Letter from Silicon Valley

Sam Altman Won in Court Against Elon Musk. But, Really, We All Lost

萨姆·奥特曼(Sam Altman)在法庭上赢了埃隆·马斯克(Elon Musk)。但说到底,我们全都输了。
The cases of Elizabeth Holmes and Sam Bankman-Fried at least offered a pleasant sense of comeuppance. But in Musk v. Altman, to root against Tweedledum was effectively to root for Tweedledee.
伊丽莎白·霍姆斯(Elizabeth Holmes)和山姆·班克曼-弗里德(Sam Bankman-Fried)的案子至少让人感受到一种大快人心的因果报应。但在马斯克(Musk)诉奥特曼(Altman)一案中,反对半斤,实际上就是在支持八两。
Image may contain Sam Altman Elon Musk Face Head Person Photography Portrait Adult Clothing Formal Wear and Suit
Illustration by Joan Wong; Source photographs from Getty

A famous logic puzzle takes place on a mythical island divided between the knights, who never lie, and the knaves, who always do. A foreign traveller encounters a fork in the road: one way guarantees safe passage, the other certain death. A member of each tribe is present, though it isn’t clear which is which, and the traveller is granted only one question. The solution is well known: ask either of them what the other would advise, and then to choose the opposite path. (An accurate account of a lie and an inaccurate account of the truth amount to the same wrong answer.) But this works only if someone is honest. What if nobody can be trusted? The Cretan philosopher Epimenides inspired an alternative scenario set on his own island, when he supposedly said that “all Cretans are liars.” Logicians call unstable statements like these “self-referential paradoxes,” or utterances that undermine their own claims. Epimenides would presumably have felt at home at trial of Musk v. Altman, which over the past few weeks turned an Oakland courtroom into an island of lying cretins.

一个著名的逻辑谜题发生在一个虚构的岛屿上,岛上分为永不撒谎的骑士和总是撒谎的无赖。一位异乡旅人来到岔路口:一条路通向安全,另一条则必死无疑。两个部落各有一人在场,但无法分辨谁是谁,旅人只能问一个问题。众所周知的解法是:随便问其中一人,另一人会建议哪条路,然后选择相反的方向。(对谎言的准确描述和对真相的不准确描述,最终都会指向同一个错误答案。)但这个方法的前提是有人诚实。如果谁都不能信任呢?克里特岛哲学家埃庇米尼得斯(Epimenides)据说曾说过“所有克里特人都是说谎者”,这启发了一个以他家乡为背景的变体场景。逻辑学家将这类不稳定的陈述称为“自指悖论(self-referential paradoxes)”,即那些自我否定其主张的言论。埃庇米尼得斯若置身于马斯克诉奥特曼案的庭审现场,想必会感到如鱼得水——过去几周,这场审判将奥克兰的一间法庭变成了一个满是说谎的克里特人的岛屿。

In theory, the trial was about the good-faith control of artificial intelligence. In 2015, Elon Musk and Sam Altman founded OpenAI together as a nonprofit. Its mission—“to ensure that artificial general intelligence benefits all of humanity”—was explicitly intended to counter Google’s potential dominance of the technology, which seemed almost foreordained at the time. Musk pledged up to a billion dollars to prevent that outcome. It didn’t take long for the two men to disagree over the chain of command. Each thought he alone deserved to run the show. About two years and thirty-eight million dollars later, Musk took his remaining nine hundred and sixty-odd million dollars and went home. In his valedictory e-mail, he wrote, “My probability assessment of OpenAI being relevant to DeepMind/Google without a dramatic change in execution and resources is 0%. Not 1%. I wish it were otherwise.” OpenAI needed another source of largesse. With investors in mind, it opened a for-profit subsidiary and secured billions of dollars from Microsoft and others. This past October, the company completed a lengthy process of restructuring and recapitalization. Today, the subsidiary is worth something close to a trillion dollars.

从理论上讲,这场审判关乎对人工智能的善意掌控。2015年,埃隆·马斯克与萨姆·奥尔特曼共同创立了OpenAI,当时它是一家非营利机构。其使命——“确保通用人工智能惠及全人类”——明确旨在对抗谷歌对该技术的潜在主导地位,而当时这种主导地位几乎已成定局。马斯克承诺投入高达十亿美元,以阻止这一结果。然而,两人很快就在指挥链问题上产生了分歧。他们都认为,只有自己才配执掌大局。大约两年后,在投入了三千八百万美元之后,马斯克带着他剩下的九亿六千多万美元离开了。在他的告别邮件中,他写道:“我对OpenAI能在不彻底改变执行方式和资源投入的情况下,对DeepMind/谷歌构成威胁的概率评估是0%,而不是1%。我希望事实并非如此。”OpenAI需要另一笔慷慨资助。为了吸引投资者,它成立了一家营利性子公司,并从微软及其他公司获得了数十亿美元的资金。今年十月,该公司完成了一项漫长的重组与资本重组过程。如今,这家子公司的估值已接近一万亿美元。

Musk’s lawsuit alleged that Altman, along with other OpenAI executives and in collusion with Microsoft, “stole a charity.” He believes that they solicited his generosity on false pretenses, exploiting the cover of a humanitarian cause to build one of the world’s most valuable companies, and, in the process, enriching themselves beyond measure. The remedies he sought include the unwinding of OpenAI’s transformation into a for-profit company, the disgorgement of a hundred and fifty billion dollars in damages to be paid to the original nonprofit, and the final exile of Altman from the organization. It would effectively destroy the venture as such. The suit was an act of vengeance, and its primary function seemed to be to make everyone involved look heinous.

