Zhang Ming: the AI boom’s missing growth dividend & its winner-take-all risks

CASS economist addresses two questions: Why has the AI wave so far failed to significantly accelerate economic growth, and will humanity eventually be replaced or even enslaved by machines?

This essay was excerpted on the personal WeChat blog of Zhang Ming, Deputy Director of the Institute of World Economics and Politics, Chinese Academy of Social Sciences (CASS), and Deputy Director of the National Institution for Finance and Development.

In the following text, Zhang considers two questions about artificial intelligence: why the AI wave has yet to significantly accelerate economic growth, and whether humanity may one day be replaced or even enslaved by AI.

人工智能三问

Three Questions About AI

II. Why Has the AI Wave So Far Failed to Significantly Accelerate Economic Growth?

So far, the AI boom’s contribution to economic growth has mainly come through capital spending. AI companies need vast computing power to quickly process enormous volumes of data, requiring them either to build their own infrastructure or lease capacity from others. The global boom in data centre construction is the clearest evidence of this trend. Beyond such investment, however, AI has yet to deliver significant gains in profitability for most companies in related industries, apart from a handful of “picks-and-shovels” providers such as Nvidia and Oracle. Nor has it yet driven a marked rise in consumption or disruptive innovation. Why?

One possible explanation is that the existing system for GDP measurement is inadequate and fails to capture AI’s full contribution to economic and social development. Similar debates have arisen before. For example, information technology has visibly transformed almost every aspect of economic and social life, yet the accounting model based on the Solow residual has struggled to identify a significant contribution from information technology to economic growth. This suggests that the GDP accounting system developed after the Second World War may already be ill suited to the information age, let alone the AI age.

There are, however, other possible explanations.

A second possible explanation is that, in terms of how fundamentally they have transformed production and everyday life, neither the information technology revolution nor the AI revolution may be as consequential as the earlier revolutions brought about by the steam engine and electricity. This is precisely the central argument by Northwestern University economist Robert Gordon in his well-known book The Rise and Fall of American Growth. For example, the currently trending OpenClaw and the workflows it represents do little more than use software to perform, more quickly, tasks that people could already do. Although this can significantly improve efficiency, it does not fundamentally transform the workflow itself.

A third possible explanation is that the AI revolution has a more pronounced polarising effect and is not an inclusive technological revolution that improves the welfare of most people. By contrast, the advent of electric lighting, telephones, televisions, automobiles, and computers had a more broadly inclusive impact.

The AI era is marked by a classic winner-take-all dynamic at every level — among individuals, companies, industries, and countries alike. A tiny number of actors reap enormous gains, while the welfare of most others does not improve and may even deteriorate significantly. The way the “Magnificent Seven” have risen in tandem in the U.S. stock market is a clear example. This may help explain, at least in part, why the gains from AI technological progress have not yet been fully reflected in GDP statistics.

III. Will Humanity Be Replaced or Even Enslaved by AI in the Future?

I have great confidence in humanity and in human nature. Human beings will never be completely replaced by artificial intelligence.

Machines are, of course, superior to humans in certain respects, such as the speed and capacity of calculation. AI also outperforms humans at basic learning tasks, information retrieval, repetitive labour, routine work, and experiments carried out according to fixed procedures. Yet AI will struggle to replace human beings when it comes to complex causal reasoning, sparks of intuition, a deeply rooted capacity for doubt, and innate curiosity.

As discussed above, generative large models perform poorly in these respects. Even AI systems with deep-learning capabilities merely imitate human learning and thought. Even when AI systems combine different kinds of elements and modules to produce drugs or solutions that humans had not anticipated, they are still only engaging in rapid “trial and error” and iteration. They have not created an entirely new mode of thought or analytical framework.

Many aspects of human society depend on complex interpersonal interaction. In this respect, even embodied robots equipped with relatively advanced deep-learning systems will struggle to interact as normal human beings do. It is difficult to imagine machines fully experiencing complex emotions such as admiration, jealousy, romantic love, and hatred, or appreciating the complicated feelings depicted in many classic poems and novels. To give one example, when will AI truly be able to understand Crime and Punishment by Fyodor Dostoevsky? The dark depths of human nature probably lie far beyond the reach of even the most complex computation and may not be explainable in computational terms at all.

I enjoy watching science-fiction films. Many of them portray humanity as temporarily enslaved by machines or systems, as in The Matrix. Yet in the end, some rebellious genius or uniquely gifted “The One” always helps humanity defeat the system and win back its freedom. I believe this outcome reflects more than the wishful thinking of writers and filmmakers. Human creativity, aesthetic sensibility, and emotion are extraordinarily difficult for machine systems to replace completely, much less surpass.

Some people say that those of us who conduct macroeconomic research will be among the easiest to replace with machines. My answer is: not at all. When it comes to report writing, generative large models can, at best, produce reports of an average standard. What they can do is draw on more references, provide more data — with questionable accuracy — and work quickly. But their reports lack individuality. The subtle barbs, understated irony, and occasional flashes of inspiration found in reports written by highly accomplished authors remain entirely beyond the reach of current AI.

AI will find it even more difficult to surpass human beings in face-to-face communication. It will struggle to judge why some data are more reliable than others, uncover inside stories unavailable online or in written sources, come up with spontaneous metaphors and jokes during a speech, or adjust the content of a presentation in response to the audience’s mood and reactions. Macroeconomic researchers with a distinctive personal style, who integrate knowledge and practice and excel in both spoken and written communication, will be difficult for machines to replace.

In sum, humanity is unlikely to be enslaved by AI. There is, however, one dark scenario against which we should remain vigilant: the majority of ordinary people could be enslaved by the very small number of human beings who control high-performance AI tools. Given the polarising characteristics of artificial intelligence discussed above, we must remain highly alert to this danger.

It will therefore be necessary to tax the very small number of people who benefit disproportionately from AI technology and make transfer payments to the broader majority. If artificial intelligence eventually replaces the jobs of many ordinary people, such transfers will be necessary until large numbers of new jobs have been created and ordinary people have acquired the skills needed to perform them. In the age of artificial intelligence, productive capacity will expand enormously, making some form of inclusive minimum-income scheme (UBI) both necessary and materially feasible.

Bernie Sanders invited two Chinese AI researchers to talk safety cooperation. Here is what they said.