尊敬的用戶您好，這是來自FT中文網的溫馨提示：如您對更多FT中文網的內容感興趣，請在蘋果應用商店或谷歌應用市場搜尋「FT中文網」，下載FT中文網的官方應用。
Another week, another record high in US equity markets. Last week’s jump was triggered by the Federal Reserve’s signal that investors can look forward to more interest rate cuts this year. But deeper market bullishness is built on two things: the cash reserves of the tech giants that now dominate the markets, and belief in their ability to monetise artificial intelligence.
又是一週，美股再創新高。上週的漲幅由美聯準的信號觸發，該信號表明投資者今年可以期待更多次的降息。但市場更深層的看漲情緒基於兩大因素：一是目前主導市場的科技巨擘們手中龐大的現金儲備，二是對它們透過人工智慧賺錢的能力抱有堅定的信心。
AI will “change the world”, we are told. It will radically increase productivity (albeit by disrupting millions of jobs). It will create a huge new wealth pie for the world to share. And, according to a breathless ARK Invest report that last week predicted a $40tn boost to global gross domestic product from AI by 2030, it will “transform every sector, impact every business, and catalyze every innovation platform”.
我們被告知，人工智慧將「改變世界」。它將根本性地提升生產率（雖然這意味著數百萬崗位的顛覆）。同時，它也將創造一個龐大的新財富「蛋糕」，供全球共享。根據方舟投資(ARK Invest)上週發佈的一份令人興奮的報告，預計到2030年，人工智慧將爲全球國內生產總值增加40兆美元，它將「轉型每一個行業，影響每一家企業，併成爲每一個創新平臺的催化劑」。
It’s the euphoria and sense of inevitability in this straightforward narrative that makes me nervous. Even if you believe AI will be today’s equivalent of electricity or the internet, we are at the very early stages of a highly complex multi-decade transformation that is by no means a done deal. Yet valuations are pricing in the entire sea change, and then some. A February report by Currency Research Associates pointed out that it would take 4,500 years for Nvidia’s future dividends to equal its current price. Talk about a long tail.
這種直接敘述中的狂喜和必然感讓我感到緊張。即使你相信人工智慧將成爲當今的電力或網路的等價物，我們仍處於一個高度複雜的、跨幾十年的轉型的早期階段，這絕不是一個已經確定無疑的事情。然而，市場估值已經預期了整個鉅變，甚至更多。貨幣研究協會(Currency Research Associates)在2月份的一份報告指出，輝達(Nvidia)未來的股息需要4500年才能與其當前價格相等。這真是一個長尾(long tail)。
While Nvidia isn’t Pets.com — it has tangible revenues from selling real things — the overall AI narrative depends on many uncertain assumptions. For example, AI requires huge amounts of water and energy. There’s a push in both the US and EU to get companies to disclose their usage. Whether via carbon pricing, or a tax on resource usage, it’s quite likely that those input costs will rise significantly in the future.
雖然輝達並非Pets.com——它透過銷售實物獲得了實實在在的收入——但整個人工智慧的發展依賴於許多不確定的假設。例如，人工智慧需要大量的水和能源。美國和歐盟都在推動企業公開其資源使用情況。無論是透過碳定價還是對資源使用徵稅，這些投入成本在未來都有可能大幅上升。
Likewise, AI developers don’t now have to own the copyright to content on which the models are trained. They don’t have to make profits on AI itself, of course; the assumption of future gains is enough to fuel the froth. Relentless techno-optimism and the illusion of inevitability is how Silicon Valley creates paper wealth. But remember, many of the proponents of “AI everywhere” were touting web3, crypto, the metaverse and the benefits of the gig economy not so long ago.
同樣，人工智慧開發者現在不必擁有其模型訓練所依賴內容的版權。他們自然不需要直接從人工智慧本身獲利；僅僅對未來潛在收益的預期就足以推動這股熱潮。不懈的技術樂觀主義和對未來不可避免性的錯覺是矽谷創造紙面財富的手段。但請記住，那些現在鼓吹「人工智慧無所不在」的人，不久前還在大力推廣web3、加密貨幣、元宇宙以及零工經濟的種種好處。
One big difference, of course, is that AI has been validated by huge, cash-rich, market-leading companies such as Microsoft, Google and Amazon. But even within those companies developers have their doubts. One senior staffer at a leading AI company recently admitted to me, when pushed, that the profit assumptions around the technology were based “more on speculation than substance”, and that it has major kinks still to be worked out.
