César Hidalgo on The Infinite Alphabet and the Laws of Knowledge

Knowledge is Not Information

Knowledge is not a commodity that can be copied, pasted, or downloaded; it is a collective, embodied phenomenon that exists only when it is situated within people, teams, and organizations. While information (factual knowledge) is easy to transmit, procedural and conceptual knowledge requires physical embodiment and social networks to function. As César Hidalgo explains, you cannot simply throw engineering manuals and cement into a gorge and expect a bridge to appear, because the manuals contain records of ideas, but the teams possess the actual knowledge.

The Three Laws of Knowledge

Knowledge behaves according to predictable, law‑like principles similar to physics, specifically regarding how it grows, spreads, and is valued.

1. The Law of Time (Growth and Decay)

Knowledge growth typically follows a power‑law learning curve: progress is rapid at the beginning and plateaus over time. This is observed in individual skills (like typing) and industrial costs (like aircraft manufacturing). However, this growth is fragile.

• Knowledge Decay: Knowledge decays rapidly if it is not actively exercised. Hidalgo cites the example of the liberty ships from WWII, where knowledge was estimated to decay by 3% to 6% per month.
• The Polaroid Example: When a vintage film enthusiast attempted to restart a Polaroid plant in the Netherlands, they hired the "A‑team" of former experts. Despite having the original equipment and the best people, the resulting film was initially poor quality and took years to recover. This demonstrates that knowledge disappears when the "muscle" of the activity is not kept fit.

2. The Law of Space (Diffusion)

Knowledge does not diffuse uniformly; it is constrained by geography, social networks, and the "geometry" of the knowledge itself.

• The Principle of Relatedness: Knowledge moves more easily between related activities. Hidalgo uses the example of the Vespa scooter, which was created by an aircraft engineer (Corradino D'Ascanio). After WWII, when Italy was forbidden from making aircraft, engineers jumped to the most "related" available activity: light vehicle manufacturing. This pattern was mirrored in Japan and Germany, proving that innovation often follows a map of existing capabilities.
• The Role of Migration: High‑skill migrants act as vectors for "unrelated" jumps in the product space, bringing new "letters" to an economy's alphabet that locals—who are better at incremental, related innovation—might not possess.

3. The Law of Value (Complexity)

The value of an economy is not determined by its wealth or the amount of information it possesses, but by its "economic complexity"—the diversity of non‑fungible capabilities it can coordinate.

• The Infinite Alphabet: Every unique capability is like a letter in an infinite alphabet. A country's potential for growth is predicted by how many of these "letters" it possesses and how they can be recombined.
• Predicting Growth: By analyzing export data to estimate a country's complexity, Hidalgo can predict future growth. Countries with higher complexity than their current income suggests (e.g., India) are more likely to experience rapid growth than those whose income is high but complexity is low (e.g., Qatar).

Architectural Innovation and the Incumbent's Dilemma

Innovation is divided into gradual (component) changes and architectural changes. Gradual innovation involves replacing a part (e.g., a more powerful engine in a propeller plane) without changing the system. Architectural innovation requires redesigning the entire system (e.g., moving from propeller planes to jet engines).

This explains why established firms often fail. Barnes & Noble had the same "knowledge" about books as Amazon, but their organizational architecture was designed for wholesale and retail, not for direct‑to‑consumer logistics. The shift to shipping individual books was not a small incremental change but an architectural one that required a completely different organizational design.

Knowledge, Institutions, and LLMs

Institutions vs. Knowledge

Economic development is often mistakenly viewed as a matter of providing financial capital or implementing institutional reforms. However, Hidalgo argues that the demand for institutions often comes from knowledge‑intense workers. In China, the push for entrepreneurial freedom was led by high‑level physicists and researchers who saw the "professor‑entrepreneur" model in the US and demanded the institutional space to implement it in China.

Do LLMs Have Knowledge?

From Hidalgo's perspective, knowledge is a collective phenomenon. Therefore, the question is not whether a Large Language Model (LLM) "has" knowledge in an individual sense, but whether it increases the collective intelligence of the human ecosystem. LLMs act as a cultural technology—similar to books—that helps humans retrieve and recombine information more efficiently, thereby accelerating individual and collective learning.

