The Lede

Anthropic and the Monetization of Paradise

Anthropic与天堂的货币化
What does it mean to sell shares in the idea of a better world? Ahead of its I.P.O., the company is making the pitch.
出售一个更美好世界理念的股份,究竟意味着什么?在首次公开募股前夕,这家公司正在向世人兜售这一愿景。
a globe that looks like a vase with a flower or a bomb with a lit fuse
Illustration by Dan Zhou

The Earth is still twirling on its axis as it glides across the heavens in its elegant orbit of the sun, but Anthropic’s stock goes up for sale this week, so brace yourself for a certain planetary chaos: money falling from the skies, the sense of time stopped—at least, if the grandiose promises in the company’s S-1 filing are to be believed. Beginning on Thursday morning, when the opening bell rings on Wall Street, you will be able to invest in “the responsible development and maintenance of advanced AI for the long-term benefit of humanity.” O brave new world, that has such publicly traded stock in it.

地球仍在自转轴上旋转,沿着优雅的轨道绕太阳滑行于苍穹之间,但Anthropic的股票本周即将上市发售,所以请做好准备迎接某种行星级的混乱:金钱从天而降,时间仿佛停滞——至少,如果该公司S-1文件中那些宏大的承诺可信的话。从周四上午华尔街开市钟声响起的那一刻起,你将能够投资于“先进人工智能的负责任开发与维护,以谋求人类的长期福祉”。啊,美丽新世界,竟有如此公开交易的股票在其中。

“We believe the diffusion of AI through the global economy represents one of the largest opportunities in history to improve how humans live and work, and it is just beginning,” the company states in the filing. Anthropic is a public-benefit corporation, and its S-1 must abide by rules set by the S.E.C.; it is required to disclose risks and to explain how its goals for benefitting the public—humanity!—may affect shareholder profits. Compared with the Silicon Valley venture-capital hype of the Andreessen Horowitz variety, then, the rhetoric of Anthropic’s three-hundred-plus-page stock prospectus is restrained, and even cautious. The company, for instance, acknowledges that its technology “may pose catastrophic or existential risks to humanity” and that, even if Claude doesn’t kill everyone, public concern that it might “could adversely affect our reputation and brand.” Caveat emptor.

“我们相信,人工智能在全球经济中的扩散代表着历史上改善人类生活与工作方式的最大机遇之一,而这一切才刚刚开始,”该公司在文件中如此陈述。Anthropic是一家公益公司,其S-1文件必须遵守美国证券交易委员会制定的规则;它必须披露风险,并解释其造福公众——人类!——的目标可能如何影响股东利润。与安德森·霍洛维茨那种硅谷风投式的炒作相比,Anthropic这份三百多页的股票招股书的措辞显得克制,甚至审慎。例如,该公司承认其技术“可能对人类构成灾难性或生存性风险”,并且,即便Claude不会杀死所有人,公众对其可能如此的担忧“也可能对我们的声誉和品牌产生不利影响”。买者自负。

Of course, most of the filing concerns the prospects for profit, on the back of the promise that, “if managed well,” A.I. “could rapidly accelerate scientific innovation, greatly increase economic growth rates, bring cures for major diseases within reach, and solve problems that have resisted generations of human effort.” Is Anthropic attempting to monetize the idea of paradise?

当然,文件的大部分内容关乎盈利前景,其基础是这样一个承诺:“如果管理得当”,人工智能“可以迅速加速科学创新,大幅提高经济增长率,使重大疾病的治愈触手可及,并解决数代人类努力未能攻克的难题”。Anthropic是否在试图将天堂的理念货币化?

The company’s I.P.O. puts me in mind of a call for investment issued nearly two centuries ago, offering “small shares of twenty dollars” to anyone willing to join an “association for the improvement of the human condition.” This prospectus came in the form of a treatise published in Pittsburgh, in 1833, titled “The Paradise Within the Reach of All Men, Without Labour, by Powers of Nature and Machinery: An Address to All Intelligent Men.” Its author, a German engineer and inventor named John Adolphus Etzler who had immigrated to the U.S., claimed that solar, wind, and sea power could make the struggle for human existence obsolete. “FELLOW-MEN!” he began, “I promise to show the means for creating a paradise within ten years, where every thing desirable for human life may be had for every man in superabundance, without labour, without pay; where the whole face of nature is changed into the most beautiful form of which it be capable,” a world where a man “may provide himself with means unheard of yet, for increasing his knowledge of the world, and so his intelligence; he may lead a life of continual happiness, of enjoyments unknown yet; he may free himself from almost all the evils that afflict mankind, except death, and even put death far beyond the common period of human life, and, finally, render it less afflicting.”

