这篇文章从古代生存策略追溯到现代竞争优势，指出机器学习已成为一种普遍的预测引擎，嵌入到招聘、商业战略、警务、司法、住房、放贷和媒体内容分发中。尽管文中未提供具体百分比或误差率，但核心趋势是预测系统如今几乎覆盖所有主要社会制度，将预测从孤立的推断扩展到治理和市场决策的常态。

文章的伦理核心在于：算法预测常被当作事实来对待，却本质上是受偏差训练数据和不透明模型影响的有噪声估计；它们可能剥夺个人机会，随后又再生产这些结果，形成自我实现的闭环。作者将此与“卡夫卡式”处境相连：决策因隐藏于模式匹配之下而变得无法挑战，而非基于可审查的透明规则。

更深层的主张是规范性：预测并非仅是描述，而是像指令一样推动人和制度朝其宣称的未来前行，甚至在保险和刑事司法等领域走向社会控制。由于缺乏公共辩论和护栏框架，文本警告说，AI可在未告知对象的情况下对任何人进行预测并据此行动，因此若不加入限制、可质疑性与问责机制，所谓“对多数有效的准确率”可能把个人推向更大的伤害。

This essay traces prediction from an ancient human survival strategy to a modern source of competitive advantage, arguing that machine learning has become a general prediction engine embedded in hiring, business strategy, policing, justice, housing, lending, and media curation. Although no percentages or error-rate figures are provided, the trend it emphasizes is that predictive systems now touch nearly every major social institution, shifting forecasting from isolated use to routine governance and market decision-making.

The core ethical claim is that algorithmic forecasts are often treated as facts even though they are noisy estimates shaped by biased training data and opaque models, and they can deny people opportunities before reproducing those outcomes in a self-reinforcing loop. The author links this to a “Kafkaesque” condition: people cannot contest outcomes because the governing criteria are hidden in pattern matching rather than transparent rules.

At a deeper level, the essay argues that predictions are normative acts that push people and institutions toward the futures they announce, with implications for social control in areas like insurance and criminal justice. Because there is little public debate or guardrail framework, predictive AI can be applied to anyone without notice, so without restraint, contestability, and accountability, accuracy-at-scale can become harmful at the individual level.

Source: AI is the new Oracle of Delphi. That’s bad news

Subtitle: Societies urgently need to confront the ethics of prediction, writes Carissa Véliz

Dateline: 4月 23, 2026 06:20 上午