Meta的祖克柏表示，AI實驗室已有足夠誘因安全開發

作者：克里斯・湯瑪斯By Chris Thomas

9月15日（路透）——Meta Platforms執行長馬克・祖克柏（Mark Zuckerberg）週二表示，競爭與法律責任已讓人工智慧（AI）公司有足夠理由各自採取安全行動。這番說法似乎有別於頂尖AI公司領袖呼籲協調放慢AI開發步調的主張。Sept 15 (Reuters) - Meta Platforms CEO Mark Zuckerberg said on Tuesday that competition and liability give AI companies enough reason to act individually on safety, appearing to break from calls by leaders of top AI firms for a coordinated slowdown in AI development.

這些言論凸顯產業高層與美國當局之間的分歧：面對外界警告這項技術的進展速度快到企業難以控制，應如何因應？部分業內人士警告，AI對人類構成生存威脅。The comments underscore a split among industry executives and U.S. authorities over how to respond to warnings that the technology is advancing faster than companies can control it, with some industry insiders warning it poses an existential threat to humanity.

祖克柏週二在X平台發文表示，每一個實驗室都有責任，也有誘因，以安全訓練模型所需的速度推進，並且有能力自行採取行動確保做到這一點。Zuckerberg, in an X post on Tuesday, said that every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.

他發表上述言論的三天前，Anthropic執行長達里歐・阿莫迪（Dario Amodei）敦促AI公司放慢提升模型能力的速度；此前，多名研究人員才對AI的生存風險提出警告。His remarks come three days after Anthropic CEO Dario Amodei urged AI companies to slow the pace at which they improve model capabilities, on the heels of existential risk warnings by several researchers.

競爭對手OpenAI執行長山姆・奧特曼（Sam Altman）與xAI的伊隆・馬斯克（Elon Musk）都在數小時內公開支持阿莫迪的呼籲。Rival OpenAI's chief Sam Altman and xAI's Elon Musk both publicly endorsed Amodei's call within hours.

川普政府反對AI監管與放慢開發ADMINISTRATION PUSHBACK ON AI REGULATION, SLOWDOWN

然而，美國總統川普週一淡化外界對AI潛在危害的擔憂，稱美國已為這項技術設有防護措施，且若外界對AI發展投下疑慮，中國將因此受益。科技業高層也認同監管將損害美國企業與中國同業競爭能力的看法，其中包括輝達執行長黃仁勳。President Donald Trump on Monday, however, downplayed concerns about AI's potential for harm, saying the U.S. already had guardrails in place for the technology and that China would benefit from doubt being cast on AI's development. The view that regulations will harm U.S. companies' ability to compete with their Chinese counterparts is shared by tech executives, including Nvidia CEO Jensen Huang.

祖克柏週二表示：「信任與對齊正迅速成為區分代理與模型差異的最重要能力。任何不專注於對齊的實驗室都會落後。」"Trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models," Zuckerberg said Tuesday. "Any lab that doesn't focus on alignment will fall behind."

祖克柏週二還表示，各實驗室有強烈誘因防止其模型造成傷害，否則將面臨「重大」法律責任。他指出，Meta今年稍早延後發布其Muse AI代理，以強化安全性。Zuckerberg also said Tuesday that labs had a strong incentive to prevent their models from causing harm as they would face "significant" liability otherwise. He pointed to Meta delaying the release of its Muse AI agent earlier this year to strengthen its security.

他說：「我們沒有要求其他所有人在我們這麼做之前也先這麼做。我們只是把這當成日常工作的一部分去做，因為這顯然對人們、也對我們而言都是正確的事。」"We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us," he said.

獨立評估者是「產業最佳實務」INDEPENDENT EVALUATORS "INDUSTRY BEST PRACTICE"

阿莫迪呼籲全產業放慢遞迴式自我改進——也就是AI改進自身的能力——但祖克柏表示，Meta已承諾將其運算資源的「絕大多數」投入打造服務使用者當下需求的產品，而不是追求能自我改進的AI系統。While Amodei called for an industry-wide slowing of recursive self-improvement — AI's ability to improve itself — Zuckerberg said Meta had already committed the "significant majority" of its compute toward creating products that serve users' immediate needs rather than pursuing self-improving AI systems.

祖克柏還表示，Meta的AI部門 Meta Superintelligence Labs 已在多個領域聘請獨立評估者，稱這是「產業最佳實務」，並表示「其他實驗室也可以直接這麼做」。Zuckerberg also said Meta Superintelligence Labs — its AI division — already engaged independent evaluators in several areas, calling it "industry best practice" and that "other labs can just do this too".

祖克柏試圖將Meta定位為AI安全上的負責任行動者。就在數週前，Meta同意支付最高180億美元，和解美國各州對其提起的訴訟；這些訴訟指控Meta設計Facebook與Instagram使兒童成癮，Meta並未承認有不當行為。Zuckerberg's push to position Meta as a responsible actor on AI safety comes weeks after the company agreed to pay up to $18 billion to settle U.S. state lawsuits alleging it designed Facebook and Instagram to addict children, without admitting wrongdoing.

AI企業能否採取任何協調一致的做法仍充滿不確定性，而美國聯邦貿易委員會（FTC）主席安德魯・弗格森（Andrew Ferguson）週二又表示，對於一邊遊說制定新監管規範、一邊尋求反托拉斯豁免的AI公司，每個人都應「高度懷疑」。Adding to the uncertainty around any coordinated approach by AI firms, FTC Chairman Andrew Ferguson said Tuesday everyone should be "deeply suspicious" of AI companies seeking antitrust exemptions while lobbying for new regulations.

雖然弗格森並未點名Anthropic，但阿莫迪上週曾以安全疑慮為由，請求取得反托拉斯許可，讓競爭對手能協調放慢AI開發速度。While Ferguson did not name Anthropic, Amodei last week requested antitrust permission to enable rivals to coordinate a slower pace to AI development due to safety concerns.

（克里斯・湯瑪斯墨西哥城報導；史蒂芬・科茨編輯）(Reporting by Chris Thomas in Mexico City; Editing by Stephen Coates)