캘리포니아 AI 규정: 혁신과 안전의 균형

### 캘리포니아의 AI 규정: 혁신과 안전의 균형
* 🤖 캘리포니아의 AI 규정의 등장은 **유럽연합의 유사한 움직임**에 이어 인공지능에 대한 규제 없는 시대가 끝났음을 알리는 신호입니다.
* ⚖ 특히 소규모 기업과 오픈소스 모델의 경우, AI 기술이 혁신을 저해하지 않고 공익에 기여할 수 있도록 **안전장치와 규정**이 필요합니다.
* 💰 SB 1047 규정은 **투자 규모가 1억 달러가 넘는 대규모 모델 또는 중요 인프라에 대한 사이버 공격을 용이하게 할 수 있는 모델**을 대상으로 하는 투자 매개변수를 도입합니다.
* 🗣 법안 내용에 대한 학계와 업계 리더들의 우려가 있지만, **책임감 있는 AI 개발을 촉진하기 위해 민관 파트너십의 중요성**이 강조되고 있습니다.
* 🔍 타사 테스트와 '킬' 개념에 관한 주요 이슈가 남아 있어 **독점 기업이 규정 준수 노력에 참여하지 못하게 할 수 있습니다.**
* 🏗 이 규정은 **사람 중심의 검증 및 품질 관리 시스템**을 구축하여 잠재적으로 업계 내 안전장치를 개선하는 것을 목표로 합니다.
* 🌍 캘리포니아의 녹색 배출 기준과 유럽의 GDPR과 같은 역사적 선례는 **규제가 기술 부문에서 규정 준수와 책임감**을 어떻게 촉진할 수 있는지 잘 보여줍니다.
* ⏳ 전반적으로 캘리포니아의 AI 규제의 미래는 **필요한 안전장치와 지속적인 혁신의 필요성 사이의 균형**을 맞추는 프레임워크를 만들 준비가 되어 있습니다.

캘리포니아의 새로운 AI 규제는 혁신과 필요한 안전장치의 균형을 맞추면서 규제되지 않은 AI의 종말을 알리는 신호탄입니다.

What do you make of this regulation going through California? So I can be a force for good. But it does need some safety guards and some guardrails put in place. I think this is really a signal that the unregulated era of AI is over. I think that first step was with the EU Act that happened and was ratified overwhelmingly last year by the majority of the European Union. The worry for some in the market open air has been a voice for this when Wendy, is that the regulation coming through will harm in particular smaller companies and also open source large language models. Is that something that you agree with or is this the necessary step taken to ensure that AI remains for good? Well, it's a tricky regulation to put in place. There's always this balance of ensuring that we're not stifling innovation while trying to reduce harm and to create transparency in AI. Within the SB 1047 regulation itself. There are some parameters related to the

amount invested in the AI, so they're trying to find a little bit of a balance of if it is a large-scale model that is being that has, you know, over 100 million of investment or $10 million to deploy that, that is what would be subject to this as well as AI that could be used to enable cyber attacks to critical infrastructure such as energy, you know, nuclear, you know, chemicals, as an example. So it's about spending. We spoke to academics who said they were consulted but didn't like the content of the bill. We spoke to Senator Scott Wiener, who said he spoke to everyone he could and tried to get input from as many people as possible. Then you have matter that's against it. From the open-source perspective, why couldn't we find a bill that basically works for a broad coalition, given all the talk about it? It's true that it is somewhat incomplete.

I think one of the best things that can be done is that public-private partnership where, you know, companies like Meta, OpenAI, and others are really on the leading edge of developing this technology. And I think there were a few things that were basically incomplete in this. One was that there, you know, is the notion of kills, which is the other is this notion of having third parties test. Well, that becomes really difficult for a proprietary AI, right? You're opening yourself up to a third party, and that's a reason why companies may not want to participate. Then there's some other important items like sandboxes. So in AI, research and development is a critical part of this. If we don't have some content around how sandboxes can be leveraged to test out, you know, whether there are actual harms, I think it could become very challenging. So really, at the end of the day, what I think everybody's voicing some concern

on is we don't want to have any unintended consequences, whether it is IP stifling of innovation. At the same time, I think we could all agree, and most Californians probably do agree, that some safeguards for this pervasive technology are necessary. Wendy Sommer provides data labeling and evaluates models. How is the bill and its contents going to impact you in practice? Well, one of the most critical things about developing responsibly AI, is having some human-centric validation. So I believe that this will ultimately drive forward some additional measures for both human-centric validation quality management systems. So I think there could be an increase ultimately over time. But what we do see is that a lot of companies, including big tech, probably do a little bit more than maybe the public is aware of in terms of trying to provide these safeguards. Now it's really about putting forth a regulation that balances that and really creates some transparency.

So we've seen it happen before. Certainly, whether it was green emission standards set up by California or, you know, laws like GDPR in Europe, where they did set forth a wave of actions that ultimately companies had to comply with. So I wouldn't say that, you know, we're seeing it in droves right now. I do think over time, companies are going to need to think about this because, as I mentioned before, I think the unregulated era is coming to an end.

California Passes Controversial AI Safety Bill

Welcome to the Bloomberg Technology YouTube channel. Subscribe for original series and mini-docs on game-changing technologies, profiles on innovators, live conversations with experts, and breaking technology news & analysis from Bloomberg TV.

Bloomberg Technology

캘리포니아의 AI 규제: 균형 잡힌 법

캘리포니아의 AI 규제는 유럽연합의 압도적인 비준에 이어 인공지능에 대한 규제 없는 시대의 종말을 알리는 신호탄입니다.

이 규정은 중요 인프라에 대한 사이버 공격을 가능하게 할 가능성이 있는 대규모 AI 투자와 모델을 대상으로 하여 혁신과 피해 감소의 균형을 맞추는 것을 목표로 합니다.

그러나 이 법안은 불완전하고 소규모 기업과 오픈소스 대규모 언어 모델의 요구를 효과적으로 수용하지 못한다는 비판에 직면해 있습니다.

AI 개발에서 민관 파트너십의 역할

AI 기술 개발을 위해서는 Meta 및 OpenAI와 같은 기업과의 민관 파트너십이 매우 중요합니다.

주요 과제로는 독점 AI에 대한 제3자 테스트의 어려움과 잠재적 피해를 평가하기 위한 효과적인 샌드박스의 필요성 등이 있습니다.

목표는 업계 관행의 혁신과 투명성을 촉진하는 동시에 AI에 대한 안전장치를 마련하는 것입니다.

규제 표준이 기업에 미치는 영향

캘리포니아의 녹색 배출법이나 유럽의 GDPR과 같은 규제 표준은 역사적으로 기업들이 규정을 준수하도록 유도해 왔습니다.

규정 준수 조치가 갑자기 급증하지는 않겠지만, 이러한 추세는 시간이 지남에 따라 증가할 것으로 예상됩니다.

기업에 대한 규제가 없던 시대가 끝나가고 있으며, 이에 따라 기업들은 새로운 규제에 적응해야 할 필요성이 커지고 있습니다.