法案: 相反，一些旨在“终结单户住宅区划”并通过改造商业街和教堂停车场来强力推动住房建设的州法律，实际上对住房供应的影响极其有限。

The National Environmental Policy Act (NEPA): Conceived in 1969 as an aspirational policy statement to protect the environment, its last-minute environmental impact statement (EIS) requirement received virtually zero congressional debate. Yet, it became the primary weapon for environmental groups to halt or delay major infrastructure projects.

The 401(k) Retirement Account: Originally introduced as an obscure, "negligible" provision within the 1978 Revenue Act by a 28-year-old junior lawyer, it transformed into the primary vehicle for retirement savings for tens of millions of Americans.

California’s Housing Laws: Conversely, several state laws designed to "end single-family zoning" and super supercharge housing construction by converting strip malls and church parking lots have had a severely limited impact on housing supply.

方法论与评分标准

为了衡量立法的可预测性，一个 AI 模型（Claude Opus 4.8 Max Thinking）分析了在 1976 年至 2023 年期间通过的 239 部联邦法律（已过滤掉诸如邮局更名等不足 10 页的法案）。

Methodology & Scoring Rubric

To measure legislative predictability, an AI model (Claude Opus 4.8 Max Thinking) analyzed 239 federal laws passed between 1976 and 2023 (filtered to exclude bills under 10 pages, such as post office renamings).

该 AI 完全基于当时的同期资料（法案文本、国会预算办公室评分、总统签署声明以及媒体报道），对预期效果与实际效果进行估算，并为每部法律分配了一个 介于 -10 到 +10 之间的偏离度得分： * 正数 (+): 影响大于预期，或出现了未预见的积极/扩大应用。 * 负数 (-): 影响小于预期、彻底失败或未得到实施。

Based exclusively on contemporaneous sources (bill texts, CBO scores, presidential signing statements, and media coverage), the AI estimated expected versus actual effects, assigning each law a Divergence Score from -10 to +10: * Positive numbers (+): Larger-than-anticipated impacts or unforeseen positive/expanded applications. * Negative numbers (-): Smaller-than-anticipated impacts, complete failures, or non-implementation.

评分量表

0: 实际影响几乎完全符合预期。
±1–2 (轻微): 总体上符合预期；存在细微的次要偏差。
±3–4 (中等): 存在明显的差距，但核心目的基本实现。
±5–6 (严重): 主要结果实质性超出或未达标，或出现了重大的意外影响。
±7–8 (重大): 核心目标被极大超越，或失败/未实施。
±9–10 (极端): 实际影响与预期毫不相干（近乎完全失败或产生了完全相反的效果）。

The Scoring Scale

0: Actual impact matched expectations almost exactly.

±1–2 (Minor): Broadly as expected; small secondary deviations.

±3–4 (Moderate): Noticeable gaps, but the core purpose was largely realized.

±5–6 (Substantial): Primary outcomes materially exceeded or unmet, or major unanticipated effects emerged.

±7–8 (Major): Central goal greatly overshot or failed/went unimplemented.

±9–10 (Extreme): Actual impact bore little resemblance to expectations (near-total failure or completely opposite effects).

(完整的提示词数据和结果可在 GitHub Repository 中找到。)

(Full prompt data and results can be found in the GitHub Repository.)

结果：立法可预测性的形态

偏离度得分呈现出具有轻微正偏态的钟形曲线： * 40% (95 部法律): 得分在 -1 到 1 之间（作用完全符合预期）。 * 20% (49 部法律): 得分在 -2 到 -4 之间（中度低于预期）。 * 28% (68 部法律): 得分在 2 到 4 之间（中度超出预期/波及范围出人意料）。 * ~11%: 遭遇了“严重偏离”（得分 $\ge \pm5$）。

Results: The Shape of Legislative Predictability

The divergence scores formed a bell curve with a slight positive skew: * 40% (95 laws): Scored between -1 and 1 (acted precisely as expected). * 20% (49 laws): Scored between -2 and -4 (moderate under-delivery). * 28% (68 laws): Scored between 2 and 4 (moderate over-delivery/unanticipated reach). * ~11%: Suffered a "substantial divergence" (scores $\ge \pm5$).

