International relations experts often think that great powers become great because they develop certain key technologies that strengthen national power. Perhaps productivity increases, or military capabilities become formidable, allowing expansion everywhere.
So maintaining great power status requires leading in new technologies. But how does one lead technologically?
Political economy experts might say protecting property rights, freedom of speech, etc., makes it easier for the private sector to develop technology.
Military determinism experts would say countries with unstable neighbors are more likely to innovate out of fear. Vested interests are less likely to block innovation to maintain their advantages.
Now with AI technology emerging, the US and China are competing for AI advantage. Which experts should we listen to?
Jeffrey Ding, author of “Technology and the Rise of Great Powers,” offers another perspective: which technology you develop matters. Previous perspectives don’t care about which actual technology to develop, treating chips and potato chips equally.
People have a blind spot when looking at technology—they only see the newest, trendiest breakthroughs. Those technologies capture media headlines and people’s attention. The author calls these “leading sectors.”
For these breakthroughs to have impact and improve productivity, they must deepen into various industries. So the author believes we must additionally identify so-called “General Purpose Technologies” (GPT for short).
Capturing leading sectors can’t make you a great power; mature general-purpose technology development can.
How do we define general-purpose technologies? They must meet three requirements:
- Have potential for continuous improvement
- Can find many different uses across various industries
- Have strong complementary effects—other industries’ technologies need to adapt to them to substantially increase productivity
Because of these three conditions, GPT “diffusion” to various industries and reflection in productivity numbers takes a long time (usually decades).
How does the author argue GPTs matter more than leading sectors? By examining history’s three industrial revolutions:
First Industrial Revolution
- International relations mainstream narrative: Leading sector innovations: cotton textiles, iron making, steam power
- General technologies: Iron making, steam engines, factory system
The problem with leading sectors is that before 1815, those technologies didn’t affect overall productivity. The key to Britain’s rise was rapidly spreading mechanization everywhere, then mechanization’s effects drove productivity growth.
Steam engines changed processes, spawning mechanical engineering. Production shifted from manual to mechanical. Impact on coal mining promoted iron making, fostering mechanization.
Why didn’t France and Netherlands beat Britain at the same time during the First Industrial Revolution? Knowledge and application disconnected. Developing GPTs requires lots of hands-on, tacit knowledge that can’t be explicitly stated. In the 19th century, Britain had over 1,000 industry associations with 200,000 members. France had famous scientists, but Napoleon trained experts for limited political and military purposes, limiting trainees’ ability to combine with industry. Netherlands also lacked science-industry integration.
Second Industrial Revolution 1870-1914
- Leading sectors: Chemicals, electrical equipment, automobiles, steel industry
- General technologies: Chemicalization, electrification, internal combustion engines, interchangeable manufacturing
- Chemicalization: Chemical processes spreading to ceramics, food processing, glass, metallurgy, petroleum and other industries
- Electrification: Central steam engine belt drives → electric motors driving individual machine units
Germany initially made important chemical breakthroughs. But America won chemical production advantages.
The reason was American chemists collaborated with mechanical experts. They developed chemical engineering departments and unit operations (breaking chemical procedures into basic operations).
Mechanically, America developed mechanical engineering curricula and standardized screw threads.
Third Industrial Revolution
- Leading sectors: Computer industry, consumer electronics, semiconductor industry
- General technology: Computerization (recording production information with computers)
Japan pioneered many leading sector technologies, and American international relations experts once worried about Japan’s computer technology lead. But ultimately America became the computer power.
- America: Created computer science departments. 1968 ACM curriculum helped schools organize computer education
- Japan: 1991 report showed Japan relied on outsourcing. Companies couldn’t maintain full-time staff
GPTs demand high human capital—both deep expertise in single domains and ability to combine across domains. The author illustrates two types of AI talent: the former is what people imagine as AI talent, the latter is what people often overlook.
In 2014, Baidu prominently poached deep learning star Andrew Ng from Google. In Alibaba’s 2019 IPO photo was Yuan Wenkai, a warehouse worker with professional automation management knowledge who improved logistics warehouse sorting capacity to 20,000 orders per hour.
Developing GPTs also requires establishing disciplines. Those subjects that delay my fellow classmates’ graduation and the majors students study turn out to be sources of historically crucial national power 🤯
However, the author also says looking at past industrial revolutions involves hindsight. People struggle to predict the next GPT. The author says if he’d written this book 20 years ago, it might have been entirely about nanotechnology.
For me, I actually don’t care much about AI or who becomes a great power, as long as it’s not authoritarian states. I care more about technologies I’ve participated in: how blockchain and cryptography become useful technologies.
Computer science has another definition of “general”: people can command computers through programs to accomplish desired business logic. Ethereum and programmable cryptography have these properties.
But this “generality” also seems like a very useless “generality.” Often when you want to use blockchain for some application, people ask why not just use a database. Even now, besides digital wallets (based on programmable cryptography), I really struggle to recommend people use blockchain or new cryptography technologies, unless you really need blockchain-appropriate tools.
It’s hard to say these technologies aren’t ready—blockchain has developed for over a decade. AI draws a picture or speaks a sentence, and people immediately sense its use. Blockchain’s value is very abstract—it’s more like a security product. Security products need corresponding threats for adoption. Signal had few installations before the Ukraine war, but downloads surged after fighting began.
Security products should ideally be like HTTPS, adopted without users knowing. Users don’t worry about threats but are already protected.
Obviously imagining blockchain as a leading sector is difficult, but can it become some “blockchainization” general technology? Let’s hammer the blockchain nail with the GPT hammer:
- Continuous improvement potential: Seems so! Architecture can improve with single computer performance or cryptography performance improvements.
- Broad use across industries: If we ask people’s expectations for online platforms, everyone wants social media algorithms and food delivery pricing not unilaterally controlled by platforms. I believe every online service we use should have some profit-distribution calculations locked on blockchain.
- Strong complementarity: Maybe. Many “why not use databases” reasons are that systems’ weakest links negate all blockchain benefits. For example: since data recording requires human input, blockchain’s fairness and transparency are meaningless with possible malicious input. Therefore all effective blockchain applications accommodate blockchain in design, replacing input with economic or cryptographic mechanisms.
I also hope GPT definitions don’t become post-hoc recognition but provide pre-development technology guidance.