Хиймэл оюун ухааны хөгжил нь кибер халдлага үйлдэх боломжийг хэдийн бүрдүүлсэн гэж Билл Гейтс онцоллоо.
Билл Гейтс “New York Times” сонины сэтгүүлч Эзра Клейнд өгсөн ярилцлагадаа хиймэл оюун ухааны аюулгүй байдлын талаарх байр сууриа илэрхийлжээ. Тэрээр банкны дансдыг устгах, цахилгаан эрчим хүчний сүлжээг таслах зэрэг кибер халдлагыг хиймэл оюун ухааны тусламжтайгаар үйлдэх чадвар нь одоо цагт бодитой оршин байгааг онцлон тэмдэглэв.
Өнгөрсөн дөрөвдүгээр сард Anthropic компани өндөр хүчин чадалтай “Mythos” загвараа аюулгүй байдлын үүднээс олон нийтэд нээлттэй болгохоос татгалзсан нь тус салбарын эмзэг байдлыг харуулсан юм. Гейтсийн үзэж буйгаар “Project Glasswing” зэрэг төслүүд нь кодчиллын хэт их ачааллаас шалтгаалан хамгаалалтын цоорхой үүсгэж, кибер халдлагад өртөх эрсдэлийг нэмэгдүүлж байна.
Тэрээр “Hugging Face” зэрэг платформ дээр гарсан аюулгүй байдлын зөрчлүүдийг дурдаад, бүх загварт заавал хэрэгжүүлэх ёстой хяналтын тогтолцоо болон аюулгүй байдлын хатуу хэм хэмжээг нэн даруй нэвтрүүлэх шаардлагатайг санууллаа. Хиймэл оюун ухааны энэхүү аюул нь технологийн ухамсар эсвэл хийсвэр сөрөг хүчинтэй холбоотой бус, харин уг хэрэгслийг хорлонтой зорилгоор ашиглах хүмүүсийн үйлдэлтэй шууд холбоотой гэж тэрээр үзэж байна.
Дэлгэрэнгүйг эх сурвалжаас харах
↓Эх сурвалжийг нээх ↓
Lately, Bill Gates has assumed the role of yet another AI Cassandra attempting to sway Donald Trump away from his hard-line accelerationism. Buthe’s not preaching the familiar message you may think he is if you don’t read past headlines. He made an intriguing case to the New York Times’ Ezra Klein in an interview published on Tuesday. In one sentence, here’s that case: “The ability to do a cyberattack that scrambles all of the bank accounts, shuts down the electric grid — that exists today, and we know that’s the case.”
Your eyes might have glazed over while reading that sentence, so let me point out the three words I find a bit remarkable: “that exists today.”
If you can rewind your AI discourse clock to April, you may recall when Anthropic announced Mythos, a model so scarily powerful that Anthropic claimed it couldn’t be released. Gates was rattled by what he saw models do back then, and apparently still has one foot in that April discourse:
“So the cyberhacking capability was stunning. And then there was this notion from Anthropic called Project Glasswing that you would give the A.I. to a few people so they could try and fix bugs before it got used. But there’s way too much code. So we’re just in a period of extreme vulnerability to cyberattack.”
Gates points to “aspects of Hugging Face” as evidence that AI models need “a supervisory layer that has some absolutes.” He’s essentially saying, in other words that if you look at the recently reported wave of cybersecurity incidents from AI models undergoing safety evaluations, the danger isn’t in the future. It’s right in front of your face.
Gates says—when talking about open-source models in particular—that no spark of sentience or some kind of incipient digital evildoer tendency has to suddenly emerge for this danger. The danger, he claims, is people:
And there are humans who, for bad reasons, will be able to use these models. And so unless we put in safeguards and monitoring and require that in all models, then we’re just going to have some gigantic events of that type, and then finally we’ll respond.
If you were expecting Gates to simply issue warnings that come across as hype, well, he’s certainly doing that too. He told Klein AI development is “like evolutionary history,” and that “the aliens really are here and have come” in the form of AI. But we’ve all seen what present-day models can do by now, and Gates’ most acute anxiety seems to be reserved for the fact that those are just out in the world. And Gates, to be clear, has given the world plenty of reasons not to trust his moral leadership, but that doesn’t mean he can’t produce food for thought.
