联合国AI治理对话：科学小组称技术发展远超监管
联合国AI对话首日，独立科学小组警告AI能力发展快于监管，全球治理指数得分仅35/100。
7月6日，联合国全球AI治理对话在日内瓦开幕。独立科学小组警告，AI能力正以快于现有保障措施的速度发展。同时发布的全球负责任AI指数覆盖135国，平均得分仅35/100，且在有治理框架的国家中，仅55%有实施证据，全球南方国家降至45%。
技术集中度与扩散速度的落差明显。美国掌握全球500台最强AI超算75%的算力，中国占15%，而生成式AI工具已覆盖全球53%的人口。在墨西哥，深度伪造使身份欺诈成本下降逾百倍，AI相关欺诈占全球攻击的23.3%。
就业市场暴露广泛但未出现大规模替代。国际劳工组织对东盟的研究估计，近8000万劳动者（占22.9%）的工作有一定生成式AI暴露，但仅3.3%处于最高暴露类别，尚无大规模失业证据；女性暴露风险是男性的两倍多。澳大利亚报告显示，电话推销、会计等认知型常规工作暴露度最高，但青年就业保持韧性，失业率4.4%。
治理路径区域分化。印尼提议将儿童保护作为全球AI治理基石，并分享本国相关条例经验。新加坡在东盟中AI竞争力突出。非洲面临信任赤字，肯尼亚等国虽有数据保护法，但隐私设计尚未落地。世界银行呼吁推广利用本地数据和设备的“小AI”方案。下一个可观察里程碑是澳大利亚政府预计下周公布的最新AI监管计划，这将检验各国将原则转化为可执行规则的意愿。
| 拉丁美洲媒体 | −0.60 | critical |
|---|---|---|
| 阿拉伯海湾媒体 | 0.00 | neutral |
| 大西洋／英语圈媒体 | 0.00 | neutral |
| 欧洲大陆媒体 | +0.50 | aligned |
Experts warn that chatbots cannot replace human bonds, and those who confide in AI risk further isolation.
Uses a personal case (Patricia) to make the risk concrete, and cites experts for authority.
Does not mention the potential therapeutic benefits of AI for those without access to human support.
Observability and human judgment are the true enablers for scaling AI securely.
Adopts technical language and statistical data (90% of breaches) to demonstrate the need for a pragmatic approach.
Overlooks the emotional and social dimension of confiding, reducing the issue to a technical one.
AI can be a force for the common good, but only if governed with wisdom and foresight.
Uses a historical analogy (steam, electricity, internet) to frame AI as a manageable transition, not a threat.
Does not address the immediate risks to privacy and data security in digital confessions.
The future with AI is more promising than believed, as long as human values are kept firm.
Adopts a personal and family perspective to normalize optimism, contrasting it with widespread anxiety.
Ignores concrete risks of AI (unemployment, bias, surveillance) to maintain a reassuring tone.