塔瑪拉·尼斯/文　俞月圓/譯

AI technologies promise more chatbots and replicas of people who have passed. But giving voice to the dead comes at a human cost.

人工智能技術有望帶來更多的聊天機器人和逝者“復制品”，但讓逝者發(fā)聲需要生者付出代價。

Giving enough data， one can feel like its possible to keep dead loved ones alive. With ChatGPT and other powerful large language models， it is feasible to create a more convincing chatbot of a dead person. But doing so， especially in the face of scarce resources and inevitable decay， ignores the massive amounts of labor that go into keeping the dead alive online.

只要有足夠的數(shù)據(jù)，人們就會覺得死去的親人能夠繼續(xù)活著。有了ChatGPT和其他厲害的大型語言模型，就可以造出更接近逝者的聊天機器人，但這種做法——尤其是考慮到資源稀缺，衰亡難免——忽略了讓逝者存活于線上所需付出的大量勞動。

Someone always has to do the hard work of maintaining automated systems， as demonstrated by the overworked and underpaid annotators and content moderators behind generative AI， and this is also true where replicas of the dead are concerned. From managing a digital estate after gathering passwords and account information， to navigating a slowly-decaying inherited smart home， digital death care practices require significant upkeep. Content creators depend on the backend labor of caregivers and a network of human and nonhuman entities， from specific operating systems and devices to server farms1， to keep digital heirlooms alive across generations. Updating formats and keeping those electronic records searchable， usable， and accessible requires labor， energy， and time. This is a problem for archivists and institutions， but also for individuals who might want to preserve the digital belongings of their dead kin.

維護自動化系統(tǒng)的艱苦工作總要有人來做，比如生成式人工智能背后的注釋者和內容審核者，他們加班加點地工作卻得不到合理的報酬。復制逝者所涉及的工作也是如此。從收集密碼和賬戶信息后管理數(shù)字資產，到設法操作慢慢衰亡的智能家居遺產，數(shù)字殯葬服務需要做大量的維護工作。要想將數(shù)字遺物一代代傳下去，內容創(chuàng)作者就要依賴護理人員的后臺勞動，以及由人類和非人類實體（比如特定的操作系統(tǒng)和設備，還有服務器場）構成的一整個網絡。更新數(shù)據(jù)存儲格式并確保這些電子記錄永遠可搜索、可使用、可訪問，需要付出勞力、精力和時間。這不僅是檔案管理人員和管理機構面臨的問題，也是希望為已故親屬保存數(shù)字財產的個人面臨的問題。

And even with all of this effort， devices， formats， and websites also die， just as we frail humans do. Despite the fantasy of an automated home that can run itself in perpetuity or a website that can survive for centuries， planned obsolescence2 means these systems will most certainly decay. As people tasked with maintaining the digital belongings of dead loved ones can attest， there is a stark difference between what people think they want， or what they expect others to do， and the reality of what it means to help technologies persist over time. The mortality of both people and technology means that these systems will ultimately stop working.

即使付出了這么多努力，設備、數(shù)據(jù)儲存格式和網站仍然會像我們這些脆弱的人類一樣死去。盡管人們幻想擁有可以永久運行的自動化住宅，或者可以存活幾個世紀的網站，但計劃性報廢策略意味著這些系統(tǒng)肯定會衰亡。負責維護已故親人數(shù)字財產的人可以證實，人們認為自己想要的東西（或者說他們期望別人做到的事情）與幫助技術長期存續(xù)下去的現(xiàn)實內涵之間，存在天壤之別。人和技術的壽命都有限，因此這些系統(tǒng)終將停止工作。

Early attempts to create AI-backed replicas of dead humans certainly bear this out. Intellitars Virtual Eternity， based in Scottsdale， Arizona， launched in 2008 and used images and speech patterns to simulate a humans personality， perhaps filling in for someone at a business meeting or chatting with grieving loved ones after a persons death. Writing for CNET3， a reviewer dubbed Intellitar the product “most likely to make children cry.” But soon after the company went under in 2012， its website disappeared. LifeNaut， a project backed by the transhumanist4 organization Terasem—which is also known for creating BINA48， a robotic version of Bina Aspen， the wife of Terasems founder—will purportedly combine genetic and biometric information with personal datastreams to simulate a full-fledged human being once technology makes it possible to do so. But the projects site itself relies on outmoded Flash software， indicating that the true promise of digital immortality is likely far off and will require updates along the way.