當然，一個顯著的區別是，人工智慧已經得到了像微軟(Microsoft)、谷歌(Google)和亞馬遜(Amazon)這樣的大型、資金充足、市場領先的公司的認可。然而，即使在這些公司內部，開發者也存在疑慮。一位在一家領先的人工智慧公司的高級員工最近在我追問下承認，關於這項技術的盈利預期「更多的是基於猜測，而非實質」，並且它還有一些重大的問題有待解決。
Anyone who’s experimented with large language models can vouch for this. I wouldn’t rely on a chatbot when doing research for my own work because I don’t want to worry about the accuracy of the data I’m being fed. I also don’t want to give up my ability to curate my own informational inputs. (I’d much rather do a Google search and see sources and citations laid out.)
任何體驗過大型語言模型的人都會同意這一點。我在進行個人工作研究時，不會依賴聊天機器人，因爲我不想擔心我所接收數據的準確性。我也不想放棄自己篩選資訊輸入的權力。（我更傾向於進行谷歌搜尋，檢視資料來源和引用。）
I’m admittedly operating at the high end of the white-collar job spectrum. But even for more rote middle-market tasks, there are lots of questions about how to integrate AI into workflows, and whether it will really be more productive than the humans it may replace. And the humans are beginning to revolt. The Hollywood writers’ strikes were at their core about control of AI, and unions are taking on the issue of technology regulation more broadly.
無可否認，我所從事的屬於白領工作的高階領域。但即便對於相對機械的中端市場任務，關於如何將人工智慧整合進工作流程，以及它是否真能比它可能取代的人類更高效，都存在諸多疑問。而且，人類已經開始抗議了。好萊塢編劇的罷工在本質上是對人工智慧控制權的爭奪，而工會則在更廣泛地對技術監管問題發起挑戰。
Meanwhile, the copyright backlash against AI is gaining steam. Last week, French regulators fined Google €250mn for failing to notify news publishers that it was using their articles to train its AI algorithms, and for not licensing fair deals. This follows similar suits against OpenAI and Microsoft brought by the New York Times. As AI works its way into proprietary corporate data sets, opportunities for litigation over copyright will increase, and possibly even dovetail with worker complaints over corporate surveillance.
與此同時，針對人工智慧的版權反彈正在加劇。上週，法國監管機構對谷歌罰款2.5億歐元，原因是谷歌未向新聞出版商通知其正在使用他們的文章來訓練其人工智慧演算法，也未就公平交易達成許可。先前，紐約時報(New York Times)對OpenAI和微軟提起了類似的訴訟。隨著人工智慧逐漸滲透到企業專有數據集，版權訴訟的機會將會增加，甚至可能與員工對企業監控的投訴相交織。
Then there’s the monopoly problem. As Meredith Whittaker, president of the Signal Foundation and the co-founder of the AI Now Institute, wrote in 2021, modern AI advances are “primarily the product of significantly concentrated data and compute resources that reside in the hands of a few large tech corporations”. Our increasing reliance on such AI, Whittaker added, “cedes inordinate power over our lives and institutions to a handful of tech firms”.
然後是壟斷問題。正如信號基金會(Signal Foundation)主席、AI Now Institute的聯合創辦人梅雷迪思•惠特克(Meredith Whittaker)在2021年所寫，現代人工智慧的進步「主要是集中在少數幾家大型科技公司手中的數據和計算資源的產物」。惠特克補充說，我們對這種人工智慧的日益依賴，「將我們生活和機構的過度權力拱手讓給了少數幾家科技公司」。
The so-called Magnificent Seven companies have driven AI enthusiasm and stock market gains over the past year. They have pushed the concentration of the S&P 500 to a historic extreme. But as a recent Morgan Stanley Wealth Management report notes, “index concentration has historically proved self-correcting, with some combination of regulatory, market and competitive forces, along with business cycle dynamics, undermining static leadership”. The report says “analysis suggests that equity returns have typically struggled following peaks in concentration”.
過去一年中，被稱爲「瑰麗七股」的公司推動了對人工智慧的熱情，並帶動了股市的上漲，使得標普500指數的集中度達到了歷史高點。但如摩根士丹利財富管理最近的報告所指出，「指數的集中度歷史上往往能自行調整，透過監管、市場和競爭等力量的結合作用，以及商業週期的變化，來削弱固定的領先地位。」該報告還指出，「分析顯示，在集中度達到頂峯之後，股權回報率通常會遇到挑戰。」
That combination of correcting factors might include the growing number of Big Tech antitrust cases and the possibility that carbon pricing and copyright fines will challenge the “free” inputs necessary to make a profit.
這些可能的糾正因素包括大型科技公司反壟斷案件的增加，以及碳定價和版權罰款對於實現利潤所需「免費」輸入材料構成的挑戰。
Whether you see AI as the next tulip bubble or the next combustion engine, it’s worth questioning how the market is pricing this story.
無論你視人工智慧爲下一場鬱金香泡沫，還是認爲它將成爲下一代內燃機，質疑市場如何給這個故事定價都是有價值的。