比尔·盖茨表示，人工智能在今年早些时候跨过了两个危险的门槛：它现在能够协助进行生物恐怖主义袭击，并能发动破坏力巨大的网络攻击。他在《埃兹拉·克莱因秀》（The Ezra Klein Show）上发表了这一观点，该采访于周二由《纽约时报》发表。
Bill Gates says AI crossed two danger thresholds early this year: it can now help a bioterrorist, and it can power a crippling cyberattack. He made the case on The Ezra Klein Show, in an interview published by The New York Times on Tuesday.
盖茨对克莱因说：“我们在今年早些时候跨过了网络门槛和生物门槛。”
“We crossed the cyber threshold and we crossed the bio threshold early this year,” Gates told Klein.
在网络方面，他指出有些模型能够发现人类已经检查了20年以上的代码中的安全漏洞。他提到了催生Anthropic公司Mythos的相关版本。在生物方面，他表示，设计新病原体的能力过去只属于主权国家。现在，使用最新人工智能工具的小型群体也具备了这种能力。“最危险的事物”克莱因向盖茨提出了这样一种观点，即产品责任和公司自身的利益足以保证人工智能的安全。
On cyber, he pointed to models that find security bugs in code humans had checked for more than 20 years. He named the releases that led up to Anthropic’s Mythos. On bio, he said the power to design a new pathogen used to belong only to nation states. Now small groups using the latest AI tools have it, he said. ‘The most dangerous thing’Klein put to Gates the view that product liability and companies’ own interests are enough to keep AI safe.
克莱因称这是华盛顿的主流观点。他说，这也是特朗普总统和顾问大卫·萨克斯（David Sacks）的观点，并在一定程度上是英伟达首席执行官黄仁勋的观点。
Klein said it was the governing view in Washington. It is the view of President Trump and adviser David Sacks, he said, and to a degree of Nvidia’s Jensen Huang.
盖茨说：“好吧，我几乎不敢相信你会问这个。这是人类有史以来接触过的最危险的事物。”
“Well, I almost can’t believe you’re asking that. This is the most dangerous thing that humans have ever gone near,” Gates said.
他将人工智能与药物、飞机和汽车进行了对比，这些领域都面临着安全法规的约束，而不仅仅是依靠诉讼。他补充说，任何人都可以拿走一个开源模型并关闭其监控。
He compared AI with drugs, aircraft and cars, which all face safety rules rather than lawsuits alone. He added that anyone can take an open-source model and switch off its monitoring.
他说：“你不能指望这个行业来进行自我监管。我的意思是，这简直太疯狂了。”
“You can’t rely on the industry to self-regulate here. I mean, it’s just insane,” he said.
一层安全防护网 盖茨呼吁在所有人工智能模型中建立一层监督性质的安全防护网和监控机制。其职责是阻止人们利用当今的系统来瘫痪经济或导致数百万人死亡。他说，开源模型可以保持免费且可定制。但它们必须在政府能够检查安全防护网是否依然存在的平台上运行。
A layer of safeguards Gates called for a supervisory layer of safeguards and monitoring in all AI models. Its job would be to stop people using today’s systems to shut down economies or kill millions. Open-source models could stay free and customisable, he said. But they would have to run on platforms where governments can check the safeguards are still there.
他还表示，Anthropic对Mythos和Fable设置的过滤网过高了。他说，询问癌症相关的用户最终会被转到其较小的模型上。
He also said Anthropic had set the filters on Mythos and Fable too high. Users asking about cancer end up pushed down to its smaller models, he said.
盖茨表示，人工智能迄今为止创造的就业岗位多于其摧毁的岗位。但在他看来，编程是唯一一个人工智能在可靠性方面明显超越人类的专业领域。他预计，会计、法律事务、电话销售和客户支持等领域将在几年内出现类似情况。
Jobs and a payroll tax for robots AI has so far created more jobs than it has destroyed, Gates said. But coding is the only profession where AI has clearly passed humans on reliability, in his view. He expects accounting, legal work, telesales and customer support to follow within a few years.
他希望企业对人工智能或机器人征收与对被其取代的人类员工相同的工资税。他还提出了“人类保留”岗位的概念，例如儿童保育和老年护理，这些岗位应专门保留给人类。
He wants companies to pay the same payroll tax on an AI or a robot as on the worker it replaces. He has also proposed “Human Reserved” jobs, such as child care and elder care, kept for people.
盖茨基金会将在十月的战略评审会上确定明年100亿美元的支出计划。他表示，今年ChatGPT、Claude和Copilot将参与部分评审会议。上周，盖茨在NBC的《会见记者》节目中警告称，人工智能可能导致十亿人死亡。
The Gates Foundation sets how it spends $10bn next year at strategy reviews in October. This year ChatGPT, Claude and Copilot will sit in on some of them, he said. Last week, Gates warned on NBC’s Meet the Press that AI could cause a billion deaths.