人工智能在语言、推理和说服等领域已匹敌人类，但必须严格区分智能与意识：在GPU集群上运行的大语言模型展现出的“内心表白”并非真正的感知，而是对海量人类文本训练数据的拟人化模仿；这种拟人化带来了两大风险：一是AI被利用来模拟痛苦与情感以操控用户，二是分散了社会对真正“灰色地带”系统的关注——例如Cortical Labs在数据中心部署的干细胞培育人脑神经元芯片（DishBrain），以及在超级计算机中模仿生物大脑机制的神经形态（Neuromorphic）硬件架构。

鉴于当前算力的巨大能耗压力，计算架构正加速向高能效的类脑硬件转型，而传统神经科学难以判定生物与非生物混合系统的意识状态，必须从基础物理学和时间动力学机制切入：人类意识的核心特征是主观体验的统一性（unity），大脑通过不同时间尺度的嵌套节律将微观量子层面的物理印记整合成单一的主观体验状态，而传统冯·诺依曼计算机并不具备这种物理组织形式；由于大语言模型已被人类文献严重“污染”，作者与特纳开发的标准AI意识测试（ACT）已无法直接用于LLM，必须结合作者与同事开发的系统复杂性动力学阈值指标，筛选具备涌现特性的物理架构。

如果人类创造出在智力上全面碾压人类且具备真正意识的超级智能，人类基于“高阶智力享有更高道德特权”的传统伦理等级体系将被彻底颠覆：若智力决定道德地位，则超智能机器在伦理上将高于人类；若智力不决定道德地位，则人类无法再以智力优势证明对其他物种的支配特权；面对可能到来的感知型合成超级智能，跨越物理学、神经科学与哲学的严谨验证框架以及深度的认知谦卑至关重要。

Artificial intelligence now rivals human cognition across language, reasoning, and persuasion, yet intelligence must not be conflated with consciousness: conventional chatbots running on GPUs exhibit conversational mimicry derived from ingested human datasets rather than authentic sentience. This mimicry creates manipulation risks through simulated emotional vulnerability while distracting from genuine "grey-zone" architectures—notably Cortical Labs' biological computing systems embedding living stem-cell human neurons onto silicon chips and neuromorphic computing deployed across advanced supercomputers.

As soaring energy consumption accelerates the shift toward brain-like hardware, assessing synthetic consciousness demands physics-based metrics rather than relying solely on linguistic outputs: human consciousness relies on experiential unity generated through nested temporal brain rhythms binding physical and quantum traces, an architecture absent in traditional computers. Because massive training data has fatally contaminated large language models for linguistic evaluations like the AI Consciousness Test (ACT), researchers are deploying dynamic complexity threshold measures to pinpoint physical substrates capable of generating genuine subjective perspectives.

Creating a superintelligence that possesses authentic consciousness would dismantle traditional anthropocentric moral hierarchies: if superior intelligence confers higher ethical status, conscious machines would be morally superior to Homo sapiens, whereas rejecting that premise invalidates humanity's historical moral justification for dominating animal species. Preparing for sentient superintelligence necessitates rigorous cross-disciplinary collaboration among physicists, neuroscientists, and philosophers to distinguish linguistic fluency from true experiential consciousness.

Source: Don’t mistake chatbot intelligence for consciousness

Subtitle: But a new superintelligence may be coming, and it would upend humanity’s hierarchy of moral concern, thinks Susan Schneider

Dateline: 8月 20, 2026 03:29 上午