实际影响远大于预期的法律示例 (+ 得分)

航空放松管制 / 航空货运法案 (得分: +5): 原本只打算调整航空保险，但一项货运放松管制的条款允许像 FedEx 这样的公司在任何航线上使用大型喷气式飞机，意外催生了现代“航空快递货运”行业。
2002 年萨班斯-奥克斯利法案 (得分: +5): 旨在安然事件后加强审计监督，其合规成本比预期高出 30-50 倍。此外，一项关于“妨碍官方程序”的冷门条款后来被用于对数百名 1 月 6 日国会山骚乱事件的被告提出刑事指控（在部分被最高法院限制之前）。
2015 年贸易便利化和贸易执法法案 (得分: +7): 虽被宣传为一项常规的海关现代化法案，却悄悄地将“最低免税限额”（免征关税的商品）从 200 美元提高到 800 美元，直接助推了来自 Shein 和 Temu 的价值数十亿美元的直接面向消费者的中国进口商品浪潮。

Examples of Laws with Much Larger Effects than Predicted (+ Scores)

Airline Deregulation / Air Cargo Act (Score: +5): Intended merely to adjust airline insurance, a cargo deregulation provision allowed companies like FedEx to use large jets on any route, accidentally birthing the modern "air express cargo" industry.

Sarbanes-Oxley Act of 2002 (Score: +5): Aimed at increasing auditing oversight after Enron, compliance costs were 30–50 times higher than expected. Furthermore, an obscure clause regarding "obstruction of an official proceeding" was later used to criminally charge hundreds of January 6 defendants (before being partially limited by the Supreme Court).

Trade Facilitation and Trade Enforcement Act of 2015 (Score: +7): Billed as a routine customs modernization bill, it quietly raised the "de minimis exemption" (goods exempt from tariffs) from $200 to $800, directly fueling the multi-billion-dollar influx of direct-to-consumer Chinese imports from Shein and Temu.

实际影响远小于预期的法律示例 (- 得分)

1976 年阿拉斯加天然气运输法案 (得分: -7): 曾被设想为史上最大规模的私人投资能源项目，旨在将天然气输送至美国本土 48 州，但却因成本上升及随之而来的天然气供应过剩而彻底搁浅。
1997 年阿拉巴马-库萨-塔拉普萨河流域公约 (得分: -5): 为促进阿拉巴马州和佐治亚州之间的共享水资源分配计划而设立，但两州未能达成协议，取而代之的是持续数十年的诉讼拉锯战。
2009 年加强与巴基斯坦伙伴关系法案 (得分: -5): 旨在通过投入数十亿美元的基础设施建设资金来换取善意和密切关系，但随着政治紧张局势和军事行动进一步使双边关系恶化，该法案在转变公众舆论方面彻底失败。

Examples of Laws with Much Smaller Effects than Predicted (- Scores)

The Alaska Natural Gas Transportation Act of 1976 (Score: -7): Envisioned as the largest privately financed energy project in history to pipe gas to the lower 48 states, it was completely derailed by rising costs and a subsequent natural gas supply glut.

The Alabama-Coosa-Tallapoosa River Basin Compact of 1997 (Score: -5): Created to foster a shared water-allocation plan between Alabama and Georgia, the states failed to reach an agreement, resulting in decades of continuous litigation instead.

Enhanced Partnership with Pakistan Act of 2009 (Score: -5): Aimed at buying goodwill and close relations through billions in infrastructure funding, it utterly failed to shift public opinion as political tensions and military raids strained relations further.

额外模式

时间不变性: 近期颁布的法律在可预测性上与几十年前通过的法律基本持平。
页数: 篇幅更长、更复杂的法律与高于预期的影响呈现出轻微的正相关性，这可能是由于“捆绑效应”——庞大复杂的综合法案增加了至少有一个条款产生超常影响的几率。
法案类型: 与建立全新结构性项目的法案相比，常规的拨款法案表现出更低的差异度，而前者面临极端偏离的风险则要高得多。

Additional Patterns

Time Invariance: Recent laws are roughly as predictable as laws passed decades ago.

Page Count: Longer, more complex laws show a slight positive correlation with higher-than-expected impacts, likely due to a "bundling effect" where complex omnibus bills increase the odds of at least one component having an outsized effect.

Bill Type: Routine appropriations bills show much lower variance compared to bills creating brand-new structural programs, which carry a much higher risk of extreme divergence.

