预测市场近期的争议显示，关键问题往往不在事件本身，而在于指标定义的边界。美国对委内瑞拉的军事行动未被认定为“入侵”，相关赌注未结算；关于美国是否会介入格陵兰的押注亦受制于条款细节；2024年12月《时代》年度人物授予“AI架构师”，而非“AI”本身，导致相关下注落空。这些案例并非孤立，而是预测市场在规模化过程中反复遭遇的结构性问题：当结果依赖语言定义而非可量化事实时，争议概率显著上升。
预测市场依赖清晰、可裁决的标准。菲利普·泰特洛克提出的“先知测试”要求问题在无需澄清的情况下即可被全知者回答，但现实世界事件往往难以满足。即便将验证权威外包给官方机构，如诺贝尔委员会，也无法覆盖所有情形。当前围绕委内瑞拉的争论，核心在于行动是否“意图建立对任何领土的控制”，显示出标准在意图、程度与结果之间的灰色地带。
这种指标困境在AI时代被放大。安德烈·卡帕斯指出，任务的“可验证性”越高，越适合自动化，因为AI可以围绕明确目标反复优化。但组织必须先设计可度量的测试与评分标准，这在实践中并非二元选择。随着AI执行、人类设定指标，人类的角色将转向解释偏差、修订标准，并在指标优化与真实价值之间做出判断。指标与价值之间的缺口不会消失，只会变得更重要。
Recent disputes in prediction markets show that the core problem is often not events themselves but the boundaries of metric definitions. A US military action against Venezuela was not deemed an “invasion,” so related bets did not pay out; wagers on potential US involvement in Greenland hinge on fine print; and in December 2024 TIME’s Person of the Year went to the “architects of AI,” not “AI,” nullifying those bets. These cases are not isolated but reflect a structural issue that intensifies as markets scale: when outcomes depend on language rather than quantifiable facts, disputes become more likely.
Prediction markets require clear, resolvable criteria. Philip Tetlock’s “clairvoyance test” holds that a question should be answerable by an omniscient seer without clarification, a standard real-world events rarely meet. Even delegating verification to official bodies, such as the Nobel committee, cannot eliminate ambiguity. The current Venezuela debate turns on whether the operation was “intended to establish control over any portion of territory,” revealing gray zones between intent, magnitude, and outcome.
This metric problem is amplified in the AI era. Andrej Karpathy argues that higher “verifiability” makes tasks more amenable to automation because AI can optimize against explicit objectives. In practice, organizations must first design measurable tests and rubrics, which are rarely binary. As AI executes and humans define metrics, human work shifts to interpreting mismatches, revising standards, and judging when optimization diverges from real value. The gap between what is measured and what matters will persist, and its importance will only grow.