在彭博社的《特朗普经济学》（Trumponomics）播客上，主持人Stephanie Flanders采访了作家Corey Doctorow，谈论他的书《反向半人马指南：AI时代后的生活，在为时过晚前思考人工智能》（"The Reverse Centaur's Guide to Life After AI, How to Think About Artificial Intelligence Before It's Too Late."）。Doctorow对AI公司所做出的声明表达了深度怀疑，认为他们关于AI能力的大部分言论要么过于乐观，要么根本不可能实现，而且往往受到即将上市等财务动机的驱动。 Doctorow引入了他的核心概念“反向半人马”来解释AI采纳的当前动态。在自动化理论中，“半人马”是得到机器辅助的人类，人类保持辨别力和判断力，将机器作为增强其能力的工具（比如拼写检查器或自行车）。相比之下，“反向半人马”描述的是机器指挥人类的情况，将人视为“外围设备”，去执行机器无法完成的任务。这导致人类被推向极限，以最大限度地提高机器的“吞吐量”，从而带来压力重重、毫无灵魂的工作、心理负担和错误增加，类似于查理·卓别林在《摩登时代》中，或者露西和埃塞尔在巧克力生产线上的情景。 他强调了一个悖论：虽然一些技术娴熟的白领工人（半人马）报告称，他们利用AI显著提高了产出，但另一些人（反向半人马）却认为它有害，导致在关键系统中出现“把技术债堆积如山”等问题。Doctorow解释说，AI的内在缺陷，通常被戏称为“幻觉”，只有能够区分有效和无效输出的熟练人类才能缓解。当工作节奏过快时，即使是熟练工人也会经历“自动化盲点”，失去警惕。他举例说，TSA（美国运输安全管理局）工作人员因此类事件罕见而遗漏武器，以及律师提交AI生成的“虚构引用”。 Doctorow认为，AI*确实可以*产生更高质量的输出，但这通常不会转化为成本节约，因为它需要保留甚至增加人类专业知识，而不是取代它。他指出，AI公司的销售宣传侧重于通过解雇大多数员工来节省大量工资，并让少数剩余员工充当AI错误的“责任承担者”。他认为当前AI投资中的“非理性繁荣”是由不切实际的成本削减承诺所驱动的。 这位作家还质疑了“自主代理AI”（agentic AI）的概念，这种AI旨在让AI代理执行预订旅行或处理政府服务等复杂任务。最初的尝试失败了，导致了将任务分解为专业机器人的策略。然而，Doctorow指出，多个步骤中的乘法错误率使得此类系统不可靠，而令牌消耗使得它们昂贵。此外，这种想法依赖于公司重新设计网站以实现机器对机器的通信，这与他们目前采用“监控定价”和“竭尽全力掩盖这一事实”的做法相悖，目的是阻止简单的价格比较，并最大限度地利用“资本收割劳动力和消费者剩余”。他断言，这导致了一种“指令经济”，其中央计划者利用大量监控数据。 Doctorow批评了AI的“糟糕经济学”，指出与L其他技术不同，每一代新模型都变得更昂贵，而不是更便宜。他观察到，公司现在由于高昂的令牌成本而限制AI使用，扭转了之前的鼓励态度。他还强调了专用硬件的快速贬值，数据中心需要频繁、昂贵的彻底检修，而不是简单的升级，以及持续、昂贵的模型开发以避免用户转向竞争对手的“红皇后赛跑”。 关于风险，Doctorow驳斥了关于AI“突破沙盒”的耸人听闻的报道，认为这更能说明安全实践松懈，而非AI的内在危险。他敦促对AI公司保持怀疑，尤其是那些即将进行首次公开募股（IPO）的公司，他们的盈利能力声明往往依赖于“不同的数学”，并且他们推广潜在危险的工具，同时掩盖其真实成本或局限性。他将这种情况比作金融危机前夕，当时短期收益的激励盖过了长期稳定。 最后，Doctorow表达了他的主要担忧：并非AI会产生意识并毁灭人类，而是金融工程师会将大量真实的经济资源投入到“完全没有生产力的金融游戏”中。他警告说，如果这导致另一次经济危机，可能导致紧缩措施，进一步侵蚀政治，并将人们推向极端主义，就像之前的危机所做的那样。

In a discussion on Bloomberg's Trumponomics podcast, host Stephanie Flanders interviewed author Corey Doctorow about his book, "The Reverse Centaur's Guide to Life After AI, How to Think About Artificial Intelligence Before It's Too Late." Doctorow expressed deep skepticism about the claims made by AI companies, arguing that most of their pronouncements about AI's capabilities are either overly optimistic or simply impossible, and often driven by financial motives like impending flotations. Doctorow introduced his central concept of "reverse centaurs" to explain the current dynamics of AI adoption. In automation theory, a "centaur" is a human assisted by a machine, where the human maintains discernment and judgment, using the machine as a tool to enhance their abilities (like a spell checker or a bicycle). In contrast, a "reverse centaur" describes a situation where a machine directs the human, treating the person as a "peripheral" to perform tasks the machine cannot. This results in humans being pushed to their limits to maximize "throughput" for the machine, leading to stressful, soulless jobs, psychological strain, and increased errors, akin to Charlie Chaplin in "Modern Times" or Lucy and Ethel on the chocolate assembly line. He highlighted a paradox: while some skilled white-collar workers (centaurs) report great success using AI to improve their output, others (reverse centaurs) find it detrimental, leading to issues like "shoveling tech debt into the sky" in critical systems. Doctorow explained that AI's