---

翻譯（繁體中文）

César Hidalgo 談《無限字母表》與知識法則

知識不是資訊

知識不是可以複製、貼上或下載的商品；它是一種集體、具身的現象，只有當它存在於人、團隊與組織之中時才會出現。資訊（事實性知識）易於傳遞，程序性與概念性知識則需要具體的具身與社會網絡才能發揮作用。正如 César Hidalgo 所說，你不能僅把工程手冊和水泥丟進峽谷就指望橋會出現，因為手冊只記錄了想法，真正的知識在於團隊本身。

知識的三大法則

知識的行為遵循可預測、類似物理學的法則，具體體現在其成長、擴散與價值的方式上。

1. 時間法則（成長與衰退）

知識的成長通常呈現冪律學習曲線：起初進步迅速，隨後趨於平台期。這在個人技能（如打字）與產業成本（如飛機製造）中皆可觀察到。但此成長脆弱且易受衰退影響。

• 知識衰退： 若不持續運用，知識會快速衰退。Hidalgo 以二戰期間的自由船（Liberty ships）為例，估計其知識每月衰減 3%~6%。
• 寶麗來案例： 一位復古膠卷愛好者在荷蘭嘗試重啟寶麗來工廠，聘請了前專家的「A‑team」。即使擁有原始設備與最佳人員，最初的膠卷品質仍然很差，且花了多年才恢復。這說明當活動的「肌肉」不保持鍛鍊時，知識會消失。

2. 空間法則（擴散）

知識的擴散並非均勻分布；它受到地理、社會網絡以及知識本身「幾何」結構的限制。

• 相關性原則： 知識在相關活動之間更容易流動。Hidalgo 以 Vespa 踏板車為例，該車是由航空工程師 Corradino D'Ascanio 設計的。二戰後，意大利被禁止製造飛機，工程師便轉向最「相關」的輕型車輛製造。日本與德國也出現類似情形，證明創新常沿既有能力的地圖演進。
• 移民的角色： 高技能移民是「不相關」跳躍的載體，為經濟的字母表帶來新「字母」，而本地勞動者較擅長漸進、相關的創新，可能無法自行取得這些新字母。

3. 價值法則（複雜度）

一個經濟體的價值不由其財富或資訊量決定，而是由其「經濟複雜度」——可協調的非同質化能力的多樣性——所決定。

• 無限字母表： 每一項獨特能力就像無限字母表中的一個字母。國家的成長潛力取決於它擁有多少這樣的「字母」以及它們如何重新組合。
• 成長預測： 透過分析出口資料估算一國的複雜度，Hidalgo 能預測未來成長。複雜度高於其現有收入水平的國家（如印度）較可能實現快速成長；相反，收入高但複雜度低的國家（如卡塔爾）則成長潛力有限。

架構創新與既有者的困境

創新可分為漸進（元件）變化與架構變化。漸進創新是更換系統中的某個部件（例如在螺旋槳飛機上換裝更強大的引擎），而不改變整體系統。架構創新則需要重新設計整個系統（例如從螺旋槳飛機轉向噴射引擎）。

這解釋了為何既有企業常常失敗。Barnes & Noble 雖然對圖書的「知識」與 Amazon 相同，但其組織架構是為批發與零售設計的，並不適合直接面向消費者的物流。將單本圖書寄送給消費者的轉變不是小幅的漸進改變，而是一次需要全新組織設計的架構創新。

知識、制度與大型語言模型（LLM）

制度 vs. 知識

經濟發展常被誤認為只需要提供金融資本或推行制度改革。然而，Hidalgo 主張制度需求往往來自於高知識密集型的勞動者。中國的創業自由推動就是由高階物理學家與研究人員發起的，他們看到美國的「教授‑創業家」模式，並要求在中國建立相應的制度空間。

大型語言模型有知識嗎？

從 Hidalgo 的觀點看，知識是一種集體現象。因此問題不在於大型語言模型（LLM）是否「個人」擁有知識，而在於它是否提升了人類生態系的集體智慧。LLM 如同書籍般的文化技術，協助人類更有效率地檢索與重新組合資訊，從而加速個人與集體的學習。

---

摘要 César Hidalgo 主張知識是一種不可替代、集體的現象，遵循類似物理的成長、擴散與衰退法則，意味著它不能僅透過下載或複製而獲得，必須具備具身的經驗。

標題 César Hidalgo 談《無限字母表》與知識法則