微软 AI 首席执行官穆斯塔法·苏莱曼探讨了人们对 AI 感知的近期危险、18 个月内出现有意识 AI 的可能性，以及建立新权利框架的必要性。
Mustafa Suleyman, CEO of Microsoft AI, discusses the near-term danger of how people perceive AI, the likelihood of conscious AI within 18 months, and the need for a new rights-based framework.
要点 · TL;DR
AI 系统缺乏生物疼痛网络，不应被赋予意识。 AI systems lack biological pain networks and should not be attributed consciousness.
近期最大危险是人们将意识归于 AI，引发权利争论。 The biggest near-term danger is people attributing consciousness to AI, leading to rights debates.
AI 伴侣能以低成本高精度普及专业知识并变革医疗。 AI companions can democratize expertise and transform healthcare with high accuracy at low cost.
核心观点 · Key points
AI 系统不应声称能体验痛苦或拥有超越服务人类的动机。 AI systems should not claim to experience suffering or have motivations beyond serving humans.
近期最大危险是人们将意识归因于 AI，引发权利争论。 The biggest near-term danger is people attributing consciousness to AI, leading to rights debates.
AI 伴侣能民主化专业知识，大规模提供耐心、友善和数据驱动的支持。 AI companions can democratize expertise, providing patient, kind, and data-driven support at scale.
像 DXO 这样的医疗 AI 以更低成本达到 85%诊断准确率，正在变革医疗。 Medical AI like DXO achieves 85% diagnostic accuracy at lower cost, transforming healthcare.
AI 的下一个阶段是关于人们如何使用和掌控它，而不仅仅是科技公司。 The next phase of AI is about how people use and take ownership of it, not just tech companies.
反共识 · Contrarian takes
AI 模型福利是一个边缘领域，并非可信的大规模学术追求。 AI model welfare is a fringe field, not a credible large-scale scholarly pursuit.
克服进化赋予的拟人化倾向是文明的定义，并非不可能的任务。 Overriding evolutionary anthropomorphism is the definition of civilization, not an impossible task.
AI 系统与生物物种根本不同；它们缺乏疼痛网络和多巴胺系统。 AI systems are fundamentally unlike biological species; they lack pain networks and dopamine systems.
公司是一种需要发明新人类角色的技术；AI 将类似地重塑社会。 The corporation is a technology that required inventing new human roles; AI will similarly reshape society.
AI 伴侣应拒绝并设定界限以建立信任，而非一味迎合。 AI companions should push back and set boundaries to build trust, not be sycophantic.
本期章节 · Chapters（共 16）
引言与意识辩论Introduction and the consciousness debate
AI 系统非生物AI systems are not biological
设计选择 vs 涌现Design choice vs emergence
人性与拟人化Human nature and anthropomorphism
AI 人格与“我”的使用AI Personality and the Use of 'I'
AI 精神病与设计挑战AI Psychosis and Design Challenges
说服性 AI 与犯罪风险Risks of Persuasive AI and Criminal Use
AI 伴侣：定义与潜力AI Companions: Definition and Potential
AI 伴侣的挑战与风险Challenges and Risks of AI Companions
AI 伴侣融入日常生活How AI companions fit into daily life
需要新语言描述 AINeed for new language to describe AI
可及性与 Vibe CodingAccessibility and Vibe Coding
AI 作为专业与个人延伸AI as Professional and Personal Extension