首页 / 财富中文网 / 正文

中国押注机器人技术与开源路线,力求在人工智能领域超越美国

财富中文网 2025-07-26 03:31:42

中国押注机器人技术与开源路线,力求在人工智能领域超越美国
在新加坡举办《财富》人工智能头脑风暴大会上,发言嘉宾们探讨了中美之间的AI竞争。图片来源:Graham Uden for Fortune

中国在AI领域的突破已引起全球关注。自今年1月的“深度求索(DeepSeek)冲击”以来,中国AI领域飞速发展,使中美之间的差距即使尚未消除,也已显著缩小。

然而,美国似乎专注于强大但专有的大语言模型、企业级AI以及半导体,而中国则采取了截然不同的路线发展本国的AI产业。

周三,Tech Buzz China的创始人马睿在新加坡召开的《财富》人工智能头脑风暴大会上表示:“机器人技术是当下中国重点发展的领域。”中国企业正齐心协力地将AI融入实体硬件,形成所谓的“具身智能”。即使美国拥有更强大的基础模型,中国在AI的工业应用领域已经取得了更大进展。

DGA-Albright Stonebridge Group的合伙人保罗·特里奥罗表示,另一个不同点是中国全力支持开源人工智能。“除少数例外情况外,中国企业已转向开源和权重模型。”

外界对中国AI模型(如深度求索和阿里巴巴的模型)的部分兴趣,源于其开源特性,这使得开发者可以在自己的设备上自由下载、使用和试验这些模型。

乔治城大学安全与新兴技术中心(Georgetown’s Center for Security and Emerging Technology)战略与基础研究资助主任海伦·托纳表示,通过免费公开源代码,AI公司正在助力构建“软实力”。

其他国家的AI开发者可以利用这些开源模型开发自己的产品,从而扩大中国的全球影响力。

但在美国,OpenAI、Anthropic和谷歌(Google)等引领AI行业的公司,仍然选择封闭源代码。就连曾凭借其Llama模型引领开源AI发展的Meta,据称也在考虑转变策略,专注于闭源开发。

托纳周三解释道:“如果你处于领先地位,那么你就拥有了这个极其宝贵的资产。”而处于领先地位的公司不愿意免费交出这一资产。

然而,托纳继续表示,对于月之暗面(Moonshot)或阿里巴巴等中国开发者而言,它们可能无法在最高级别展开激烈竞争,但仍提供高性能的模型,那么开源模式是一个“赢得大量信誉”的有效途径。

托纳表示,美国国内对于走开源还是闭源路线尚未达成共识,这场竞争仍是“胜负难料”。不过她承认,在开源领域的竞争中,“中国无疑占据优势”。(*)

译者:刘进龙

审校:汪皓

中国在AI领域的突破已引起全球关注。自今年1月的“深度求索(DeepSeek)冲击”以来,中国AI领域飞速发展,使中美之间的差距即使尚未消除,也已显著缩小。

然而,美国似乎专注于强大但专有的大语言模型、企业级AI以及半导体,而中国则采取了截然不同的路线发展本国的AI产业。

周三,Tech Buzz China的创始人马睿在新加坡召开的《财富》人工智能头脑风暴大会上表示:“机器人技术是当下中国重点发展的领域。”中国企业正齐心协力地将AI融入实体硬件,形成所谓的“具身智能”。即使美国拥有更强大的基础模型,中国在AI的工业应用领域已经取得了更大进展。

DGA-Albright Stonebridge Group的合伙人保罗·特里奥罗表示,另一个不同点是中国全力支持开源人工智能。“除少数例外情况外,中国企业已转向开源和权重模型。”

外界对中国AI模型(如深度求索和阿里巴巴的模型)的部分兴趣,源于其开源特性,这使得开发者可以在自己的设备上自由下载、使用和试验这些模型。

乔治城大学安全与新兴技术中心(Georgetown’s Center for Security and Emerging Technology)战略与基础研究资助主任海伦·托纳表示,通过免费公开源代码,AI公司正在助力构建“软实力”。

其他国家的AI开发者可以利用这些开源模型开发自己的产品,从而扩大中国的全球影响力。

但在美国,OpenAI、Anthropic和谷歌(Google)等引领AI行业的公司,仍然选择封闭源代码。就连曾凭借其Llama模型引领开源AI发展的Meta,据称也在考虑转变策略,专注于闭源开发。

托纳周三解释道:“如果你处于领先地位,那么你就拥有了这个极其宝贵的资产。”而处于领先地位的公司不愿意免费交出这一资产。

然而,托纳继续表示,对于月之暗面(Moonshot)或阿里巴巴等中国开发者而言,它们可能无法在最高级别展开激烈竞争,但仍提供高性能的模型,那么开源模式是一个“赢得大量信誉”的有效途径。

托纳表示,美国国内对于走开源还是闭源路线尚未达成共识,这场竞争仍是“胜负难料”。不过她承认,在开源领域的竞争中,“中国无疑占据优势”。(*)

译者:刘进龙

审校:汪皓

China’s AI breakthroughs have captured worldwide attention. Since January’s “DeepSeek shock,” the country’s AI development has accelerated, narrowing—if not eradicating—the gap with the U.S. when it comes to technological advancement.

Yet while the U.S. appears to focus on powerful yet proprietary large language models, enterprise AI, and semiconductors, China is taking a vastly different approach to cultivating its AI industry.

“Robotics is where it’s at in China today,” Rui Ma, founder of Tech Buzz China, said at the Fortune Brainstorm AI Singapore conference on Wednesday. Chinese companies have made a concerted push to integrate AI into physical hardware in what’s called “embodied AI.” Even if the U.S. has the more powerful foundational models, China has gone further in applying AI to industrial uses.

Another difference is China’s wholehearted embrace of open-source AI, noted Paul Triolo, a partner at the DGA-Albright Stonebridge Group. “Chinese companies, with a few exceptions, have moved to open-source and weight models.”

Some of the interest in China’s AI models, like those from DeepSeek and Alibaba, is the result of their open-source nature, which allows developers to freely download, use, and experiment with these models on their own devices.

By making their source code freely available, AI companies are helping build “soft power,” said Helen Toner, director of strategy and foundational research grants at Georgetown’s Center for Security and Emerging Technology.

AI developers in other countries can use these open-source models to develop their own products, spreading China’s global influence.

But in the U.S., companies leading the AI industry, like OpenAI, Anthropic, and Google, have continued to keep their source codes closed. And even Meta, which helped lead open-source AI development with its Llama model, is reportedly considering a shift in strategy to focus on closed-source development.

“If you are in the lead, then you have this incredibly valuable asset,” Toner explained on Wednesday. And leading companies don’t want to give away that asset for free.

Yet for Chinese developers like Moonshot or Alibaba—which may not quite compete at the highest levels but still offer high performance—then an open model is a good way to “buy a lot of goodwill,” she continued.

Toner said there isn’t a consensus in the U.S. about whether to go open- or closed-source, and that the battle is still “anyone’s game.” Yet she admitted that the fight for open-source is “certainly one where China has the edge.”

*