Amid the buzz surrounding artificial intelligence (AI) and its potential to revolutionize industries, the economic implications remain elusive. While substantial investments flow into AI development, the tangible outcomes are still unclear.
The exploration of AI and its economic ramifications is central to the research of MIT’s Institute Professor Daron Acemoglu, a Nobel Prize-winning economist. Acemoglu has a long history of examining how technology shapes societal structures, from analyzing how innovations spread to studying the effects of robotics on employment.
In a notable achievement, Acemoglu, alongside his colleagues Simon Johnson (MIT Sloan School of Management) and James Robinson (University of Chicago), was awarded the 2024 Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel. Their groundbreaking research investigates how political institutions influence economic growth, revealing that democracies with strong rights foster better long-term growth than other governance systems.
Since technology is a primary driver of growth, Acemoglu is keenly interested in how AI will be integrated into society. Recently, he has published several papers addressing the economics of AI, prompting questions about the future of human tasks within this transformative landscape. “Where will the new opportunities for humans arise with generative AI?” he posits. “The future applications that genuinely transform our workflows remain uncertain.”
Understanding the Economic Impact of AI
Historically, the United States has maintained an average GDP growth rate of about 3 percent per year since 1947, with productivity growth at roughly 2 percent. Some forecasts suggest that AI might double this growth rate or propel it to new heights. However, in his study “The Simple Macroeconomics of AI,” published in the August issue of Economic Policy, Acemoglu estimates a more modest projection: an expected GDP increase of 1.1 to 1.6 percent over the next decade, translating to an annual productivity gain of around 0.05 percent.
Acemoglu’s analysis incorporates recent research highlighting the potential impact of AI on the workforce. For instance, a 2023 study from OpenAI, OpenResearch, and the University of Pennsylvania indicates that approximately 20 percent of job tasks in the U.S. are susceptible to AI intervention. Meanwhile, a 2024 report from MIT FutureTech, the Productivity Institute, and IBM finds that roughly 23 percent of computer vision tasks may soon be automated profitably. Furthermore, data suggests an average operational cost saving of about 27 percent from AI implementations.
On the topic of productivity, Acemoglu notes, “While a 0.5 percent gain over ten years may not seem significant, it is still an improvement. However, it falls short of the ambitious promises circulating in the tech industry.”
While these projections might seem conservative, they are noteworthy given the potential emergence of new AI applications. Acemoglu emphasizes that his calculations do not account for emergent technologies, like AI solutions for protein structure predictions, for which other scholars received a Nobel Prize in October.
Some experts speculate that AI-related displacement may spur job reallocations that enhance growth, but Acemoglu cautions, stating that changes in job allocation typically yield minimal benefits: “The direct productivity gains are the critical aspect,” he asserts.
The Future of Employment
Assessing the implications of AI can refine our perceptions of its impact. While many predict a groundbreaking revolution, Acemoglu’s work prompts a more measured outlook on expected transformations. “Considering AI’s potential effects by 2030, how fundamentally different will the U.S. economy truly be?” he raises. “That’s open to interpretation; some may foresee mass job losses from chatbots, while others envision enhanced productivity across job roles. My belief is that businesses will largely maintain their current operations. Essential roles, such as journalists, financial analysts, and HR personnel, will continue to persist.”
If this perspective holds, AI’s influence is likely confined to specific white-collar tasks where massive computational power can outperform human input, predominantly concerning data analysis, visual matching, and pattern recognition—activities constituting about 5 percent of the economy.
Though Acemoglu and Johnson have sometimes been labeled AI skeptics, they identify as pragmatic observers. “I aim to adopt an optimistic stance,” Acemoglu states. “Generative AI possesses potential, but I believe that its current application trends are misdirected.”
Harnessing AI for Enhanced Worker Productivity
Acemoglu expresses concern regarding the direction of AI development, particularly whether it will prioritize “machine usefulness”—technologies that enhance worker productivity—over replacing human jobs. He contrasts the productive augmentation of a biotechnologist with the replacement of customer service personnel through automation. Currently, he argues, companies are leaning more towards the latter approach.