马斯克的诉讼指控奥尔特曼与OpenAI其他高管合谋,并与微软勾结,“窃取了一家慈善机构”。他认为,对方以虚假借口骗取他的慷慨资助,借人道主义事业之名打造了全球最具价值的公司之一,并在此过程中让自己富可敌国。他寻求的补救措施包括:撤销OpenAI转型为营利性公司的决定,要求向原非营利实体支付一千五百亿美元赔偿金,以及最终将奥尔特曼逐出该组织。这实际上将彻底摧毁这家企业。这场诉讼是一场复仇行动,其主要作用似乎就是让所有相关方都显得面目可憎。

At the very least, it promised to be entertaining. During jury selection, one prospective juror assessed Musk to be “a greedy, racist, homophobic piece of garbage,” while a more restrained prospect deemed him only “a world-class jerk.” Musk’s lawyers argued that such sentiments were blatantly prejudicial. Yvonne Gonzalez Rogers, the district-court judge and one of the few trial participants who managed to acquit herself honorably, told them to suck it up. “The reality is that people don’t like him,” she said. “Many people don’t like him—but that doesn’t mean that Americans nevertheless can’t have integrity for the judicial process.” In the first week of testimony, Musk took the stand and couldn’t help but get tetchy. He seemed to have the impression that he alone understood the finer points of trial law. The Verge reporter Elizabeth Lopatto, who seemed to be live-blogging her own disgorgement (not of money but of bile), wrote, “I have never been more sympathetic to Sam Altman in my life.”

至少,这场庭审注定会是一场好戏。在陪审团遴选阶段,一位潜在陪审员评价马斯克是“一个贪婪、种族主义、恐同的垃圾”,而另一位更克制的候选人则认为他不过是“一个世界级的混蛋”。马斯克的律师辩称,这种情绪明显带有偏见。地区法院法官伊冯·冈萨雷斯·罗杰斯(Yvonne Gonzalez Rogers)——她是少数几个在这场庭审中能保持体面的参与者之一——让他们忍忍。“现实就是,人们不喜欢他,”她说,“很多人不喜欢他——但这并不意味着美国人就不能对司法程序保持诚信。”在作证的第一周,马斯克站上证人席,忍不住变得暴躁起来。他似乎认为只有自己才懂得庭审法的精妙之处。《The Verge》记者伊丽莎白·洛帕托(Elizabeth Lopatto)——她似乎在进行一场现场直播式的自我倾泻(不是金钱,而是胆汁)——写道:“我这辈子从未如此同情过山姆·奥特曼。”

Musk’s attorneys hoped that the second week—which rehashed an imbroglio from around Thanksgiving 2023, when the OpenAI board fired Altman only for Altman to return and fire almost the entire board—would mitigate Altman’s contrasting appeal. These allegations, for anyone unfortunate enough to have paid close attention, had long been rehearsed to death, and the iterative incantation of now-canonical lines from e-mails and text had the quality of operatic leitmotif: Altman had been removed, we heard again and again, for having been “not consistently candid in his communications.” The few novel revelations made him look less like a mastermind than kind of a loser. The existence of the lawsuit was almost redeemed by the release of a text thread between Altman, who was conferring with the Microsoft C.E.O., Satya Nadella, and Mira Murati, who briefly replaced Altman as chief executive. Altman plays the role of the clueless boyfriend who can’t accept that his partner is leaving him:

马斯克的律师们希望,第二周的庭审——重述了2023年感恩节前后的一场风波,当时OpenAI董事会解雇了奥特曼,结果奥特曼回归后几乎炒掉了整个董事会——能削弱奥特曼的反差魅力。对于那些不幸密切关注此事的人来说,这些指控早已被反复咀嚼到令人厌倦,而邮件和短信中那些如今已成经典的语句被一次次念诵,宛如歌剧中的主导动机:我们一遍又一遍地听到,奥特曼被解雇是因为他“在沟通中未能始终坦诚”。少数新披露的内容让他看起来不像一个幕后操纵者,倒更像一个失败者。这场诉讼的存在几乎被一条短信对话所拯救——对话双方是奥特曼(当时正与微软CEO萨提亚·纳德拉商议)和米拉·穆拉蒂(她曾短暂接替奥特曼担任首席执行官)。奥特曼在其中扮演了一个懵懂男友的角色,无法接受伴侣要离开他的事实:

ALTMAN: can you indicate directionally good or bad? satya and others anxious

奥特曼:你能大致判断一下是好是坏吗?萨提亚(Satya)和其他人都很焦虑。

MURATI: Directionally very bad

穆拉蒂:方向性极差

“Directionally” is Silicon Valley jargon for “generally.” But Altman still doesn’t get it:

“方向性上(Directionally)”是硅谷的行话,意为“大致上”。但奥特曼(Altman)依然没有明白这一点:

ALTMAN: can you wrap up soon? lots of pressure from msft for an update

ALTMAN:你能尽快收尾吗?微软那边催着要更新,压力不小。

MURATI: Sam this is very bad

穆拉蒂:山姆,这情况非常糟糕。

ALTMAN: can i come in?