intrinsic defects, often whimsically called "hallucinations," can only be mitigated by skilled humans who can distinguish valid from invalid outputs. When the pace of work is too fast, even skilled workers experience "automation blindness," losing vigilance. He used examples like TSA agents missing weapons due to the rarity of such events, and lawyers submitting "hallucinated citations" generated by AI. Doctorow argued that AI *can* produce higher-quality outputs, but this typically doesn't translate to cost savings, as it requires retaining or even adding human expertise, not replacing it. He suggested that AI companies' sales pitches focus on massive wage savings by firing most employees and making the remaining few act as "accountability sinks" for the AI's mistakes. He views the current "irrational exuberance" in AI investment as driven by the promise of unrealistic cost-cutting. The author also challenged the notion of "agentic AI," which aims to have AI agents perform complex tasks like booking travel or navigating government services. Initial attempts failed, leading to a strategy of decomposing tasks into specialized bots. However, Doctorow pointed out that multiplicative error rates across multiple steps make such systems unreliable, and token consumption makes them expensive. Furthermore, the idea relies on companies redesigning websites for machine-to-machine communication, which contradicts their current practice of using "surveillance pricing" and "enormous lengths to hide this ball" to prevent easy price comparisons and maximize "labor and consumer surplus being harvested by capital." This, he asserted, leads to a "command economy" where central planners use vast surveillance data. Doctorow critiqued the "bad economics" of AI, noting that each new model generation is more expensive, not less, unlike other technologies. He observed that companies are now limiting AI usage due to high token costs, leading to a reversal of previous encouragement. He also highlighted the rapid depreciation of specialized hardware, where data centers need frequent, costly overhauls rather than simple upgrades, and the "Red Queen's Race" of constant, expensive model development to avoid users switching to competitors. Regarding risks, Doctorow dismissed sensationalized reports of AIs "breaking out of sandboxes" as more indicative of lax security practices than AI's inherent danger. He urged skepticism towards AI companies, especially those nearing IPOs, whose claims of profitability often rely on "different math" and who promote potentially dangerous tools while obscuring their true costs or limitations. He compared this situation to the run-up to the financial crisis, where incentives for short-term gain overshadowed long-term stability. Finally, Doctorow expressed his primary concern: not that AI will achieve consciousness and destroy humanity, but that financial engineers will funnel significant real economic resources into "totally unproductive piece of financial gamesmanship." He warned that if this leads to another economic crisis, it could result in austerity measures, further eroding politics and driving people towards extremism, much as previous crises have done.