Philosophy has long suffered an unfortunate reputation as pedantic and abstruse. In one of the most prominent debates of the 20th century, philosophers spent a great deal of energy arguing over what the means. Paul Graham, the legendary tech investor, studied philosophy as a college student, which seemed “an impressively impractical thing to do,” as he later wrote. “Sort of like slashing holes in your clothes or putting a safety pin through your ear.” But over time, Graham became disillusioned: “I kept taking philosophy courses and they kept being boring,” he explained. And so, eventually, he switched to studying artificial intelligence.
哲学长期以来背负着一种不幸的声誉，被视为迂腐且晦涩难懂。在20世纪最著名的辩论之一中，哲学家们耗费了大量精力去争论“那意味着什么”（指代无意义的语义争论）。传奇科技投资人保罗·格雷厄姆在大学时曾修读哲学，他后来写道，这看起来“是一件令人印象深刻的不切实际的事”，“简直就像是在衣服上剪洞，或者把安全别针穿过耳朵一样”。但随着时间的推移，格雷厄姆感到幻灭：“我不断修读哲学课程，它们却总是枯燥乏味，”他解释道。于是，他最终转而去研究人工智能。
Like Graham, the field of philosophy has lately turned its attention to AI. At major tech companies, a growing rank of philosophers with Ph.D.s and flush compensation packages are helping shape the technology’s future. Meanwhile, universities are pouring resources into hiring philosophers who study AI. In 2013, 1 percent of roles on PhilJobs, the field’s primary job board, were related to the technology. Last year, that figure hit 16 percent.
与格雷厄姆一样，哲学界最近也将目光转向了人工智能。在大型科技公司中，越来越多的拥有博士学位且薪酬丰厚的哲学家正协助塑造这项技术的未来。与此同时，大学也在投入大量资源招聘研究人工智能的哲学家。2013年，该领域主要招聘网站PhilJobs上仅有1%的职位与该技术相关。去年，这一比例已飙升至16%。
In some ways, it is philosophers who got us into this AI mess in the first place. For centuries, they have contemplated the creation of artificial minds. And the Swedish philosopher Nick Bostrom’s 2014 book, Superintelligence, helped bring attention to the potential dangers of all-powerful AI. Bostrom’s work has influenced research agendas across all of the major labs. Sam Altman once described the book as “the best thing” he had read on the risks of AI.
在某种程度上，正是哲学家们让我们陷入了这场人工智能的混乱局面。几个世纪以来，他们一直在沉思创造人造心智的可能性。瑞典哲学家尼克·波斯特洛姆2014年的著作《超级智能》有助于人们关注全能人工智能的潜在危险。波斯特洛姆的作品影响了所有主要实验室的研究议程。山姆·奥特曼曾形容这本书是他读过的关于人工智能风险“最好的东西”。
But the two disciplines have never been quite as entangled as they are now. As the AI boom has exploded, Silicon Valley has looked to philosophers to help the industry build what are, at least in theory, more virtuous machines. AI companies have to make all kinds of difficult decisions about how their bots should interact with humans—decisions that philosophers, experts in parsing such dilemmas, are uniquely well equipped to inform. Last fall, in an interview with Tucker Carlson, Altman said that OpenAI consulted “hundreds of moral philosophers” and tech-ethics experts when designing rules for ChatGPT’s behavior. (An OpenAI spokesperson was unable to provide additional information about what this consulting involved.)
但这两个学科从未像现在这样纠缠不清。随着人工智能热潮的爆发，硅谷转向哲学家，希望他们能帮助行业构建至少在理论上更具美德的机器。人工智能公司必须就其机器人应如何与人类互动做出各种艰难的决定——对于这些困境的解析专家，哲学家们拥有独特的优势来提供指导。去年秋天，在接受塔克·卡尔森的采访时，奥特曼表示，OpenAI在设计ChatGPT的行为规则时咨询了“数百名道德哲学家”和科技伦理专家。（OpenAI发言人无法提供有关这些咨询具体内容的更多信息。）
Perhaps the most philosophy-drunk of the major AI firms is Anthropic. It wants Claude, in addition to being a helpful assistant, to have “good character,” Amanda Askell, a philosopher at the company, told me last year. In January, under Askell’s leadership, the company published Claude’s constitution, an 84-page philosophical treatise that outlines Anthropic’s intentions for the bot’s personality and behavior. That document, which includes dense philosophical sections on meta-ethics and epistemology, is then used to train Claude.
在主要的人工智能公司中，最“醉心于哲学”的或许是Anthropic。该公司哲学家阿曼达·阿斯克勒去年告诉我，他们希望Claude除了是一个有用的助手外，还具备“良好的品格”。今年一月，在阿斯克勒的领导下，公司发布了《Claude宪法》，这是一份84页的哲学论文，概述了Anthropic对机器人性格和行为的意图。这份文件包含了关于元伦理学和认识论的密集哲学章节，随后被用于训练Claude。
Not all philosophers working with AI companies do so full-time. Sam Elgin, who studies logic and metaphysics at the University of Pennsylvania, told me that a firm, which he declined to name, recently asked him to be a consultant. “The general goal was to train large language models to reason more rigorously about ethics,” he said. He fed ethical dilemmas into the AI system, and then evaluated the logic the model used to produce its response, searching for unstated assumptions and gaps in its reasoning.
并非所有与人工智能公司合作的哲学家都是全职工作。在宾夕法尼亚大学研究逻辑学和形而上学的萨姆·埃尔金告诉我，一家他拒绝透露名称的公司最近聘请他担任顾问。“总体目标是训练大型语言模型更严谨地推理伦理问题，”他说。他将伦理困境输入AI系统，然后评估模型生成回答所使用的逻辑，寻找其推理中未言明的假设和漏洞。