AI could help lower-skill economies gain more, but adoption cost is key, says economist
As AI reduces the probability of failure, firms tend to deploy it where the potential returns are highest – typically at the weakest link in a production chain.
Artificial intelligence could deliver larger productivity gains in lower-skill economies than in advanced economies, potentially helping narrow global income gaps, but high adoption costs could limit those benefits, economist Jean-Louis Arcand said today (10 August).
Speaking at a public lecture at BRAC Centre Inn in Dhaka, Arcand said, "Any divergence must run through the adoption-cost channel," while presenting a research paper on AI and productivity.
The lecture, titled "These Aren't the Droids You're Looking For: Endogenous AI, O-Rings, and the Bottleneck Reallocation Theorem", presented Arcand and Balasubramanyam Pattath's extension of Michael Kremer's O-ring theory of production to the age of AI.
Keep updated, follow The Business Standard's Google news channel
Arcand, president of the Global Development Network (GDN), said AI does not eliminate the O-ring mechanism but instead relocates it. The theory holds that production depends on a chain of complementary tasks, where failure in one task can undermine the value of the entire production process.
As AI reduces the probability of failure, firms tend to deploy it where the potential returns are highest – typically at the weakest link in a production chain. Once that constraint is eased, another task becomes the new bottleneck, according to the research.
The researchers tested their model using data from 227 radiologists alongside occupation-level evidence on AI use. They found that AI intensity declined as radiologists' skill levels increased, while 38.8% of the radiologists in the sample were made worse off by AI assistance.
Across occupations, AI use increased with wages and was also higher where the consequences of failure were greater.
After accounting for AI adoption costs, the model estimated a one-off aggregate output gain of around 0.5%, with virtually no wage compression at current costs. Wage compression would emerge only if AI became cheaper, the research found.
The model also suggested that lower-skill economies could gain more from AI than higher-skill economies at every level of task complexity, provided adoption costs were equal. However, this advantage diminished as production chains became longer.
Under the researchers' calibrated assumptions, the comparative advantage of skill-scarce economies would explain only around 0.35% of the income gap between Sweden and Chile.
Arcand said the findings place the cost of AI adoption at the centre of the policy challenge for poorer economies. Digital infrastructure, connectivity, reliable electricity and complementary institutional investment will be crucial for such countries to capture AI-driven productivity gains.
The lecture was opened by Selim Raihan, executive director of the South Asian Network on Economic Modelling (Sanem) and professor of economics at the University of Dhaka, and was followed by an open discussion with participants.