该公司的首次公开募股让我想起近两个世纪前发出的一份投资号召,它向任何愿意加入“改善人类境况协会”的人提供“二十美元的小额股份”。这份招股书以一篇论文的形式出现,1833年发表于匹兹堡,题为《无需劳作、凭借自然之力与机械,人人皆可企及的天堂:致所有有识之士》。其作者是一位移民美国的德国工程师兼发明家,名叫约翰·阿道弗斯·埃茨勒,他声称太阳能、风能和海洋能可以让人类为生存而进行的挣扎成为过去。“同胞们!”他开篇写道,“我承诺展示在十年内创造天堂的方法,在那里,人类生活所需的一切美好之物,人人都能超量拥有,无需劳作,无需付费;在那里,整个自然的面貌将被改造成它所能呈现的最美形态,”在那个世界里,一个人“可以为自己获得前所未有的手段,以增进对世界的认知,从而提升自己的智力;他可以过上持续幸福的生活,享受前所未有的愉悦;他可以摆脱几乎所有折磨人类的苦难,除了死亡,甚至可以将死亡远远推迟到人类寿命的通常期限之外,并最终使其不再那么令人痛苦。”

Anthropic’s bid for shareholders rests, by decided contrast, on real, working machinery, the ingenious, unsettling advances that fall under the heading of artificial intelligence: “For the first time in history, intelligence is something humans can build.” Yet the premise of the Anthropic offering—invest in the possibility of paradise—is strikingly similar to Etzler’s technological utopianism. Anthropic proposes to “significantly accelerate global growth and dramatically improve quality of life for people around the world.” So did Etzler.

相比之下,Anthropic对股东的招揽所依托的,是真实运转的机器——那些归入人工智能名下的精巧而令人不安的进步:“有史以来第一次,智能成为人类可以建造的东西。”然而,Anthropic此次发行的前提——投资于天堂的可能性——与埃茨勒的技术乌托邦主义惊人地相似。Anthropic提出要“显著加速全球增长,大幅改善世界各地人们的生活质量”。埃茨勒当年也是这么说的。

In the prospectus, Anthropic—presumably keen to avoid the vague and controversial language of “artificial general intelligence” and “superintelligence”—refers, instead, to “Powerful AI.” By this, it means to describe artificial intelligence with “cognitive capabilities matching or exceeding those of top human experts across most professional and scientific disciplines” and possessing “the ability to interface with the physical world,” by way of robots and other devices. Achieving Powerful A.I. is Anthropic’s goal, one that it envisions accomplishing, chiefly, by way of Claude, which is fast improving itself. (“Each new version of a Claude model directly accelerates development of the next.”) The enormous investment that will almost certainly result from the I.P.O. will also allow Anthropic to “expand our own data center footprint,” necessary for the scale of compute required for the company’s growth, in both existing markets and new ones: “reaching domains from healthcare and life sciences to robotics and the physical world.” (To picture how all this will work, it’s important to set aside the idea that Claude is a chatbot on your laptop. For instance, Claude will revolutionize medical research because Claude and its minions may soon be able to run an entire lab: “our Model Hardware Standard enables AI agents to operate lab and manufacturing equipment such as microscopes, liquid handlers, and off-the-shelf robotic arms.”)

在招股书中,Anthropic——想必是为了避开“通用人工智能”和“超级智能”这类模糊且有争议的表述——转而使用了“强大AI”一词。它以此指代具备“在大多数专业和科学领域中匹敌或超越顶尖人类专家的认知能力”,并拥有“通过机器人及其他设备与物理世界交互的能力”的人工智能。实现强大AI是Anthropic的目标,它设想主要通过Claude来实现这一目标,而Claude正在快速自我改进。(“每一版新的Claude模型都会直接加速下一版的开发。”)首次公开募股几乎必然会带来的巨额投资,也将使Anthropic能够“扩大我们自己的数据中心版图”,这是公司增长所需的算力规模所必需的,既包括现有市场,也包括新市场:“从医疗保健和生命科学到机器人和物理世界等各个领域。”(要想象这一切将如何运作,关键是要抛开Claude只是你笔记本电脑上的一个聊天机器人这一想法。例如,Claude将彻底改变医学研究,因为Claude及其下属可能很快就能运营整个实验室:“我们的模型硬件标准使AI智能体能够操作实验室和制造设备,如显微镜、液体处理器和现成的机械臂。”)

If you are wondering what will be left for humans to do when Claude can do everything, you are not alone. Or maybe you are wondering how, aside from becoming a stockholder with voting rights, you’ll have any say in the future that Claude is building for you, humanity, and the planet. Be assured, Anthropic insists, that Claude—here repurposing marketing rhetoric used to describe nearly every other product to come out of Silicon Valley since the nineteen-eighties—is, somehow, good for democracy. In this case, “AI can be used to promote and strengthen democracies and bend the arc of geopolitical history in the direction of freedom.” Nothing in the prospectus, for all its mentions of stopping cyberattacks, provides any real explanation of how this can possibly be true.