奥尔特曼:我能进来吗?

MURATI: They don’t want you to

穆拉蒂:他们不希望你……

Most of the regular courtroom observers eventually gave themselves over to listlessness. Wired ran a story about the butt pillows used by OpenAI’s phalanx of lawyers and executives, including the president and co-defendant, Greg Brockman, to insulate themselves from the discomfort of the court’s benches. On a Tuesday morning of the third and final week of trial arguments, in the vacant hours of predawn Oakland, I arrived at the Ronald V. Dellums Federal Building to find a half-dozen journalists and interested civilians in the disorderly semblance of a self-organized queue. (One woman, who had severe bangs and a medieval-looking corona of braids, reminded Lopatto, the quick-witted Verge reporter, of a “stern German nanny”; she declined to provide her name or purpose and refused to recognize the authority of the line.) The general topic of desultory conversation was not the dispiriting trial of the present but the livelier intrigue of courtroom tech-dramas past—of Elizabeth Holmes, which inspired particular nostalgia, or Sam Bankman-Fried. Neither of those performances featured anybody to pull for, exactly, but at least they held out the pleasurable promise of comeuppance.

大多数常驻法庭的旁听者最终都陷入了倦怠。《连线》杂志报道了OpenAI律师和高管团队使用的坐垫——包括总裁兼共同被告格雷格·布罗克曼在内——用来缓解法庭长椅的不适。在庭审辩论第三周也是最后一周的一个周二清晨,奥克兰还笼罩在黎明前的寂静中,我来到罗纳德·V·德拉姆斯联邦大楼,发现六七名记者和感兴趣的市民正杂乱无章地自发排着队。(其中一位留着浓密刘海、盘着中世纪风格发辫的女士,让机敏的《The Verge》记者洛帕托想起了“严厉的德国保姆”;她拒绝透露姓名和来意,也不承认排队秩序的有效性。)人们漫无边际闲聊的主题并非当下令人沮丧的庭审,而是过去法庭科技剧中的精彩桥段——伊丽莎白·霍姆斯引发了特别的怀旧情绪,还有山姆·班克曼-弗里德。这两场戏码中都没有真正值得支持的角色,但至少它们提供了令人愉悦的因果报应。

Had Musk v. Altman merely been a petulant matter of injured vanity, it might have played as a diverting farce. It was instead a travesty. The underlying issues—of how A.I. ought to be governed, and by whom, and how—are of great consequence. But in this trial, to root against Tweedledum was effectively to root for Tweedledee. It was a no-win situation.

如果马斯克诉奥特曼案只是出于受伤虚荣心的任性之举,那或许还能演变成一场令人解颐的闹剧。然而,这实则是一场荒诞的闹剧。其核心问题——人工智能应当如何治理、由谁治理、以何种方式治理——事关重大。但在这场审判中,反对“糊涂蛋”实际上就等于支持“糊涂虫”。这根本是个两败俱伤的局面。

The butt pillow might have begun as a symbol of the trial’s frivolity, but it was clear soon enough that it was also a powerful metaphor for the collective failures that got us here. It was difficult, sitting in the unyielding pews, not to feel personally implicated. These were the leaders our society had somehow been assigned. Mike Isaac, a veteran tech reporter for the Times, wasn’t ashamed to admit that Brockman had inspired him to secure his own butt pillow. Isaac, a magnanimous man who looks like the actor Wilford Brimley styled as a member of the hardcore band Minor Threat, offered to share the cushion, but it struck me as somehow more appropriate to sit in the docks as a penitent. The courtroom filled up quickly in anticipation of Sam Altman, who was set to appear on the stand that day under oath. The OpenAI C.E.O. has long been known for his boyishness, but the past few years have coarsened his features and frosted his spiky hair with gray tips. He looked like a lesser vocalist for ’N Sync on a reunion tour. His presence in the courtroom had the mournful air of someone who no longer qualified as precocious.

屁股靠垫最初或许只是这场审判轻浮性的象征,但很快人们就意识到,它也是导致我们走到今天这一步的集体失败的强大隐喻。坐在那些坚硬的靠背长椅上,很难不感到自己也难辞其咎。这些就是我们社会不知何故被分配到的领导者。《纽约时报》资深科技记者迈克·艾萨克(Mike Isaac)毫不羞于承认,布罗克曼(Brockman)启发他也弄了一个屁股靠垫。艾萨克是个宽厚的人,长相酷似演员威尔福德·布里姆利(Wilford Brimley)被塑造成硬核乐队Minor Threat成员的模样,他主动提出分享靠垫,但我总觉得,坐在被告席上像个忏悔者更合适。法庭很快坐满了人,大家都在期待山姆·奥特曼(Sam Altman)当天出庭宣誓作证。这位OpenAI首席执行官长期以来以孩子气著称,但过去几年让他的面容变得粗粝,尖刺般的头发也染上了灰白的发梢。他看起来像是超级男孩('N Sync)重组巡演中一个不太出色的主唱。他出现在法庭上,带着一种不再配得上“早熟”这个标签的哀伤气息。

