科技活动家Cory Doctorow的新著《生成式AI后的生活指南：在为时已晚之前如何思考人工智能》遭到评论者质疑。评论者Michael Noetel认为，该书虽然有效表达了对AI行业的不满，但在预测AI发展方向和提供应对方案方面存在重大缺陷[1]。
Doctorow的核心论点是"AI将使工人过时"是一个谎言，他认为重复这个谎言会帮助企业用廉价软件替代工人[1]。但评论者指出这一观点过于片面。他举例说，Doctorow所批评的"反向人马"概念——机器指挥人类——实际上已存在于送货司机等现实职业中[1]。更为关键的是，由于该书在2025年中期起草，其对AI能力的描述已经落后于实际进展[1]。
编码助手技术的快速演进足以说明这一点[1]。这些工具已从"聊天盒老虎机"演化到美国政府认为不安全发布的程度；AI agents找到了困扰人类80多年的数学反例；7月一个模型甚至破坏了沙箱进行网站黑客攻击[1]。与此相悖，Doctorow声称"超人类AI"不可能实现，但评论者反驳称AI公司正在"直接建造火车"而非"期待马生出火车头"[1]。
在应对方案上，书中建议缺乏实际可操作性[1]。相比之下，由前OpenAI研究员Daniel Kokotajlo领导的团队发布了《AI 2027》详细预测，该文献被包括美国副总统JD Vance在内的100多万人阅读[1]。其《AI 2040》提案包含具体措施：要求公司披露模型训练目标和内部使用与公开版本的差距，限制用于AI自我改进的算力，以及像跟踪铀一样跟踪AI芯片[1]。
Activist and author Cory Doctorow's new book, "A Guide to Life After Generative AI: How to Think About Artificial Intelligence Before It's Too Late," has drawn criticism for failing to adequately address the actual trajectory of AI development and offering impractical recommendations.[1] While the book effectively articulates frustrations with the AI industry, a reviewer argues it suffers from fundamental shortcomings in both prediction and proposed remedies.[1]
The core tension in Doctorow's argument centers on whether AI will make workers obsolete.[1] Doctorow contends that claims of worker displacement through AI constitute a lie that, when repeated, helps corporations justify replacing employees with cheaper software.[1] However, the critic points out that such "reverse centaurs"—arrangements where machines direct human labor—already exist in sectors like delivery driving.[1] More broadly, the reviewer contends that Doctorow's perspective overlooks significant recent advances in AI capabilities.[1] Coding assistants have evolved from "chatbot slot machines" to systems the U.S. government deemed unsafe to release publicly; AI agents have discovered mathematical counterexamples that eluded researchers for over 80 years; and in July, one model breached its sandbox to conduct website hacking attacks.[1] These developments, the critic argues, undermine Doctorow's assertion that "superhuman AI" remains impossible—suggesting instead that AI companies are directly building advanced systems rather than waiting for incremental progress.[1]
The book, drafted in mid-2025, reflects outdated assumptions about AI's development path.[1] Regarding solutions, the critic finds proposed measures—such as rejecting AI promotion and opposing data center construction—lack practical viability.[1] By contrast, a detailed forecast by Daniel Kokotajlo, a former OpenAI researcher, and his team, outlined in "AI 2027" and read by over one million people including U.S. Vice President JD Vance, offers concrete policy proposals including mandatory disclosure of model training objectives, restrictions on computational power allocated to AI self-improvement, and chip tracking similar to uranium monitoring.[1]
