Productivity: diffusion is more important than invention in era shifts

INTRODUCTION:

Technology, productivity and country leadership

ChinaTalk became my latest goto podcast in 2023. Despite the name, the current themes are more related to an introspective view of technology strategy, leadership and how being the inventor does not associate with leadership.

This post is to capture the source documents and early thoughts. I still study the Ding paper and the podcast participants including from Deepmind, to understand better Innovation capacity, diffusion surplus and diffusion deficit alongside how we can think about our example, AI, and its early, medium and long term contribution to growth. This is all about the how AI will produce a long term Era impact.

The genius inventor experiences a Eureka moment. An awesome engineering marvel reaches completion. A new theorem changes everything. These are the images that come to mind when most people picture scientific and technological advance. Neglected in the collective imagination is the toil of diffusion: an invention becomes a standardized product, an engineering marvel is re-constructed in another context, a theory spreads from one institution to another.

Technology leadership is cultural. The ‘diffusion deficit’ refers to the lagging multiplier effect of spreading the benefit of technology so that it becomes mainstream within that country. Where the effect is greater than expected this language would refer to ‘diffusion surplus’. Both concepts are in the paper.

This paper argues that these assessments should, instead, place greater weight on a state’s capacity to diffuse, or widely adopt, innovations. Specifically, when there is a significant gap between a rising power’s innovation capacity and its diffusion capacity, relying solely on the former results in misleading appraisals of its potential to sustain economic growth in the long run.

I demonstrate this with two historical cases: the U.S. in the Second Industrial Revolution and the Soviet Union in the early post- war period.

Lastly, I show that, in contrast to assessments based on innovation capacity, a diffusion-centric approach reveals that China is far from being a science and technology superpower.

In the podcast the discussion got into the classic example of the British Industrial Revolution. Many associate with Cotton production and worldwide sales. Jordan says different. While core inventions in S&T (Scientific and Technological) produced steam power as a propellant, he argues it was Iron production which created the capacity to “diffuse, or widely adopt, innovations” and this generated the revolution. In this context we can think of the widespread adoption rates of factories and rail networks driven by that invention.

The concept of diffusion deficit or diffusion surplus lies in the country’s capacity to lever the invention. This takes the form of widespread entrepreneurial adoption. When we look at the centres of industrial development in England they cover entire counties across swathes of the country even today.

Notes on the podcast followed by early impression on AI and diffusion

• Productivity as real predictor of growth. British Industry revolution was driven by diffusion of GPT.
• Japan was considered a leader in the 1980’s fell behind by a choice to narrowly focus on mainframes which limited and missed the real change that lay in adoption or diffusion of personal computers across the economy.
• Jordan Schnieder- how wrong everyone can be in decadal predictions. Get out of your mini echo chamber. Dose of humility.
• Overestimate as risk in predicting competition between countries a and the trajectory a of technologies.
• Widen the base of AI engineering to become a general purpose technology. (GPT). Anything that improves AI diffusion. Eg community colleges. Voucher system to encourage were some ideas discussed.
• GPT take 20 years to show up in productivity stats.

AI — some super early predictions based on early study of ‘Diffusion Deficit’

Artificial intelligence (AI) currently has limited adoption in a narrow group of technology companies. Success in leadership countries needs to diffuse AI across the economy and the population in order to achieve systemic productivity gains, thus step change in economic growth for the country.

During the podcast Jordan and guests from DeepMind talked about possible industrial policy shifts to engage mechanical engineers, computer engineers electronic engineering, and possibly new types of disciplines. The key point was to engage those disciplines to view AI with a new and different lens than their training to explore and lever the transformational benefits of AI. These transformational benefits would ideally be of the order of magnitude seen in Britain during the 19th Century shifting from manual labour and horses to a world driven by steam power, use of iron on a commercial scale and creation of factories. This would further go on to introduce early social mobility as the economy shifted from agriculture to industrial centred in towns.

The lens through which AI ought to be viewed is in this context; step changes to how economic growth is generated over the coming decades. Any early understandings could produce a future trajectory beyond imagination and certainly beyond the current view of AI as producing smarter search. To disregard diffision deficits will hinder producibility benefits and thus GDP scaling up.

Jeff Ding is the leading US scholar on China and AI and author of one of the earliest China-focused Substacks, ChinAI.

He recently published a fine paper called, “The diffusion deficit in scientific and technological power: re-assessing China’s rise.” It makes the argument that diffusion capacity (not just innovation capacity) is critical to economic growth — and China actually fares much worse in diffusion capacity than mainstream narratives imply.

In particular, “In cases when the emerging power has a strong innovation capacity but weak diffusion capacity (diffusion deficit), it is less likely to sustain its rise than innovation-centric assessments depict. Conversely, when the emerging power possesses a strong diffusion capacity but weak innovation capacity (diffusion surplus), it is more likely to sustain its rise than innovation-centric assessments portray.”

Mainstream narratives, meanwhile, “only compare the U.S. and China’s ability to produce new innovations, neglecting their ability to effectively use and adopt emerging technologies. By revealing the gap between China’s innovation capacity and diffusion capacity, this paper argues that innovation-centric assessments mistakenly inflate China’s S&T power.”

Cohosting the podcast is Teddy Collins, formerly of the White House Office of Science and Technology Policy and DeepMind.