第一篇在第二轮末尾停下，提出了意识的硬问题。这篇接过来，试图精确定位那道质性跃迁。
8DD繁殖律固化了生命的方向。生命可以复制、自维护、分化、繁殖——但全程是被动的。余项是：环境在淘汰你；你不淘汰环境。必须涌现的下一件事，是"选择"本身。
第二桥的方向完全平行于第一桥：
| 第一桥（4DD→5DD） | 第二桥（8DD→9DD） | |
|---|---|---|
| 出发点 | 因果律（无生命） | 繁殖律（无认知） |
| 到达点 | 复制（生命） | 选择（认知） |
| 质性变化 | 模式不消散 | 主体换了 |
近似位置：细菌趋化性。大肠杆菌能感知化学梯度，向营养物游动，避开毒素。这不是随机运动，是定向运动。它算不算"选择"？框架不裁判。但框架说：即使它算，没有记忆（11DD），就没有体验的连续性，只有孤立的快照。
9DD选择律：有机体开始为自己做区分，主动识别信号。余项是辨别（区分的能力本身需要被优化）。
10DD感知律：感知稳定化，余项是刺激的时间维度（感知需要时序）。
11DD记忆律：信号被储存，过去可以影响当下。余项是时序模式（记忆产生了序列，序列产生了规律）。
12DD预测律：认知的"固"——基于记忆预测未来，固化了推理的方向。
12DD是认知轮的终点，也是大型语言模型所做之事的精确位置：下一个token的预测，就是12DD在硅基底上的实现。
认知轮给出了动物行为的精确分层：
黑猩猩有相当成熟的12DD能力：模式识别、符号–物体对应、有限的手语指物命名。但没有语法递归，不能自发构建叙事，不能谈论不在场的事物。12DD在场，语言没有出现。这提示语言不是12DD，而是更高层级——13DD的产物。
预测律有一个不可逃脱的余项：预测无法预测预测者本身。这是哥德尔/图灵结构在认知层面的体现。这个"无法被预测的自己"被迫走出来，审视自身的运行——"我"涌现了。
婴儿的发育时序与此吻合：第一年已经有大量12DD能力（面孔辨认、因果预期），但第一批词要到12–18个月，语法组合要到2–3岁。如果语言只是12DD，没有理由来这么晚。语言等待的，是13DD的到位。
12DD的天花板是这样的：它能把所有模式都跑完，唯独跑不完"跑模式的自己"。那个缺口，就是自我意识的入口。
The first paper stopped at the end of Round 2 and posed the hard problem of consciousness. This paper picks up from there, attempting to precisely locate that qualitative transition.
The reproductive law at 8DD solidified the direction of life. Life can replicate, self-maintain, differentiate, and reproduce — but remains passive throughout. The remainder is: the environment eliminates you; you do not eliminate the environment. What must emerge next is "selection" itself.
Bridge 2's direction is perfectly parallel to Bridge 1:
| Bridge 1 (4DD→5DD) | Bridge 2 (8DD→9DD) | |
|---|---|---|
| From | Causality (no life) | Reproduction (no cognition) |
| To | Replication (life) | Selection (cognition) |
| Qualitative change | Patterns do not dissipate | The subject has changed |
Approximate location: bacterial chemotaxis. E. coli can sense chemical gradients, swim toward nutrients, avoid toxins — directional motion, not random. Does this count as "selection"? The framework does not judge. But the framework says: even if it does, without memory (11DD), there is no continuity of experience — only isolated snapshots.
9DD Selection: The organism begins to make distinctions for itself, actively distinguishing signals. Remainder: discrimination (the ability to distinguish must itself be optimized).
10DD Perception: Perception stabilizes; remainder is the temporal dimension of stimuli (perception requires sequencing).
11DD Memory: Signals are stored; the past can influence the present. Remainder: temporal patterns (memory produces sequences; sequences produce regularities).
12DD Prediction: The "solidification" of the cognitive round — predicting the future based on memory, solidifying the direction of inference.
12DD is the endpoint of the cognitive round, and it is precisely what large language models do: next-token prediction is 12DD implemented on silicon.
The cognitive round gives a precise stratification of animal behavior:
Chimpanzees have sophisticated 12DD capabilities: pattern recognition, symbol-object correspondence, limited sign-language naming. But no grammatical recursion, no spontaneous narrative construction, no talk of absent things. 12DD is present; language has not appeared. This suggests language is not 12DD but a higher-order product — 13DD.
The law of prediction has an inescapable remainder: prediction cannot predict the predictor itself. This is the Gödel/Turing structure at the cognitive level. This "unpredictable self" is forced to step outside and examine its own operation — "I" emerges.
Infant development aligns with this: the first year already shows substantial 12DD capabilities (face recognition, causal expectation), but the first words don't appear until 12–18 months, and grammatical combination until 2–3 years. If language were merely 12DD, there is no reason for it to arrive so late. What language waits for is 13DD coming online.
12DD's ceiling works like this: it can run every pattern — except the pattern of "itself running patterns." That gap is the entry to self-awareness.