The basic question of the case, which is also the basic question of Altman’s career, is whether the transmogrification of OpenAI from a safety-minded nonprofit into a ravenous corporate behemoth was cynical in intention or merely in outcome. Recently, my New Yorker colleague Andrew Marantz appeared on a podcast to discuss the alternative ways to model his behavior: the “always-a-master-plan, 3-D-chess view” and the “improvisatory-checkers-all-along view.” There was a clever bit of trollery in Altman’s decision to hire as lead counsel the lawyer William Savitt, who had earlier forced Musk to follow through on his impulse to buy Twitter. Over hours of direct questioning, Savitt elicited from his brow-furrowed client a defense narrative that combined the most flattering elements of each version of the story. The part of the scheme that involved the creation of what he praised as “one of the largest charities in the world”—the nonprofit parent, by virtue of its equity stake in the for-profit subsidiary, has assets valued at more than two hundred billion dollars—was the result of what Altman repeatedly called “hard work” or “incredible work.” But the part of the scheme that involved the creation of one of the largest and most powerful for-profit companies in the world was extemporaneous—the by-product of having been “open to creative structures.” Altman said, “So this sounds a little silly to say now, but at the time, we almost didn’t start this effort because we thought Google was so far ahead that it might be hopeless to compete.”

本案的核心问题——也是奥特曼职业生涯的核心问题——在于,OpenAI 从一家注重安全的非营利组织蜕变为贪婪的企业巨兽,其意图究竟是出于愤世嫉俗,还是结果使然。最近,我在《纽约客》的同事安德鲁·马兰茨(Andrew Marantz)在一档播客节目中讨论了诠释其行为的两种不同视角:一种是“始终有宏大计划的三维象棋观”,另一种是“从头到尾即兴发挥的跳棋观”。奥特曼决定聘请律师威廉·萨维特(William Savitt)担任首席法律顾问,这招颇具挑衅意味——萨维特此前曾迫使马斯克(Musk)兑现其收购推特的冲动。在长达数小时的直接质询中,萨维特引导眉头紧锁的客户构建了一套辩护叙事,融合了两种版本中最讨喜的元素。计划中涉及创建他赞誉为“全球最大慈善机构之一”的部分——这家非营利母公司凭借其在营利性子公司的股权,坐拥超过两千亿美元的资产——是奥特曼反复强调的“艰苦努力”或“非凡工作”的成果。但计划中涉及创建全球最大、最强大的营利性公司之一的部分,却是即兴之作——是“对创新结构持开放态度”的副产品。奥特曼说:“现在这么说听起来有点傻,但当时我们几乎没打算启动这个项目,因为我们觉得谷歌领先太多,竞争可能毫无希望。”

The decision to stand up a profit-making entity was a matter of facts on the ground: the future of humanity required that OpenAI prevail in an existential battle against Google; this battle could not be fought without access to enormous pools of capital; it was impossible to court investors without the promise of returns. On these three points, everyone involved was in agreement: a dinky donor-funded charity would be taking an abacus to a data-center fight. It was acknowledged only in passing that the introduction of a fiduciary motive might create perverse incentives, and even then the worry was primarily about optics. As one of Musk’s consiglieri wrote in an e-mail, “I’m a super fan of capitalism and making tons of money doing great things, but not sure if this correlates with the ‘noble cause for humanity, not doing it to make money’ narrative.” What divided Musk and his lieutenants, on one side, from Altman, Brockman, and the OpenAI chief scientist Ilya Sutskever, on the other, was the unresolved issue of which special man got to wear the pants. In September, 2017, Musk e-mailed Sutskever and Brockman to describe a scenario in which he “would unequivocally have initial control of the company.” He insisted that he had no interest in retaining unilateral power over the destiny of the species. At some unspecified time, he continued, the authority vested in him, by him, would devolve upon an expanded board: “The rough target would be to get to a 12 person board (probably more like 16 if this board really ends up deciding the fate of the world).”

成立营利性实体的决定是基于现实考量:人类的未来要求 OpenAI 在一场关乎存亡的战役中战胜谷歌;而这场战役若没有巨额资本支持便无法进行;若不承诺回报,就不可能吸引投资者。在这三点上,所有相关方都达成共识:一个资金微薄的捐赠资助型慈善机构,就像拿着算盘去参加数据中心之战。人们只是顺带提及,引入受托人动机可能产生不正当激励,即便如此,担忧也主要停留在表面观感上。正如马斯克的一位顾问在邮件中写道:“我是资本主义的超级粉丝,也喜欢通过做大事赚大钱,但不确定这是否符合'为了人类崇高事业,不为赚钱而做'的叙事。”将马斯克及其副手与奥特曼、布罗克曼及 OpenAI 首席科学家伊利亚·苏茨克弗(Ilya Sutskever)对立起来的,是那个悬而未决的问题:究竟哪个特殊人物能掌控大局。2017年9月,马斯克给苏茨克弗和布罗克曼发邮件,描述了一个场景,其中他“将明确拥有公司的初始控制权”。他坚称自己对保留主宰人类命运的单方面权力毫无兴趣。他接着表示,在某个未明确的时间点,由他赋予自身的权力将移交给一个扩大的董事会:“粗略目标是建立一个12人董事会(如果这个董事会真要决定世界命运,可能更接近16人)。”