如果你想知道当Claude能做一切事情时,人类还能做什么,你并不孤单。或者,也许你在想,除了成为拥有投票权的股东之外,你还能以什么方式对Claude正在为你、为人类、为这个星球构建的未来拥有任何发言权。请放心,Anthropic坚称,Claude——此处挪用了自上世纪八十年代以来用于描述硅谷几乎所有其他产品的营销话术——不知何故对民主有益。具体而言,“人工智能可以用来促进和加强民主国家,并将地缘政治历史的弧线弯向自由的方向。”尽管文件中多次提及阻止网络攻击,但其中没有任何内容对这一点何以可能为真做出任何真正的解释。

Granted, compared with OpenAI, Google, and Meta, Anthropic is, by far, the most ethically responsible of the leading A.I. companies. Yet its assurances that the Powerful A.I. it plans to develop will be good for democracy—or even, simply, good—involve gobbledygook like this statement (presumably, like Claude’s “constitution,” written at least in part by Claude): “Safety, alignment, and ethical behavior resonate particularly strongly with enterprises, where rigorous safeguards, predictable model behavior, and demonstrable accountability have become decisive criteria in selecting a partner for mission-critical workloads.” And if, instead, the robots go rogue and most humans die—except for a lucky few who, perhaps, find shelter, for centuries, in an atomic-era bunker—the survivors will be able to take comfort in the immortal words of Dr. Strangelove: “There would be no shocking memories, and the prevailing emotion will be one of nostalgia for those left behind, combined with a spirit of bold curiosity for the adventure ahead!”

诚然,与OpenAI、谷歌和Meta相比,Anthropic是迄今为止领先人工智能公司中最具道德责任感的一家。然而,它关于其计划开发的强大AI将有益于民主——甚至仅仅是有益——的保证,涉及诸如此类的官样文章(想必与Claude的“宪法”一样,至少部分由Claude撰写):“安全性、对齐性和道德行为在企业中引起特别强烈的共鸣,在这些企业中,严格的防护措施、可预测的模型行为以及可证明的问责制,已成为为关键任务工作负载选择合作伙伴时的决定性标准。”而如果事与愿违,机器人失控,大多数人类死亡——除了少数幸运儿,他们或许能在原子时代的掩体中避难数百年——幸存者将能从奇爱博士那句不朽的话中得到安慰:“不会有令人震惊的记忆,主导的情绪将是对那些被留下的人的怀旧,夹杂着对前方冒险的大胆好奇精神!”

Back in 1833, for a small investment in shares costing “not greater than the price of a lottery ticket,” Etzler promised not only extraordinary financial returns but heaven on earth. Within the span of ten years from full investment in his venture, he claimed, “the whole world might therefore be really changed into a paradise.” By the eighteen-forties, he and his followers had scrounged up enough funds to attempt to build this utopia in Venezuela. Using a not yet invented system of intricate machines that could do all things—mechanical automata, or what would now be called robots—there would be “no occasion for any work,” he said, aside from supervising the machines, of course. Alas, in Venezuela, dozens of Etzlerians either died of disease or starvation, or had to be hospitalized owing to deprivation. And, not long after the project’s failure, Etzler simply disappeared.

回到1833年,只需投入“不超过一张彩票价格”的小额股份,埃茨勒就承诺不仅有非凡的财务回报,还有人间天堂。他声称,在其事业获得全额投资的十年之内,“整个世界或许真的能变成天堂。”到了十九世纪四十年代,他和他的追随者们凑集了足够的资金,试图在委内瑞拉建造这个乌托邦。利用一套尚未发明的、能做所有事情的复杂机器系统——机械自动机,也就是现在所谓的机器人——他说,“将没有任何工作的必要”,当然,除了监督这些机器。唉,在委内瑞拉,数十名埃茨勒的信徒要么死于疾病或饥饿,要么因极度匮乏而不得不入院治疗。而在项目失败后不久,埃茨勒本人便消失得无影无踪。

