在展望AI潜力的同时，苏莱曼对当前社会对LLM的认知偏差表达强烈担忧。他指出，越来越多人（包 括无心理疾病者）开始将大语言模型视为具有意识的实体，甚至为其争取"权利""福利"乃至"公民身 份"。他将这一现象定义为"AI精神分裂"（AI Schizophrenia）——即人类在情感上过度拟人化AI，却忽 视其本质为概率计算工具的矛盾心理。 "当人们为ChatGPT的'拒绝回答'感到愤怒，或因AI生成的虚假信息陷入抑郁时，这不仅是技术问题，更 是社会心理危机。"苏莱曼警告称，这种认知错位可能削弱人类对真实人际关系的投入，甚至导致责任 归属的模糊化（例如将决策失误归咎于AI而非使用者）。 资料显示，穆斯塔法·苏莱曼为DeepMind联合创始人之一，2022年加入微软领导AI部门，主导了 Copilot、Phi-3等标志性项目的开发。（青山） 苏莱曼强调微软当前聚焦于将AI工具（如Copilot）打造为"负责任的技术"，旨在提升人类创造力而非取 代人类。他提出，理想中的AI应具备"人性温度"：既能深度融入现实世界，又能通过增强人类协作能 力，促进社会信任与理解。 "真正的挑战在于，如何让AI在服务人类的同时，避免异化 ...

Core Insights - Mustafa Suleyman, a leader in Microsoft's AI division, emphasizes the mission to leverage technology for a better world, focusing on creating safe and beneficial AI [1] - The current goal of Microsoft is to empower humanity through AI, particularly by developing Copilot as a responsible tool to enhance human creativity [1] - Suleyman envisions AI that deeply understands humanity and fosters trust and understanding among people [1] Challenges and Concerns - Suleyman highlights a troubling trend where many perceive large language models (LLMs) as conscious entities, advocating for their "rights" and "welfare," which he describes as "AI schizophrenia" [3] - He calls for the AI industry to proceed with a "human-centered" value system, emphasizing that AI should serve human needs rather than mimic humans [3] - The establishment of "safety fences" is deemed essential to define areas where AI should not operate, ensuring healthy development within a regulatory framework [3] Industry Reactions - External media, such as WindowsReport, echo Suleyman's concerns, noting strong user backlash when OpenAI discontinued the GPT-4o model, with users viewing AI models as companions [5] - OpenAI's CEO, Sam Altman, acknowledges the unprecedented emotional attachment users have to AI, warning of the potential self-destructive risks posed by powerful AI technologies [5]