经济学周刊现代版恩格斯停顿不是人工智能的必然结果

William Blair There is no shortage of confusion around AI and whatimpact it will have on economic growth and, more im-portantly, the labor market. The reality, of course, is thatno one really knows how this will play out. What we cando is construct a few stylized scenarios, select the onewe mostly agree on as being the best for humanity, andwith government intervention (nudges, guardrails, taxpolicy, and regulation) start to drive in that direction.A recentviral postseems to suggest that the economicstagnation experienced during the first industrial revolu- Probably the closest historical example we can draw totoday’s scenario is Engels’ pause, which took place duringthe first industrial revolution and is named after the so- The pause lasted for approximately 50 years (from 1790 to1840) and started almost 80 years after the introductionof the first commercially successful steam engine in 1712. What caused the pause is the fact that the gains fromthose improvements, the corporate profits, were notshared equally across all divisions of labor. A rising tidedid not lift all boats. Rather, the spoils were overwhelm-ingly reaped by the owners of capital (e.g., the mill and The standard neoclassical or Solow-Swan economic growth Output = Labor Force Growth + Growth in the Capital During this period, labor was abundant and possessedvery little bargaining power; workers’ rights were virtu-ally nonexistent. Today, the labor market is structurally The equation simply tells us that economic growth isdriven by a combination of how much capital (i.e., plant,equipment, and machinery) we have and how many work-ers are available to use that stock of capital. Productivity, ortechnological progress, comes in at the end as an exoge- Back then, labor was increasingly displaced, and realwages fell despite deflation across much of the economy(exhibit 1). Unfortunately, that was not the case for foodand accommodation costs, which remained high due tothe Corn Laws and severe overcrowding. That “surplus”population led to poor sanitation, while the belching fac-tories resulted in a new problem—pollution. These were In one future scenario—the one depicted by many ofthe techno-pessimists (or are they optimists?) and thatfinancial market participants have started to discount overthe last month—the labor component of the Solow growth (where they maintain expert-level intelligence across allfields), move, speak, and act just like humans (e.g., kung fufighting robots from China), and yield their own techno-logical progress, human life faces a clear existential threat.Certainly it is hard to see how they find relevance enoughto feature as a cost (or income for those labor recipients) in William Blair suitable guardrails around the use of AI. This is a moreflexible scenario that explicitly recognizes the lump of The fallacy rests on the belief that there is only a finiteamount of work in an economy and, therefore, only afixed number of jobs to carry out those tasks. It takes theview that the introduction of new technology, or new im- In reality, while these factors can often be extremely dis-ruptive and harmful to segments of the labor market, weknow that labor is ultimately capable of evolving and creat-ing jobs that never previously existed. Recent introductionsinclude drone pilots, robot engineers, specialist technology The “godfather of AI,” Geoffrey Hinton, famously stated in2016 that medical schools should stop producing radi-ologists as it was “completely obvious” that within aboutfive years machine learning would be able to outperformthem at both image recognition and interpretation. Sincethen, Hinton has admitted he was wrong about this pre-diction, as the number of radiologists has been steadily The Invisible HandEngels’ pause should be seen as a clarion call. If nothing changes, we too could face a similar future, only with data Left unchecked, this pause could be even worse, giventhat technologically advanced machines are targetingnot just physical capabilities, but cognitive ones as well.It is quite possible that in this scenario we emerge into aworld where only a small handful of dominant technol-ogy companies control vast swaths of the tech sector and Indeed, a recentstudyfrom theHarvard Business Reviewhas found that rather than reducing work, AI has intensi- In an eight-month study of how generative AIchanged work habits at a U.S.-based technologycompany with about 200 employees, we foundthat employees worked at a faster pace, took ona broader scope of tasks, and extended work intomore hours of the day, often without being askedto do so. Importantly, the company did not man-date AI use (though it did offer enterprise sub- The Visible FistAn alternative scenario, and the one some countries seem to already be heading toward, is where the state retainscontrol of the data, algorithms, and production. Evengreater and more intrusive surveillance is introduced, andAI is used to manage and optimize labor and ca