In their acclaimed book, “Power and Progress” (PublicAffairs), Acemoglu and Johnson delve into this topic, posing critical questions about the nature of technological growth: Who reaps the benefits of technological advancements—society at large, or a select few elites?
As advocates for innovations that enhance worker productivity while maintaining employment, Acemoglu and Johnson emphasize the need to focus on technology that sustains inclusive growth. However, Acemoglu views the trend toward generative AI as prioritizing human mimicry rather than complementary utility. He refers to this pattern as “so-so technology,” which leads to only modest improvements over human performance while providing cost benefits to firms. He sees the emphasis on solutions that supplement human roles as underexplored in the industry’s current landscape.
Lessons from History on AI’s Workforce Transformation
Acemoglu and Johnson’s recent paper, “Learning from Ricardo and Thompson: Machinery and Labor in the Early Industrial Revolution — and in the Age of AI,” published in Annual Reviews in Economics, scrutinizes historical patterns amidst current debates about AI. They argue that the widespread benefits of technology are not guaranteed. The experience of 19th-century England, which saw technological advancements following prolonged social struggle, illustrates this point.
“When workers lack leverage to claim their share of productivity gains, wages are unlikely to rise,” Acemoglu and Johnson explain. “AI may elevate average productivity, yet it risked replacing many positions and diminishing job quality for those who remain.” Their insights emphasize that the relationship between productivity enhancement and wage improvement is far more intricate than commonly assumed.
Acemoglu highlights the evolution of David Ricardo, who initially celebrated technological advances for their potential societal benefits, yet later acknowledged their capacity to harm workers when they replaced rather than supported them. This historical context underscores the importance of evidence-based understanding of AI’s impact today.
The Optimal Pace for Innovation
While rapid technological advancement may seem beneficial for economic growth, Acemoglu and doctoral student Todd Lensman suggest a more cautious approach in their paper “Regulating Transformative Technologies,” published in the September issue of American Economic Review: Insights. They argue that beneficial technologies that also pose risks should be implemented gradually until potential drawbacks are mitigated.
A flush with caution, Acemoglu asserts, “Market and technology fundamentalism may insist on unrestrained innovation speed, but economics does not dictate such rules. Thoughtful consideration, especially regarding potential harms, is warranted.”
These risks could include negative effects on job availability, the proliferation of misinformation, or consumer exploitation in areas such as online advertising. Acemoglu investigates these scenarios further in an upcoming paper titled “When Big Data Enables Behavioral Manipulation,” which he co-authored with counterparts from Duke University and the University of Toronto.
“If we direct AI towards manipulation or excessive automation instead of enhancing knowledge and expertise among workers, we likely need a course correction,” he emphasizes.
Some may argue that the benefits of rapid innovation outweigh potential drawbacks, leading to unpredictable outcomes. In their September study, Acemoglu and Lensman challenge the notion that technology should always be welcomed for its supposed inevitability, advocating for a balanced evaluation of its trade-offs to foster meaningful discussions on technology adoption.
Navigating the Path for AI Adoption
As we ponder the prospect of a gradual technology adoption process, Acemoglu points out that “government regulation plays a key role” in this transition. However, the framework for long-term AI regulation remains uncertain in both the U.S. and globally.
Furthermore, he suggests that a cooling of the current AI hype could naturally slow the adoption rate, possibly yielding more sustainable growth, particularly if investments do not yield quick profits for firms. “The rapid pace we observe stems from investor mania anticipating breakthroughs in artificial general intelligence,” he notes. “This hysteria often leads to misallocation of resources in technology and prompts hasty business decisions.”
Acemoglu underscores that the fervor surrounding AI can significantly impact its trajectory. “The greater the speed and hype, the more challenging it becomes to enact necessary course corrections,” he warns. “It’s much harder to make a pivotal change at breakneck speeds.”
Photo credit & article inspired by: Massachusetts Institute of Technology
