(Roughly) Daily

The estimable Tim O’Reilly reminds us to think deeply about how AI could and should turn out. He suggests that Jeff Ding‘s diffusion theory of the role of technology in great-power competition also applies to AI adoption– and that it suggests that companies obsessed with the frontier might be optimizing for the wrong thing…

In the 1980s, Japan led the world in semiconductors, consumer electronics, and computer hardware, the industries everyone assumed would decide the next phase of economic power. Japan won them and still did not overtake the United States in the information revolution that followed. Jeff Ding, a political scientist at George Washington University, opens his book Technology and the Rise of Great Powers with the history of the first and second industrial revolutions and the third, the information revolution. The explanation he gives for who wins and who loses applies to companies as well as it does to nations, and very much to the current trajectory of AI.

Ding contrasts two theories of how technological revolutions reshape economic power. The conventional one he calls the leading sector model, or LS theory. It goes like this: New technologies create fast-growing new industries like steel and railroads and automobiles and semiconductors, and the country that dominates invention in those sectors captures the monopoly profits and the upstream and downstream economic linkages that come with them. As the story goes, if you win the leading sector, you win the era. Britain won in the first industrial revolution through its mastery of steam power, and then was surpassed by the US in the second through its leadership in electrification, the internal combustion engine, and mass manufacturing. The US kept its lead over Japan in the information systems revolution not by competing in the “leading sector” of electronic hardware but by diffusing “up the stack” via software that took the power of computing into every sector of the economy. (OK, that last bit is my explanation of what happened rather than Ding’s, but it’s consistent with his theory.)

Leading Sector theory is pretty clearly the working hypothesis of today’s AI industry and the national strategy that is forming around that industry. The company and the country with the biggest and best models wins. Everyone else is an also-ran.

Ding offers another explanation, which he calls diffusion theory. He points out that general-purpose technologies, foundational ones like the steam engine, electricity, and the computer, don’t just create massive profits and productivity gains in a single industry but instead spread across the whole economy. National economic leadership comes not from inventing the new sector but from diffusing the general-purpose technology more quickly and more broadly than your rivals. This happens over decades. The win goes to whoever most successfully embeds the technology into a wide range of ordinary productive work. This is how the US kept its lead over Japan rather than being surpassed by it.

This is obviously aligned with the thinking of Arvind Narayanan and Sayash Kapoor in “AI as Normal Technology,” which Ding cites in his book.

A big part of what enables diffusion is what Ding calls skill infrastructure, the education and training systems that widen the pool of people who can actually work with the technology. When the priority is widespread adoption rather than invention, he argues, the institutions that matter are the ones that build engineering skill at scale, standardize good practice, and tie research to industry. He writes:

GPT diffusion theory highlights the importance of GPT [General Purpose Technology] skill infrastructure. Education and training systems that widen the pool of engineering skills and knowledge linked to a GPT. When widespread adoption of GPTs is the priority, it is ordinary engineers, not heroic inventors, who matter.

Music to my ears, as it should be to yours: “It is ordinary engineers, not heroic inventors, who matter.”

That is not how the current AI narrative goes. Everyone is fixated on the labs, the frontier models, and the most famous researchers. And that fixation shapes enterprise strategy. Inside many companies AI strategy is a procurement decision: Which model and which vendor and which flagship tool should we choose? Or it’s a moonshot to stand up a lab and build an impressive demo and hire your own famous developer. Both approaches treat AI as a sector to be won. Ding’s argument is that the breakthrough sector itself is not where the long-term value for national power lives. And I believe that the same applies to corporate success. The value is in how widely and how well the technology gets embedded into the work of the people you already employ. The company that puts AI to work in finance and support and legal and sales and operations, across every unglamorous process, as well as in product and engineering, outperforms its competitors and drives its industry forward.

The reason diffusion takes a long time is that it is an organizational problem and not a technical one…

[Tim elaborates, and specifies the requirements for successful management of what is an “enterprise transformation problem”; he then unpacks the geopolitics of AI. He concludes…]

… Sovereign AI is not just a matter of national power. It is a predictable consequence of diffusion. A technology that diffuses widely will be adapted by different societies, firms, and institutions to suit their own needs, values, and constraints. Sovereign AI is AI designed for diffusion, not just raw increases in capability.