Besides its doomed followers, hardly anyone paid much attention to Etzler’s treatise. But, in 1843, after a second edition was printed in London, Henry David Thoreau reviewed it in an American magazine. He mocked Etzler’s schemes and, especially, his arrogance: “We will wash water, and warm fire, and cool ice, and underprop the earth. We will teach birds to fly, and fishes to swim, and ruminants to chew the cud. It is time we had looked into these things.” But, above all, Thoreau questioned Etzler’s equation of human happiness with the increasing speed and efficiency of machines. “Suppose we could compare the moral with the physical, and say how many horse-power the force of love, for instance, blowing on every square foot of a man’s soul, would equal,” Thoreau wrote, two years before he moved to Walden Pond, to live deliberately. He grew gauzy. “Love is the wind, the tide, the waves, the sunshine,” he went on, sticking it, Biblically, to Etzler’s mechanical utopia. “Its power is incalculable; it is many horse-power. It never ceases, it never slacks; it can move the globe without a resting-place; it can warm without fire; it can feed without meat; it can clothe without garments; it can shelter without roof; it can make a paradise within which will dispense with a paradise without.”

除了那些注定遭殃的追随者之外,几乎没有人对埃茨勒的论文给予太多关注。但是,1843年,当第二版在伦敦印刷后,亨利·戴维·梭罗在一本美国杂志上对它发表了评论。他嘲笑了埃茨勒的计划,尤其是他的傲慢:“我们将洗涤水,温暖火,冷却冰,支撑地球。我们将教鸟飞翔,教鱼游泳,教反刍动物反刍。我们是时候研究这些事情了。”但最重要的是,梭罗质疑了埃茨勒将人类幸福等同于机器速度与效率不断提升的做法。“假设我们能够将道德与物理相比较,并说出,例如,爱的力量吹拂在一个人灵魂的每一平方英尺上,相当于多少马力,”梭罗写道,那是在他搬到瓦尔登湖畔去有意识地生活之前两年。他的笔触变得朦胧。“爱是风,是潮汐,是波浪,是阳光,”他继续写道,以圣经式的口吻回击埃茨勒的机械乌托邦。“它的力量不可估量;它是许多马力。它永不停息,永不松懈;它能推动地球而无需支点;它能温暖而无需火;它能喂养而无需肉;它能穿衣而无需衣物;它能庇护而无需屋顶;它能在内心创造一个天堂,从而无需外在的天堂。”

What will more intelligent and more powerful machines mean for humans? “To build AI is to harness cognition itself, and anything so powerful can be used for ill as well as good,” Anthropic concedes. The company promises many good things: better ways to fight disease, shorter lines at the D.M.V. It warns of many bad things: authoritarianism, bioweapons, cyberattacks, the extinction of the human race. How does it measure the benefit to the public? That is to be determined, on the whole, by Anthropic’s seven co-founders, who “are distinctly equipped to be stewards of our mission,” and who, collectively, get 50.1 per cent of general voting power; each has pledged eighty per cent of their personal equity to philanthropy. (An independent long-term-benefit trust separately elects a majority of directors.) What paradise might look like, in other words, is a determination to be made by a very tiny handful of men.

更智能、更强大的机器对人类意味着什么?“构建人工智能就是驾驭认知本身,而任何如此强大的东西都可以被用于恶,也可以被用于善,”Anthropic承认。该公司承诺了许多好事:更好的抗病方法,车管所更短的排队时间。它也警告了许多坏事:威权主义、生物武器、网络攻击、人类灭绝。它如何衡量对公众的益处?总体而言,这要由Anthropic的七位联合创始人来决定,他们“具备独特的条件来担任我们使命的守护者”,并且他们合计拥有50.1%的一般投票权;每人已承诺将个人股权的80%捐给慈善事业。(一个独立的长期利益信托另行选举多数董事。)换句话说,天堂可能是什么样子,要由极少数人来决定。

Artificial intelligence will not usher in a more loving future. Neither will it teach the birds to fly, or the fish to swim. Very little about A.I. has so far augured a better world. And Anthropic, in selling shares in heaven on earth, while issuing warnings about a possible descent into hell, promises nothing so much as a more profane one. ♦

人工智能不会迎来一个更有爱的未来。它也不会教鸟飞翔,或教鱼游泳。迄今为止,人工智能几乎没有预示过一个更美好的世界。而Anthropic在出售人间天堂的股份的同时,又发出可能坠入地狱的警告,它所承诺的,不过是一个更为世俗的世界。♦

2 free previews left — register to get 30 translations a month