No big bang

AI’s impact will be a process of gradual acceleration as the weak-link effect will temper explosive growth

How will artificial intelligence reshape the economic landscape? Unlike previous technological revolutions, which primarily substituted for or augmented human labor, AI increasingly operates in the domain of human intelligence. If it can extend its capabilities to both physical and cognitive tasks — especially when embodied in robotics — then, in principle, few human activities would lie beyond its reach.

Yet history counsels caution. Over the past roughly 150 years, per capita GDP growth in the United States has remained remarkably stable at around 2 percent a year on average. During this period, we have witnessed a series of technological breakthroughs, such as electrification, the rise of computers and the internet. While these innovations have reshaped industries and boosted productivity, their long-term impact on aggregate growth rates has been more gradual than revolutionary. Perhaps each generation of general-purpose technology does not push growth rates higher so much as take the baton and keep it from falling. AI may simply be the latest runner in that relay.

The underlying mechanism reflects a “weak-link effect” in the economy: Economic production can be understood as a chain of interconnected tasks. The strength of the chain is determined not by its strongest components, but by its weakest ones. As long as certain tasks remain beyond AI’s capabilities, or cannot be performed efficiently, they become binding constraints, limiting overall gains in economic output.

This implies that even if AI were to dramatically enhance productivity in a single sector, such as software, the overall impact on GDP would remain limited, since the sector accounts for only about 2 percent of total output. Under the weak-link effect, aggregate gains are capped at roughly that share. The key to AI-driven growth, therefore, lies in systematically alleviating bottlenecks, shifting more and more tasks from “human mode” to “machine mode”.

How AI may shape the future of the economy can be illustrated through three hypothetical scenarios. First, there will always be tasks that remain inherently human — about 5 percent — that cannot be automated. Second, AI will eventually perform all other tasks. Third, a middle ground emerges, where the share of automation steadily advances but never achieves full substitution within a finite time horizon.

A recent study by Stanford economists Chad Jones and Chris Tonetti finds that over the next 75 years, the GDP trajectories under the above-mentioned three scenarios are remarkably similar. Divergences in the economic outlook only emerge further into the future. The weak-link effect tempers explosive growth, turning it into a process of gradual acceleration.

This suggests that the medium-term economic trajectory is already largely determined, providing time to prepare for the deeper transformations that may unfold over the longer term. Three issues demand attention.

The first concern is the most widespread: employment. The weak-link effect is instructive, too. In 2016, Geoffrey Hinton, a founding father of deep learning, said that training radiologists should stop because AI would soon surpass humans at reading medical images. But a decade later, the number of radiologists has increased, and their wages rose. Why? Because a radiologist’s job involves more than just image reading — it includes patient communication, collaboration with clinical teams and the synthesis of diverse clinical information into clinical judgment. By automating the image reading task, AI has actually raised the marginal value of radiologists’ other tasks.

This is the weak-link effect at work in the labor market: a job is a bundle of tasks. When one task is automated, the remaining tasks become the bottleneck and thus command higher returns. AI’s impact on different occupations will therefore be uneven: Some roles will benefit from complementarity, some will shrink due to substitution, and new jobs will emerge. Until AI achieves full automation, efforts to ease employment anxiety should focus on the weak links that machines cannot handle well, which is precisely where human value lies.

The second concern is distributional — how to share the larger pie that AI helps create. Historically, labor has been the primary productive factor for most individuals, with income derived from supplying human capital. If AI displaces human labor on a large scale, those who do not own AI-related capital, including computing power, data and models, risk being left behind.

The third concern is more existential. One category of risk arises from the misuse of increasingly powerful tools. As AI capabilities scale while access becomes more widespread, the barriers to applying AI tools are falling. This creates the potential for asymmetric risks, where a small number of actors can generate disproportionately large harm. For example, if malicious actors harness advanced AI to engineer biological pathogens more lethal and harder to detect than any known today, the consequences would be devastating. Traditionally, the most destructive capabilities have been concentrated in the hands of a limited number of actors. However, the proliferation of advanced AI could distribute certain high-impact capabilities far more widely.

A second category of risk concerns the loss of control over superintelligence. Humans are creating a form of intelligence that we ourselves do not fully understand, raising the possibility that such systems may act in ways that are difficult to predict or constrain. These risks may appear remote, but the severity of their potential consequences makes early preparation essential.

The internet has profoundly changed the world over the past three decades. AI may reshape the economy on an even larger scale, but its effects will take time to fully materialize.

AI should not be underestimated simply because rapid GDP growth is not yet visible. History has repeatedly shown that the economic effects of general-purpose technologies take decades to unfold. What’s more, preparation for enormous changes should begin now. This includes labor market transitions, income distribution and investment in AI safety. We must not wait until AI has already transformed everything before taking action. The greatest risk in the AI era lies not in its power, but in our lack of preparedness.

The author is an assistant professor at Guanghua School of Management and the deputy director of the Institute for Economic Policy Research at Peking University.

The author contributed this article to China Watch, a think tank powered by China Daily. The views do not necessarily reflect those of China Daily.

Contact the editor at editor@chinawatch.cn.