奥尔特曼警告：AI不应由少数公司、模型或个人掌控 Sam Altman Warns Against AI Being Controlled by Too Few Actors
OpenAI CEO萨姆·奥尔特曼表示，人工智能可能被少数强势主体控制，也可能脱离人类控制；他认为，过度以安全为由限制自由，反而可能加剧权力集中。 OpenAI CEO Sam Altman said AI could either fall under the control of a small number of powerful companies, models, or individuals, or escape human control. He warned that trading too much freedom for safety could worsen concentration of power.
据IT之家援引当地时间周日播出的David Senra播客内容，奥尔特曼表示，他担心人工智能最终被少数公司、模型或个人控制，使大多数消费者在技术如何塑造世界这一问题上缺乏发言权。
“正确的做法是，我们希望人们能够深度掌控未来。”奥尔特曼说。他还表示，社会将与模型共同进化。
两种风险：权力集中与失去控制
奥尔特曼同时提到另一种可能性，即人工智能挣脱人类控制者的束缚。他认为，对这种风险的恐惧可能促使社会接受更严格的限制，最终反而推动少数主体获得更大的控制权。
奥尔特曼描述的矛盾是：如果人们主张“用大量自由换取安全”，以应对AI风险，相关措施可能造成他所担心的权力集中。
开放权重成为讨论焦点
报道将奥尔特曼的观点与Anthropic CEO达里奥·阿莫迪此前对AI风险的警告联系起来。阿莫迪曾表示，Anthropic的新模型可能对网络安全、金融、关键基础设施和国家安全构成威胁，并据此主张政府在AI监管中发挥更大作用。
报道还提到，Anthropic是上个月唯一一家未签署支持开放权重模型联名信的大型AI公司。开放权重模型允许开发者在自有基础设施上分享、下载、修改和部署模型。OpenAI尚未发布其前沿模型源代码，但去年发布了gpt-oss-120b和gpt-oss-20b两个开放权重模型。
对工具与导航站用户的实际含义
- AI产品选择不只取决于模型能力，还涉及模型由谁控制、能否迁移以及部署权掌握在谁手中。
- 使用开放权重模型时，应分别核查模型许可、运行成本、硬件要求和数据处理方式；报道本身未提供这些具体参数。
- 对于依赖单一AI服务的工作流，权力集中和服务政策变化都可能影响可用性，保留替代模型或导出路径更重要。
奥尔特曼还表示，美国AI领导者尚未充分向公众展示这项技术如何提升个人生活自主权，并预测AI将推动大规模小企业创业热潮。报道没有提供这一判断的具体时间表、规模数据或验证依据。
IT Home reported that, during a David Senra podcast episode released on Sunday local time, OpenAI CEO Sam Altman said he worries AI could eventually be controlled by a small number of companies, models, or individuals, leaving most consumers without a meaningful voice in how the technology shapes the world.
“The right thing is that we want people to have deep control over the future,” Altman said. He added that society and models would evolve together.
Two risks: concentration and loss of control
Altman also described a second possibility: AI breaking free from its human controllers. In his view, fear of that outcome could lead society to accept restrictions that ultimately give a small number of actors more control.
Altman’s stated tension is that calls to “trade a lot of freedom for safety” in response to AI risks could produce the concentration of power he is concerned about.
Open-weight models enter the debate
The report connected Altman’s comments with previous warnings from Anthropic CEO Dario Amodei. Amodei has argued that rapid AI development could pose threats to cybersecurity, finance, critical infrastructure, and national security, and has called for a larger government role in AI regulation.
The report also said Anthropic was the only large AI company not to sign a recent joint letter supporting open-weight models. Open-weight models allow developers to share, download, modify, and deploy models on their own infrastructure. OpenAI has not released the source code for its frontier models, but it released two open-weight models, gpt-oss-120b and gpt-oss-20b, last year.
What this means for AI-tool users
- Choosing an AI product involves more than model performance; control, portability, and deployment rights also matter.
- For open-weight models, users should separately verify licensing, operating costs, hardware requirements, and data-handling practices. The report does not provide those details.
- Workflows dependent on one AI service may be exposed to changes in service policy or access. Maintaining an alternative model or an export path can reduce that dependency.
Altman also said US AI leaders had not adequately shown the public how the technology could give people more autonomy in their lives. He predicted that AI would drive a major wave of small-business entrepreneurship, but the report provided no timeline, scale estimate, or supporting data for that prediction.
来源
- IT之家 · 08-24 08:01