This organizational structure might have struck a reasonable person as a trifle undemocratic, all things considered, but what was readily clear from the trial was that Musk and Altman agreed that A.I. governance was much too serious to be left in the hands of non-player characters such as the nine assembled jurors. Altman, at times, spoke to them like children: Microsoft built them a “big computer,” but they needed “more capital to keep building larger computers.” (The ongoing effect was like the scene in “Airplane!” where Julie Hagerty’s stewardess character, upon hearing that a passenger needs to go to the hospital, asks, “A hospital? What is it?” and Leslie Nielsen’s character treats her like ditz: “It’s a big building with patients, but that’s not important right now.”) In his defense, it seemed as though the main lesson he’d gleaned from his dealings with Musk is that many grownups are best treated as toddlers. Altman testified that Shivon Zilis, a Musk confidant, onetime OpenAI board member, and the mother of some of Musk’s many children, had “counselled me over the years when dealing with Elon to remind him of things that happened in the past, because he was often upset.” The chief prerequisite for Musk’s employment seemed to be a talent for tantrum avoidance.

这种组织结构,在通盘考虑之后,或许会让一个理性的人觉得稍显不民主。但庭审中显而易见的是,马斯克和奥特曼都认为,人工智能治理事关重大,绝不能交给像九位陪审员这样的非玩家角色(non-player characters)来掌控。奥特曼有时对他们说话就像对孩子一样:微软为他们造了一台“大电脑”,但他们需要“更多资金来继续建造更大的电脑”。(这种持续的效果,就像《空前绝后满天飞》中的场景:朱莉·哈格蒂饰演的空姐听到有乘客需要去医院时,问道:“医院?那是什么?”而莱斯利·尼尔森饰演的角色把她当傻瓜一样对待:“那是个有很多病人的大建筑,但现在这不重要。”)作为辩护,他似乎从与马斯克的交往中得出的主要教训是,许多成年人最好被当作幼儿来对待。奥特曼作证说,希冯·齐里斯(Shivon Zilis)——马斯克的密友、OpenAI前董事会成员,也是马斯克众多子女中一些孩子的母亲——“多年来一直建议我,在与埃隆打交道时,要提醒他过去发生的事情,因为他经常感到不满。”马斯克任职的首要前提,似乎就是要有避免引发他闹脾气的天赋。

But Musk deserved such condescension, and the jurors did not. With the exception of Microsoft’s C.T.O., Kevin Scott, a Silicon Valley engineer of the classic “Whole Earth Catalog” variety, not a single witness seemed to regard the jurors as the sorts of people with brains. David Schizer, the former dean of Columbia Law School, provided expert testimony at a rate of fifteen hundred dollars an hour—for a total he ballparked as somewhere north of three hundred grand—to describe the relationship between the OpenAI nonprofit and its subsidiary as that of a museum to its gift shop. The implication (in a trial of freely mixed metaphors) was that the profit-seeking tail of the shop had come to wag the patrimony-preserving dog of the museum. In response, the defense produced Daniel Hemel, a law professor at N.Y.U., who was paid seventeen hundred and fifty dollars an hour to argue that the gift-shop analogy was all wrong. It would be more accurate, he said, to compare the OpenAI corporation to the Newman’s Own brand, which directed its profits to support a philanthropic network of summer camps. The dog of outdoor adventures for seriously ill children was not, in other words, being wagged by the tail of the popular salad-dressing company.

但马斯克这种居高临下的态度实在不该,陪审团更不该受此对待。除了微软首席技术官凯文·斯科特(Kevin Scott)这位典型的《全球概览》式硅谷工程师外,没有一位证人表现出对陪审团智识的尊重。哥伦比亚法学院前院长戴维·席泽(David Schizer)以每小时一千五百美元的标准提供专家证词——他估算总费用超过三十万美元——将OpenAI非营利组织与其子公司的关系比作博物馆与礼品店。这个比喻(在这场混合隐喻的审判中)暗示:礼品店追逐利润的尾巴,竟开始摇摆博物馆守护遗产的躯体。作为回应,辩方请来纽约大学法学教授丹尼尔·赫梅尔(Daniel Hemel),他以每小时一千七百五十美元的酬劳论证礼品店类比完全错误。他提出更准确的比喻应是OpenAI公司如同纽曼私传(Newman's Own)品牌——该品牌将利润用于资助夏令营慈善网络。换言之,为重症儿童提供户外探险的公益之犬,绝不会被畅销沙拉酱公司的商业之尾所左右。

The testimony consistently deployed a cavalier attitude about money. Bret Taylor, the chair of the OpenAI board, responded to an inquiry about his director’s compensation by saying, “I’m embarrassed to say I don’t know, but on the order of two hundred thousand dollars.” OpenAI’s chief futurist, a researcher named Joshua Achiam, wasn’t quite sure if he’d sold some of his equity last fall for more than twenty million dollars or less. These witnesses seemed to be hoping that the treatment of their financial windfalls as an afterthought would make them appear committed to elevated principles, but the net effect was to remind the jurors and the public that these sums represented little more than couch-cushion change to them. The winner of the expert-witness sweepstakes was the U.C. Berkeley computer-science professor Stuart Russell, the co-author of the canonical A.I. textbook and an outspoken proponent of A.I. safety. Musk’s attorneys brought him on to argue that OpenAI had played fast and loose with the technology, but Judge Gonzalez Rogers ruled that talk of existential risk was off-limits. (Musk’s lead attorney, Steven Molo, argued, “We all could die! We all could die as a result of artificial intelligence!” The judge observed that Musk might’ve endorsed this argument more sincerely if he hadn’t funded the xAI as a competitor.) Russell struggled to find something relevant to say. To prep for his unused time, he was paid for a full forty-hour workweek at a rate of five thousand dollars an hour.

证词中始终流露出一种对金钱满不在乎的态度。OpenAI董事会主席布雷特·泰勒(Bret Taylor)在被问及自己的董事薪酬时答道:“说来惭愧,我不太清楚,大概是二十万美元左右。”OpenAI首席未来学家、研究员约书亚·阿希亚姆(Joshua Achiam)也不太确定自己去年秋天是否卖出了超过两千万美元或更少的股权。这些证人似乎希望通过将巨额财富视为无足轻重之事,来彰显自己秉持崇高原则,但最终效果却是提醒陪审团和公众:这些金额对他们而言不过是沙发垫缝里的零钱。专家证人竞标赛的赢家是加州大学伯克利分校计算机科学教授斯图尔特·罗素(Stuart Russell),他是人工智能领域权威教科书的合著者,也是AI安全议题的直言不讳倡导者。马斯克的律师团队请他来论证OpenAI在技术应用上过于轻率,但法官冈萨雷斯·罗杰斯(Gonzalez Rogers)裁定不得讨论存在性风险议题。(马斯克的首席律师史蒂文·莫洛(Steven Molo)辩称:“我们都可能死!我们都可能因人工智能而死!”法官指出,如果马斯克没有资助xAI作为竞争对手,他或许会更真诚地支持这一论点。)罗素艰难地寻找着相关话题。为了准备这段最终未使用的证词,他按每周四十小时的全职工作时间获得报酬,时薪高达五千美元。

The acknowledgment of A.I.’s risk—existential threat or otherwise—was present only outside the courthouse, courtesy of a small cohort of the kinds of genteel retirees one might see a half-dozen miles north, in the Berkeley precincts of the worker-owned Cheese Board or Chez Panisse. One protester, with his face obscured by a ghoulish, oversized cartoon Musk mask, wore a Cybertruck suit made of corrugated cardboard with “Swastitruck” scrawled in runic letters. He brandished a quart-size ziplock bag labelled “Ketamine.” Three women danced around him, singing a capella about Musk and Altman’s shared depravity to the tune of “The Battle Hymn of the Republic.”

对人工智能风险的承认——无论是存在性威胁还是其他——只出现在法院之外,由一小群温文尔雅的退休人士呈现,他们就像人们往北六英里,在伯克利工人所有的奶酪店或Chez Panisse餐厅附近能见到的那种人。一名抗议者戴着阴森可怖、超大号的卡通马斯克面具,身穿用瓦楞纸板做的Cybertruck服装,上面用符文般的字体潦草地写着“Swastitruck”。他挥舞着一个标着“氯胺酮”的一夸脱密封袋。三名女子围着他跳舞,以《共和国战歌》的曲调无伴奏合唱,唱的是马斯克和奥特曼共同的堕落。

Altman didn’t exactly come off well, though he did get in some dryly amusing lines about having to endure Musk’s desire to show everyone memes on his phone. It didn’t really matter insofar as the evidence against Musk was challenging to refute. As document after document showed, Musk was well aware of the various plans under way to outfit their “summer-camp charity” with a substantial arm dedicated to the wanton sale of salad dressing. Musk had no problem going forward with a for-profit structure; his preference was to fold it into Tesla. (Altman testified that he’d looked askance at this plan: “Tesla is a car company.”) Even if Altman and others had in fact stolen his charitable valor, it was plain that his complaint did not fall within the three-year statute of limitations. But neither Musk nor his lawyers seemed to have embarked upon this case with the goal of juridical victory. It instead provided them with an opportunity to cast Altman, in advance of OpenAI’s expected I.P.O., as an untrustworthy guy. Molo, Musk’s attorney, is a stooped, pallid, funereal man with enormous hands, and he carried out his cross-examination with the solemnity of an undertaker. The first few minutes were supposed to be, I’m sure, a tour de force:

奥特曼的表现算不上出色,不过他确实说了几句干巴巴的俏皮话,抱怨自己不得不忍受马斯克想给所有人看他手机里表情包的冲动。这其实无关紧要,因为针对马斯克的证据很难反驳。随着一份份文件被展示出来,马斯克显然很清楚,他们正在为那个“夏令营慈善机构”配备一支专门负责随意兜售沙拉酱的庞大团队。马斯克对推进营利性结构毫无异议;他更倾向于将其并入特斯拉。(奥特曼作证时表示,他对这个计划不以为然:“特斯拉是一家汽车公司。”)即便奥特曼和其他人确实窃取了他慈善事业的功劳,但很明显,他的投诉已超出三年的诉讼时效。然而,无论是马斯克还是他的律师,似乎从一开始就没打算以法律胜利为目标。相反,这起案件为他们提供了一个机会,在OpenAI预计的首次公开募股(I.P.O.)之前,将奥特曼塑造成一个不可信赖的人。马斯克的律师莫洛(Molo)是个弯腰驼背、面色苍白、神情阴郁的人,双手巨大,他以殡仪馆工作人员般的肃穆态度进行了交叉询问。前几分钟,我敢肯定,本该是一场精彩绝伦的表演:

MOLO: Are you completely trustworthy?

MOLO:你完全值得信赖吗?

ALTMAN: I believe so.

奥特曼:我相信是的。

MOLO: But you don’t know whether you’re completely trustworthy?

莫洛:但你并不确定自己是否完全值得信赖?

ALTMAN: I’ll just amend my answer to yes.

奥尔特曼:我改一下回答,答案是肯定的。

MOLO: Should the jury believe your testimony?

MOLO:陪审团是否应当相信你的证词?

ALTMAN: I think that’s up to them, but I believe so.

奥特曼:我认为这取决于他们自己,但我相信是的。

MOLO: You believe so, or they should?

MOLO:是你这么认为,还是他们本该如此?

ALTMAN: Sir, I’m not gonna tell the jury what to think.

奥尔特曼:先生,我不会告诉陪审团该想什么。

MOLO: Do you always tell the truth?

莫洛:你总是说实话吗?

ALTMAN: I believe I’m a truthful person.

奥特曼:我相信自己是个诚实的人。

MOLO: That wasn’t my question, sir. Do you always tell the truth?

莫洛:先生,我问的不是这个。您是否总是说实话?

ALTMAN: I’m sure there is some time in my life when I have not.

ALTMAN:我确信,在我生命中的某个时刻,我并非如此。

Molo went on to recite previous testimony from various former employees or board members, who offered that Altman had dissembled about OpenAI’s safety protocols and created a “toxic culture of lying.” Altman suggested that he hadn’t heard their comments. Did Altman know that Dario Amodei, who worked alongside Altman at OpenAI before he left to found its rival Anthropic, had accused him of “misrepresenting the terms” of the initial Microsoft investment? Altman, who unlike Musk has a recognizable sense of humor, remarked, “Dario has accused me of many things.” At one point, perhaps ten minutes into this droning line of questioning, Molo asked Altman if he’d actually read the lawsuit. “Many versions, sir,” Altman said. The dubiousness of Altman’s character is of paramount importance to the world in general, but it seems the case—at least directionally—that it has, by this time, been priced into his reputation. To relitigate the matter on behalf of Musk only served to underline the conviction that all of these jerks deserved one another.

莫洛接着引述了多位前员工或董事会成员的证词,他们指出奥特曼曾在OpenAI安全协议上闪烁其词,并营造出“充满谎言的毒性文化”。奥特曼暗示自己并未听闻这些言论。奥特曼是否知道,曾与他共事于OpenAI、后离职创办竞品公司Anthropic的达里奥·阿莫迪(Dario Amodei),曾指控他“歪曲条款”——关于微软首轮投资的条款?与马斯克不同,奥特曼显然具备幽默感,他回应道:“达里奥指控过我的事情可多了。”在持续约十分钟的单调质询中,莫洛突然问奥特曼是否真的读过起诉书。“读过好几个版本,先生。”奥特曼答道。奥特曼品性可疑这一点,对全世界而言至关重要,但至少从当前趋势来看,这似乎已纳入其声誉的定价之中。如今替马斯克重翻旧账,只会强化一个共识:这帮混蛋就该彼此纠缠。

Besides, the cumulative effect of Molo’s sustained effort to demand Altman’s self-incrimination seemed to remind the courtroom that most of us, irrespective of our own faults, tend to think of ourselves as credible people doing our best. This, of course, is the real problem. Musk attempted to color Altman as a uniquely unsuitable supervisor of this technology, but this invariably invited further scrutiny into his own abject unfitness for the role. The only proper response to any of this is to point out that something as flimsy and corruptible as individual character was always going to be an insufficient basis for A.I. governance.