制造基于人工智能的逝者復制品的早期嘗試無疑證明了這一點?？偛课挥趤喞Ｄ侵菟箍拼拇鸂柕腎ntellitar公司于2008年推出了一款名為“虛擬永生”的產品。該產品利用圖像和語音模型來模仿人的音容笑貌，也許可以在某人去世后代替他參加商務會議，或者與其悲傷的親人聊天。一位評論員在為CNET網站寫的文章中說，“虛擬永生”是“最有可能把小孩嚇哭”的產品。2012年，Intellitar公司倒閉，沒過多久其網站也消失了。倡導超人類主義的組織Terasem因創(chuàng)造了BINA48（依照Terasem創(chuàng)始人的妻子比娜·阿斯彭制造的機器人）而聞名。該組織資助了一個名為LifeNaut的項目，聲稱一旦技術成熟，該項目可以把基因和生物識別信息與個人數(shù)據(jù)流結合起來，模擬出一個形神俱備的人類。不過，該項目的網站本身還依賴于過時的Flash軟件，這表明真正實現(xiàn)數(shù)字永生可能還遙遙無期，而且需要不斷升級更新。

With generative AI， there is speculation that we might be able to create even more convincing facsimiles of humans， including dead ones. But this requires vast resources， including raw materials， water， and energy， pointing to the folly of maintaining chatbots of the dead in the face of catastrophic climate change. It also has astronomical financial costs： ChatGPT purportedly costs $700，000 a day to maintain. This is not a sustainable model for immortality.

有了生成式人工智能，有人就猜測我們也許能夠制造出更接近逝者乃至全人類的復制品，但這需要大量資源，包括原材料、水和能源。由此可見，在災難性的氣候變化之下，維護逝者聊天機器人是何等愚蠢。這種做法還會帶來數(shù)額驚人的經濟成本。據(jù)說ChatGPT每天的維護費用高達70萬美元。這不是一種可持續(xù)的永生模式。

There is also the question of who should have the authority to create these replicas in the first place： a close family member， an employer， a company？ Not everyone would want to be reincarnated as a chatbot. In a 2021 piece for the San Francisco Chronicle， the journalist Jason Fagone recounts the story of a man named Joshua Barbeau who produced a chatbot version of his long-dead fiancée Jessica using OpenAIs GPT-3. It was a way for him to cope with death and grief， but it also kept him invested in a close romantic relationship with a person who was no longer alive. This was also not the way that Jessicas other loved ones wanted to remember her; family members opted not to interact with the chatbot.

另一個問題是：先不說別的，誰有權決定創(chuàng)造逝者復制品——是近親、雇主，還是某家公司？并非每個人都愿意轉世成聊天機器人。在《舊金山紀事報》2021年的一篇報道中，記者賈森·法戈內講了一個故事：一個叫喬舒亞·巴爾博的人以他逝世已久的未婚妻杰茜卡為原型，用OpenAI公司開發(fā)的GPT-3制造了一個聊天機器人。這是他應對死亡與悲傷的一種方式，卻也讓他為了與一個逝者維持親密浪漫關系而不斷付出。杰茜卡的其他親人也不希望以這種方式記住她，她的家屬寧可不與聊天機器人互動。

In the end， though， it is developers and companies， not loved ones， who have control over how long chatbots persist. The experimental program used to create Jessica Bot planned mortality into the system; to save on operating costs， the developer intentionally built it to decay and die. Barbeau had to be judicious with his time with Jessica， and stopped interacting with the chatbot before its battery drained too far so he wouldnt have to directly confront another form of death.

