白宫 AI 政策：与迈克尔·克拉齐奥斯的对话

AI Policy at the White House: A Conversation with Michael Kratsios

本期速览 · Overview

迈克尔·克拉齐奥斯讨论他在塑造美国 AI 政策中的角色，以及他选择公共服务的原因。

Michael Kratsios discusses his role in shaping US AI policy and why he chose public service.

美国必须在开源和闭源 AI 之间取得平衡以保持全球领先，并建立统一的国家监管框架。
The US must balance open and closed source AI to lead globally, with a single national regulatory framework.
AI 监管应避免设定硬性红线，因为它们很快就会过时；应关注输出而非训练。
AI regulation should avoid firm red lines as they quickly become outdated; focus on outputs not training.
继 AI 之后，量子计算是华盛顿下一个前沿技术优先事项。
Quantum computing is the next frontier tech priority for Washington, following AI.

美国必须维持一个充满活力的开源和闭源AI生态系统，才能在全球保持领先。
The US must maintain a vibrant open and closed source AI ecosystem to lead globally.
清晰的监管和统一的全国性AI框架对于支持初创企业至关重要。
Regulatory clarity and a single national framework for AI are essential to support startups.
政府应避免在AI监管中设定过于僵化的红线，因为它们很快就会过时。
The government should avoid setting firm red lines in AI regulation because they quickly become outdated.
AI将改变科学发现的方式，政府必须相应调整其研发资助策略。
AI will transform scientific discovery, and the government must adapt its R&D funding accordingly.
量子计算是华盛顿将优先考虑的下一个前沿技术。
Quantum computing is the next frontier technology that Washington will prioritize.
在政府工作提供了对国家产生重大影响的独特机会。
Working in government offers a unique opportunity to have a large impact on the country.

欧盟AI法案在ChatGPT出现之前就已制定，如今已经过时，这显示了过早监管的风险。
The EU AI Act was written before ChatGPT and is already outdated, showing the danger of early regulation.
AI带来的生物风险被夸大了；尽管有警告，但三年来并未成为现实。
Biological risk from AI is overblown; it hasn't materialized in three years despite warnings.
政府不应直接指导发现，而应塑造创新的环境。
The government should not direct discovery but shape the arena for innovation.
优先于州AI法律是支持初创企业的政策，因为大科技可以应对零散的法规，而初创企业不能。
Preemption of state AI laws is a pro-startup policy because big tech can handle patchwork regulations, startups cannot.
美国政府在研发资助中的角色已经反转；私营部门现在占主导，需要新的方法。
The US government's role in R&D funding has inverted; private sector now dominates, requiring a new approach.
AI监管应关注模型输出，如防止输出米老鼠，而不是过度监管训练过程。
AI regulation should focus on model outputs, like preventing Mickey Mouse outputs, rather than over-regulating training.