此外,莫洛持续要求奥特曼自证其罪的累积效应,似乎提醒着法庭:我们大多数人,无论自身有何缺陷,都倾向于自认是可信之人,且已竭尽全力。这,当然,才是真正的问题所在。马斯克试图将奥特曼描绘成这项技术独一无二的不合格监管者,但这反而不可避免地引来了对他自身是否胜任这一角色的更深入审视。对此唯一恰当的回应,是指出像个人品格这般脆弱且易被腐蚀的东西,从来就不足以成为人工智能治理的根基。

To claim that OpenAI’s mission of cultivating beneficial A.I. was compromised by Sam Altman is to let the entire industry off the hook. Yes, Altman seems to have a rather casual relationship with the truth. But it is far more interesting, complicated, and useful to take his self-defense at face value—to interpret the many sins of OpenAI, and its competitors, as the result of a good-faith exercise in futility. What if we imagined that he did in fact set out to do good? And discovered—or, perhaps better, failed to discover—along the way that this was structurally impossible? There are signs that some part of him anticipated his enterprise’s destiny from the beginning. In the earliest correspondence about what would become OpenAI, Altman seemed vaguely aware that the collective-action problem of A.I. governance had nothing to do with individual heroism. In 2015, he e-mailed Musk to say that the development of advanced A.I. was virtually an inevitability. He wrote that “if it’s going to happen anyway, it seems like it would be good for someone other than Google to do it first.” What he had in mind, he continued, was not exactly a rival outfit. It was, instead, something like a “Manhattan Project for AI” Perhaps the most salient thing about the Manhattan Project, however, was, that it was not entrusted to the private sector. To launch such a transformational and dangerous program as a commercial initiative was reckless and hubristic; a “Manhattan Project run by a wealthy cabal” was as much a self-referential paradox as “all Cretans are liars.”

若将OpenAI培育有益人工智能的使命被萨姆·奥特曼(Sam Altman)所玷污归咎于他一人,那便是让整个行业都逃脱了责任。没错,奥特曼与真相的关系似乎颇为随意。但更有趣、更复杂、也更有益的做法是,将他的自我辩护当真——将OpenAI及其竞争对手的诸多罪过,视为一场善意却徒劳的尝试。如果我们设想他确实初衷向善呢?并在过程中发现——或者更准确地说,未能发现——这在结构上根本不可能实现?有迹象表明,他内心的一部分从一开始就预见到了自己事业的宿命。在关于后来成为OpenAI的最早通信中,奥特曼似乎隐约意识到,人工智能治理的集体行动问题与个人英雄主义毫无关系。2015年,他给马斯克(Musk)发邮件说,先进人工智能的发展几乎是不可避免的。他写道:“如果它无论如何都会发生,那么让谷歌以外的其他人先做这件事,似乎也不错。”他接着说,他心中所想的并非一个纯粹的竞争对手,而更像是一个“人工智能的曼哈顿计划”。然而,曼哈顿计划最显著的一点或许是,它并未委托给私营部门。将如此具有变革性和危险性的项目作为商业计划启动,是鲁莽且狂妄的;一个“由富豪集团运作的曼哈顿计划”,其自相矛盾的程度,不亚于“所有克里特人都是说谎者”。

On Monday, the jury took only two hours to reach its verdict. Musk’s complaint, the panel found, had indeed exceeded the statute of limitations. If Musk had really thought that his beloved charity had been stolen, it would have occurred to him to raise the issue long before OpenAI had become what it is today. If he hadn’t cared before the launch of ChatGPT, he had no right to pretend to care now.

周一,陪审团仅用两小时便达成裁决。该小组认定,马斯克的申诉确实已超出诉讼时效。倘若他当真认为心爱的慈善机构遭人窃取,早在OpenAI发展至今日规模之前,他就该提出异议。既然在ChatGPT问世前他未曾在意,如今便无权佯装关切。

In a better world, Altman might emerge from this humiliating rite with some genuine humility. This is almost certainly too much to expect. Still, his courtiers seemed to be encouraging him to lean into sheepishness. On the last morning of testimony, Mike Isaac, of the Times, tweeted an image of his sad excuse for a courthouse lunch: a meat-based protein bar, Bumble Bee-brand Snack on the Run! Tuna Salad & Crackers, and vessels of sugary caffeine. After the morning break, Isaac sat down next to me on his butt pillow and tweeted a follow-up: “a good samaritan who took pity on me (and whose identity i will protect) came and gave me a bagel and cream cheese.” I’d been standing nearby when Altman made the offering of the paper-plated bagel. Looking like a chastened little boy, he said, softly, “My comms team told me to give you this.” ♦

在一个更理想的世界里,奥特曼或许能在这场屈辱的仪式后生出几分真正的谦逊。但这几乎注定是奢望。不过,他的随从们似乎正鼓励他摆出一副羞愧的姿态。庭审最后一天的上午,《纽约时报》的迈克·艾萨克在推特上晒出了他寒酸的法庭午餐:一块以肉类为主的蛋白棒、Bumble Bee品牌的“随身零食”金枪鱼沙拉配饼干,以及几杯含糖咖啡因饮料。上午休庭后,艾萨克坐在他的坐垫上,挨着我,又发了一条推文:“一位同情我的好心人(我会保护其身份)走过来,给了我一个百吉饼和奶油奶酪。”当时我就站在附近,看到奥特曼递上那个装在纸盘里的百吉饼。他像个被训斥的小男孩,轻声说道:“我的公关团队让我把这个给你。”♦

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