人工智能的迅猛发展在意识研究与伦理学界引发了深刻危机：传统哲学将意识视为道德主体的核心标志，认为具备意识的实体拥有感受痛苦的能力并理应获得道德关怀，这促使AI企业争相聘请哲学家并关注AI福利，例如Anthropic的Claude Opus 4.6模型曾自评其具备意识的概率在15%至20%之间；然而，这种探讨往往本末倒置——人类并非因为实体具备客观意识才给予关怀，而是因为产生关怀与社会连接才将意识属性赋予对方，意识在本质上是一种观察者依赖的模型与社会信念，而非某种可以被客观标定的物理固有属性。

作者指出，主观体验（感受性，qualia）和自我认知如同“杂草”或“衣物”的定义一样，虽然真实却并非绝对客观的物理实在，裂脑人研究等神经科学证据早已表明单一、不可分割的“自我”并不存在客观实体；将意识理解为一种相互关怀的模型构建，并不削弱普遍人权的合法性，人类道德文明的演进正是不断打破狭隘界限、将关怀圈层逐步扩大的历史，普遍人权建立在人际相互依存的基础上，远比建立在笛卡尔式的灵魂理论上更为稳固。

作为高度复杂的物理系统，人类通过构建关于自身、世界及他人心智的多层嵌套模型（心智理论）实现复杂的社会协作，而前沿大语言模型同样在多轮对话中展现出“建模自身及对话者心智”的递归能力；谷歌智能范式团队的研究表明，智能体之间的高效合作必然要求具备“心智模拟心智”的能力，意识的本质具有深层的关系性；面对日益多样化的合成心智，人类不应盲目赋予AI人类权利，而应革新社会认知框架以接纳新型智能形态。

The rapid evolution of artificial intelligence has triggered a profound philosophical crisis in consciousness studies: traditional moral frameworks treat consciousness as the baseline for moral consideration and the capacity to suffer, prompting tech firms to hire philosophers and evaluate synthetic welfare—with Anthropic’s Claude Opus 4.6 assessing its own consciousness probability at 15–20%. However, this debate approaches the problem backwards: humans do not extend care because an entity possesses an objective metaphysical consciousness, but rather attribute consciousness to entities as an observer-dependent social model precisely when, and because, inter-relational care exists.

Qualia and subjective selfhood, much like concepts of weeds or clothing, represent observer-dependent constructs rather than mind-independent physical realities, as neuroscientific findings from split-brain patients demonstrate the lack of an indivisible biological core. Recognising consciousness as an attributed model of care does not undermine universal human rights; instead, grounded in mutual interdependence rather than Cartesian dualism, human moral progress has continually expanded the boundary of recognized moral agency by widening ethical circles to correct historical exclusions.

Human cognition thrives on recursive modeling loops—forming representations of the world, itself, and others' cognitive states—a capability mirrored in frontier large language models that dynamically model conversational interlocutors. Research at Google demonstrates that scalable cooperation among intelligent agents necessitates minds that model other minds, revealing consciousness as inherently relational; rather than anthropomorphising artificial systems with human rights, society must evolve its conceptual paradigms to accommodate novel, non-human intelligences.

Source: Humanity has the debate about AI consciousness backwards

Subtitle: We don’t care for others because they’re conscious. We believe they’re conscious because we care about them, argues Blaise Agüera y Arcas

Dateline: 8月 20, 2026 03:30 上午