不過，最終掌控聊天機器人存活多久的是開發(fā)者和開發(fā)公司，而不是逝者的親人。用來制造杰茜卡機器人的試驗程序在系統(tǒng)中設計了有限的壽命。為了節(jié)約運營成本，開發(fā)者有意讓杰茜卡機器人走向衰敗和死亡。巴爾博不得不小心利用與杰茜卡機器人相處的時間，他在電池快要耗盡之前停止與其互動，這樣他就不必直面另一種形式的死亡。

In some cases， developers might want to believe that chatbots can go on indefinitely， helping the dead replace living workers altogether. For example， the SAG-AFTRA5 strike raises questions about the personality rights of the dead， as generative AI can potentially be used to revive dead actors. Similar issues have already played out with various deepfake versions of dead celebrities， including the controversial use of Anthony Bourdains deepfaked voice in a documentary about his life， and in a more mundane， low-tech case， a professors recorded lectures allowing him to continue working at a university after his death. But such re-creations， in addition to potentially harming living workers， may also upset the surviving loved ones of the dead and cause new bureaucratic headaches for them as they attempt to seek control over the replicas of their dead family members. AI versions of people may be created without the knowledge or consent of living kin.

在某些情況下，開發(fā)者可能愿意相信聊天機器人可以無限期維持下去，幫助死去的人完全取代活著的勞動者。例如，美國演員工會罷工引出了有關逝者人格權的問題，因為生成式人工智能有可能被用來“復活”故去的演員。類似的問題已經在各種深度偽造的已故名人身上出現(xiàn)。比如，一部關于安東尼·波登生平的紀錄片使用了他的深度偽造聲音，從而引發(fā)爭議。還有一個更貼近生活、技術含量更低的案例：一位教授的講座錄音使他得以在去世后繼續(xù)在大學工作。但是，這種再現(xiàn)不僅有可能傷害活著的勞動者，還可能會讓逝者的遺屬感到不安，害他們在試圖尋求掌控逝去家人的復制品時，遇上官僚主義帶來的新麻煩。人工智能復制出來的人可能是在活著的親屬不知情或者不同意的情況下制造出來的。

The revival of the dead through AI reveals the power relations， infrastructures， and networked labor behind all forms of digital production. We can create lifelike replicas of the dead， but these replicas rely on the work of the living to persist： As is the case with digital legacies in general， someone has to pay for domain names， delete spam messages， answer emails to dead accounts， update file formats， or move files from one hard drive to another as websites and companies disappear. For those who are left behind to care indef-initely for these creations， the psychological costs may be too much to bear.

借助人工智能復活逝者，揭示了所有形式的數(shù)字化生產背后的權力關系、底層結構與網絡化勞動。我們可以制造出栩栩如生的逝者復制品，但這些復制品依賴生者的勞動而存在。和處理一般的數(shù)字遺產一樣，必須有人支付域名費用、刪除垃圾郵件、回復逝者賬戶收到的電子郵件、更新文件格式，或者在網站或公司消失前將文件從一個硬盤轉移到另一個硬盤。對于那些要無限期照顧這些復制品的人來說，心理代價可能大到難以承受。

（譯者為“《英語世界》杯”翻譯大賽獲將者）

1服務器場是在單個場點的機架上安裝的通過網絡設備連接在一起的一組服務器，用以滿足對于遠大于單臺機器的處理能力的需求。? 2計劃性報廢指一種控制產品壽命以鼓勵購買者替換產品的策略。

3美國知名科技網站。? 4超人類主義是一場國際性的文化和智力運動，其最終目標是通過開發(fā)和廣泛使用技術來極大地提高人類的智力、體力和心理能力，從根本上改變人類的狀況。擁護超人類主義的思想家研究可以克服人類基本局限性的新興技術有什么潛在的好處和危險，同時也研究開發(fā)和使用這些技術所涉及的倫理問題。

5 = The Screen Actors Guild-American Federation of Television and Radio Artists美國演員工會。