为什么法律会发生偏离？

立法的错位通常源于两个主要的摩擦点：法律的操作者（法院、机构和未来的立法机构）以及更广泛的外部世界（市场和人类行为）。

Why Do Laws Diverge?

Legislative misalignments generally stem from two primary friction points: legal operators (courts, agencies, and future legislatures) and the broader world (markets and human behavior).

司法解释: 法院常常会以原创作者从未想过的方式来解释法律。例如，1989 年的《道德改革法案》禁止联邦雇员因发表演讲或撰写文章而获得报酬，直到最高法院以违反第一修正案为由否决了其中的部分条款。
机构与检察自由裁量权: 政府机构频繁改变法律的用途。2005 年的《DNA 指纹法案》（包含在一项家庭暴力法案中）后来被美国移民及海关执法局 (ICE) 用来收集移民拘留者的 DNA。相反，2012 年的《STOCK 法案》规定国会议员内幕交易为非法，但至今未产生任何起诉案例。
未来的立法行动: 后来的国会要么会抹杀某项法律的初衷（例如，1984 年的《量刑改革法案》彻底废除了联邦假释，使得 1976 年的《假释重组法案》变成一纸空文），要么会大规模扩展它（例如，2021 年的《REPLANT 法案》将一个仅有 3000 万美元的微小植树信托基金转变为一项种植 12 亿棵树的国家级倡议）。
市场与公众反应: 受监管实体的适应方式往往变幻莫测。2006 年的《信用评级机构改革法案》未能打破标普、穆迪和惠誉的寡头垄断，而贸易法也未能预见全球快时尚供应链的巧妙变通。

Judicial Interpretation: Courts often interpret laws in ways original authors never imagined. For instance, the 1989 Ethics Reform Act banned federal workers from being paid for speeches or articles until the Supreme Court struck down parts of it as a First Amendment violation.

Agency & Prosecutorial Discretion: Government bodies frequently repurpose laws. The DNA Fingerprinting Act of 2005 (housed within a domestic violence bill) was later leveraged by ICE to collect DNA from immigration detainees. Conversely, the 2012 STOCK Act made congressional insider trading illegal, yet has resulted in zero prosecutions.

Future Legislative Actions: Subsequent Congresses can either kill a law's intent (e.g., the 1984 Sentencing Reform Act entirely abolished federal parole, rendering the 1976 Parole Reorganization Act moot) or massively expand it (e.g., the 2021 REPLANT Act transformed a tiny $30M tree-planting trust fund into a 1.2-billion-tree national initiative).

Market and Public Response: Regulated entities adapt in unpredictable ways. The Credit Rating Agency Reform Act of 2006 failed to break up the oligopoly of S&P, Moody’s, and Fitch, while trade laws failed to predict the clever adaptations of fast-fashion global supply chains.

结论：作为技术的法律

将立法比作技术提供了一个有用的框架。当你发明了一项技术——比如马可尼发明了无线电报——你就创造了一种新的能力。然而，一旦这种能力被部署到大千世界中，人类总能找到完全出乎意料的方式来利用它（比如广播电台）。

Conclusion: Laws as Technology

Comparing legislation to technology offers a useful framework. When you invent a technology—like Marconi inventing wireless telegraphy—you create a new capability. However, once that capability is deployed into the wild, humanity finds entirely unanticipated ways to exploit it (such as broadcast radio).

同样，法律引入了组织、制度或监管方面的新能力。有时，它们会引发巨大的、意想不到的文化或经济变革（比如 NEPA 或 401(k)）；而在其他时候，它们则被证明相当于法律界的“气味电影 (Smell-O-Vision)”——完全无用或被市场彻底无视。尽管预测法律的长期影响通常比预测技术更容易，但持续存在的 11% 的偏离率证明，起草立法始终包含着巨大的未知因素。

Similarly, laws introduce organizational, institutional, or regulatory capabilities. Sometimes they spark massive, unintended cultural or economic revolutions (like NEPA or the 401(k)); other times, they turn out to be the legal equivalent of "Smell-O-Vision"—utterly useless or ignored by the market. While predicting the long-term impact of laws is generally easier than predicting technology, the persistent 11% divergence rate proves that writing legislation always carries a heavy element of the unknown.