This is one reason the arms-race framing is unhelpful. It encourages us to treat AI as if it were a weapons system or a scarce strategic asset. But if AI is closer to electrification, computing, or the written word, the important thing is how the technology is embedded into the ordinary life of economies and institutions, and whether that embedding happens in ways that increase agency broadly rather than concentrating it in a few hyperpowerful companies.

There are a few additional lessons we can take from the history of electrification. While motors became decentralized, factories stopped generating their own power and bought it from a centralized grid. The unit-drive revolution decentralized application, not generation. This limitation, which we are now working to overcome to some extent with decentralized solar generation, is perhaps ironically showing up most strongly in the strain that AI data centers are placing on the grid. Let’s learn from that misstep. You can diffuse AI into every workflow via API calls to a big centralized model, or it can be diffused by a network of smaller models that turbocharge every part of the economy.

We should design for a future of multiple AIs, not a single universal system. Different countries will want systems shaped by different legal regimes, languages, histories, and cultural assumptions. So will companies. So will professions and communities of practice. The instinct of some frontier labs is to imagine that the right answer is to homogenize the technology, purge it of bias, and offer a single sanitized intelligence layer for the world. But AI is a social and cultural technology. The differences are not a defect to be smoothed away.

We do need to think about standards and interoperability. The historical analogy that comes to mind is railroad gauge. When real world systems are built to incompatible standards, the result is not healthy diversity but decades of friction, kludges, and retrofitting. The same may prove true for AI. If we force the future into a choice between one universal model and a patchwork of disconnected sovereign systems, we will get the worst of both worlds. We need a layer between uniformity and fragmentation, which can come from standardized protocols that allow different models, tools, and institutions to interoperate without requiring them to become identical.

This is also why open source matters, but only if it is properly understood. Open source is not just about licenses. My earliest introduction to the shared development of software that now goes by that name came from the research community that grew up around Bell Labs’ Unix operating system despite AT&T’s proprietary (albeit permissive) licensing. Because of that experience, I became convinced that it was the modular, protocol-centric architecture of Unix that was a key driver of collaborative, internet-enabled software development.

Open source AI depends on far more than open models. It depends on the architecture of participation built into the systems above and around them: the protocols, servers, interfaces, and shared technical conventions that let many different actors build on common foundations. The Open Source AI Gap Map shows just how rich that open source AI ecosystem is becoming. But open source can also coexist with proprietary, de facto standards like the OpenAI and Anthropic APIs. Like the electric grid we are now beginning to rebuild, the AI future will be a mix of centralized and decentralized systems. Cooperation and competition can coexist. Different actors can build different systems, for different purposes, under different forms of governance, while still participating in a shared technical and economic order.

This is how the future can belong not just to the inventors of AI but to the people who make it usable, adaptable, interoperable, and worth adopting.

Eminently worth reading in full. AI for all of us: “Ordinary Engineers, Not Heroic Inventors,” from @timoreilly.bsky.social

Apposite: “How to talk about “AI” without adding to the anthropomorphization“

* Allan Dafoe

###

As we amplify access, we might we might spare a thought for someone who launched more than one central technology into braod diffusion: the Serbian-American electrical engineer and inventor Nikola Tesla; he died on this date in 1943.  Tesla is probably best remembered for his rivalry with Thomas Edison:  Tesla invented and patented the first AC motor and generator (c.f.: Niagara Falls); Edison promoted DC power… and went to great lengths to discredit Tesla and his approach.  In the end, of course, Tesla was right.

Tesla patented over 300 inventions worldwide, though he kept many of his creations out of the patent system to protect their confidentiality.  His work ranged widely, from technology critical to the development of radio to the first remote control.  At the turn of the century, Tesla designed and began planning a “worldwide wireless communications system” that was backed by J.P. Morgan…  until Morgan lost confidence and pulled out.  “Cyberspace,” as described by the likes of William Gibson and Neal Stephenson, is largely prefigured in Tesla’s plan.  On Tesla’s 75th birthday in 1931, Time put him on its cover, captioned “All the world’s his power house.”  He received congratulatory letters from Albert Einstein and more than 70 other pioneers in science and engineering.  But Tesla’s talent ran far, far ahead of his luck.  He died penniless in Room 3327 of the New Yorker Hotel.

 source