许多科学家认为，基于传统架构的现行人工智能模型无论算力多么强大，在通往意识的道路上都可能是死胡同，其核心限制在于缺乏生物活性物质及传统计算架构的固有缺陷。自1945年以来，现代计算机一直遵循冯·诺依曼架构，计算单元与存储单元相互分离，导致数据在两极之间高频搬运并产生巨大的能耗瓶颈；相比之下，生物大脑的神经元在同一物理位置同时完成信息处理与存储，典型人类大脑的运行功率仅约20瓦，而执行等量并行计算的现行人工神经网络则需要耗电数百万瓦的数据中心。

为了克服能效瓶颈并逼近生物智能机制，“神经形态”（Neuromorphic）计算利用忆阻器（memristors）等新型器件，在外加电压改变导电性后保留物理记忆，实现在同一物理节点上的存算一体化；这种架构不仅极大地提升了能效，还引发了心灵哲学家的浓厚兴趣，他们认为决定机器能否产生意识的关键在于计算实现方式（即机器“思考”的物理组织形式）而非单纯的算力堆叠。

在生物计算与前沿神经科技领域，澳大利亚初创企业Cortical Labs将由人类干细胞培育的数十万个活体神经元集成在硅基芯片上，成功训练这种“生物计算机”玩《乓》（Pong）和《毁灭战士》（Doom）等视频游戏；与此同时，实验室培育的三维脑类器官（neural organoids）规模已超过蜜蜂的神经元数量，尽管哲学家和神经科学家指出这些系统尚未具备产生意识的正确组织结构，但正如萨塞克斯大学神经科学家阿尼尔·塞斯（Anil Seth）所述，随着计算系统在生物材料与组织属性上越来越接近真实大脑，人工实体孕育意识的可能性将显著提升。 

Many scientists view current AI models as a dead end for machine consciousness regardless of raw compute power, pointing to their lack of biological substrates and structural design flaws. Since 1945, modern computing has relied on the von Neumann architecture, where continuous data transfer between separated processing and memory units incurs massive energy costs; conversely, biological brains process and store data within the same neurons, consuming a mere 20 watts compared to millions of watts required by data centres running artificial neural networks of equivalent parallel capacity.

Neuromorphic computing addresses this computational bottleneck by deploying memristors that alter conductance under voltage and retain that state as memory, achieving true in-memory processing; this architecture dramatically boosts electrical efficiency while captivating philosophers who argue that the physical method of computation, rather than computational prowess alone, constitutes the essential pathway to synthetic consciousness.

Pioneering biological computing, Australian startup Cortical Labs integrates hundreds of thousands of living stem-cell-derived human neurons onto silicon chips, successfully training them to play video games like Pong and Doom. Concurrently, 3D laboratory-grown neural organoids now surpass the neuron count of a honeybee; while neuroscientists and philosophers like Anil Seth and Peter Godfrey-Smith note that these clusters currently lack the precise structural organization required for conscious experience, engineering systems that increasingly replicate biological brain properties offers the most plausible path toward conscious synthetic entities.

Source: Could more brain-like chips provide a path to consciousness?

Subtitle: Some experts believe computers will need biological aspects to become self-aware

Dateline: 8月 20, 2026 03:30 上午