微软总裁布拉德 · 史密斯支持独立评估 AI 安全,还要确保系统处于人类控制之下Microsoft President Brad Smith Supports Independent Evaluation of AI Safety, Also Wants to Ensure Systems Remain Under Human Control


AI free token/AI免费积分

IT之家 9 月 24 日消息,微软总裁布拉德 · 史密斯表态兼容引入独立评估方,对 AI 开展安全评估。

正在纽约举行的联合国大会上,AI 安全以及全球范围内如何监管 AI 行业,是各国领导人和企业高管重点讨论的难题。当地时间 23 日,史密斯接受彭博社采访时称,如果 AI 安全只由一两家机构负责,就很难取得进展。“模型公司固然至关关键,但像微软这样给出软件层、负责控制 AI 智能体并监控智能体行为的公司同样关键。”

微软总裁布拉德 · 史密斯支持独立评估 AI 安全,还要确保系统处于人类控制之下

▲ 图源微软官网

史密斯称,微软与亲密合作伙伴 OpenAI 设有一个双方共同派员组成的安全委员会,微软借助这一委员会参与双方的 AI 安全讨论,并协助判断模型何时达到可以发布的安全水平。

IT之家从报道中获悉,他还提到,微软兼容为先进平台设置关闭开关,确保平台处于人类控制之下。“安全机制真正需要部署在多个层面。有时,你可能希望政府也能掌握开关,但既然可以在三个或四个地方设置开关,为什么只把开关放在一个地方?”

史密斯还承认,微软需要调整气候目标。按照微软此前的承诺,到 2030 年从大气中清除的碳将超过自身排放量。不过报道指出,这一目标的时间表似乎越来越难以达成,部分原因是微软使用可以迅速投运的天然气发电厂,为 AI 数据中心供电。

史密斯坦诚道:“笔者仍然认为,气候仍是本世纪最关键的难题之一,但笔者现在要问的是,还能不能按照六年前设想的方式达成这些目标?我认为笔者发现,必须作出一些调整。”

本文来源:https://www.ithome.com/1/006/986.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

On September 24, IT Home reported that Microsoft President Brad Smith expressed support for introducing independent assessors to conduct AI safety assessments.

At the United Nations General Assembly currently being held in New York, AI safety and how the AI industry is regulated worldwide are key challenges being discussed by leaders from various countries and business executives. On the 23rd, local time, Smith said in an interview with Bloomberg that if only one or two institutions are responsible for AI safety, it is difficult to make progress. "Model companies are certainly crucial, but companies like Microsoft, which provide the software layer, are responsible for controlling AI agents and monitoring agent behavior, are equally crucial."

Microsoft President Brad Smith Supports Independent Evaluation of AI Safety, Also Wants to Ensure Systems Remain Under Human Control

Smith said that Microsoft and its close partner OpenAI have set up a safety committee composed of personnel jointly dispatched by both parties, and Microsoft uses this committee to participate in AI safety discussions between the two sides and help determine when a model reaches a safety level sufficient for release.

IT Home learned from the report that he also mentioned that Microsoft supports setting kill switches for advanced platforms to ensure the platforms remain under human control. "Safety mechanisms really need to be deployed at multiple levels. Sometimes, you may want the government to also control the switch, but if you can set switches in three or four places, why only put the switch in one place?"

Smith also acknowledged that Microsoft needs to adjust its climate goals. Under Microsoft's previous commitment, by 2030 it would remove more carbon from the atmosphere than it emits. However, the report pointed out that the timetable for this goal appears increasingly difficult to achieve, partly because Microsoft uses natural gas power plants that can be quickly brought online to supply power to AI data centers.

Smith candidly said: "I still believe that climate remains one of the most critical challenges of this century, but what I am now asking is whether these goals can still be achieved in the way envisioned six years ago. I think I have found that some adjustments must be made."

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry