Focus
May 24, 2024 | 13:56
AI, Caramba: Productivity and Job Impacts
Economics
Economics
Economics
Fixed Income Strategy
Economics
Economics
Economics
Economics
Economics
Economics
Economics
Fixed Income Strategy
Economics
Economics
Economics
Economics
AI, Caramba: Productivity and Job Impacts |
|
|
Artificial intelligence (AI) threatens to change the way we work in fundamental ways. The implications are wide-ranging and, at the extreme, border on science fiction. Whether AI is revolutionary or evolutionary will depend on how fast new systems are developed and adopted. This note focuses on how intelligent machines and systems could impact productivity and workers. AI and ProductivityArtificial intelligence packs the potential to improve productivity growth. Labour productivity (output per hour worked) is driven by three factors: (1) capital deepening or increasing capital intensity; (2) labour skill upgrading or increasing the quality of labour; and (3) the growth in multifactor productivity. What makes AI a potentially powerful driver of productivity is that it can influence all three factors. First, capital intensity is increased via investment in equipment, structures, and intellectual property products, which includes the software and hardware needed to implement AI systems. Productivity should get a persistent direct boost as more companies implement such systems. According to last year’s Worldwide Artificial Intelligence Spending Guide from IDC, global outlays on AI-centric systems (including software, hardware, and services) was more than $120 billion in 2022 and is expected to surpass $300 billion by 2026, though more recent estimates point to more rapid growth. It should be noted that this specific spending still pales in comparison to the global outlays on information technology generally. Gartner forecasts this figure will hit $5 trillion in 2024, with spending on devices and software topping $1.7 trillion. AI’s profound potential lies in the other two channels of productivity growth. In a March 2024 survey by Deloitte, private companies were asked to rank the biggest ways in which they expected AI to increase productivity in their enterprises. The second most common answer (at 39%) was via workforce learning and development. (Barely nudging to the top at 40% was reducing product manufacturing cycles and service delivery times.) An often-cited Stanford/MIT study (April 2023) of call centre workers found that the introduction of a generative AI-based conversation assistant lifted average productivity (measured by customer issues resolved per hour) by 14%. This reflected a 34% improvement among novice and low-skilled workers, with a more modest impact among experienced and highly skilled employees. AI improved the quality and efficiency of customer service relatively quickly, which could be replicated across other occupations. Finally, multifactor productivity growth measures the degree to which labour and capital inputs are used more efficiently in the production process. Growth here is associated with organizational change and technological progress. Examples of the former include the shift to big-box stores boosting productivity in the retail industry, while the hub-and-spoke model did the same for the transportation sector. In terms of technological progress, the largest and longest-lasting lifts to multifactor productivity come from emerging technologies with general purpose technology (GPT) characteristics. GPTs apply to, and are easily adaptable within, multiple industries and sectors. They have the potential to transform economies and societies. Historical examples of GPTs include electricity, the automobile and information technology (such as computers and the Internet). (And, in case you’re wondering, the acronym GPT in ChatGPT stands for something else… Generative Pre-trained Transformer.) What sets AI apart from other GPTs is its self-learning ability, which implies a faster rate of improvement of the technology itself. AI, a derivative of IT, is being touted as a stand-alone GPT because of its role as a ‘method of invention’, packing a pervasive two-pronged productivity punch. First, there should be a hefty (and perhaps long-lived) flow of one-off productivity gains as AI is adapted to the vast array of production processes for both goods and services. Limited studies of generative AI (GenAI) already show 35% to 55% improvement in outcomes, such as the number of coding tasks completed in software development, and the speed and quality of professional writing tasks. In these cases, AI is augmenting labour, but in other cases the tasks will be completely automated. Second, GenAI can enhance research and development and result in more innovation among enterprises and industries. New drug discoveries in the pharmaceutical industry are an often-cited example. For the global economy, McKinsey (June 2023) estimated that GenAI and other automation technologies could boost productivity growth by anywhere from 0.5-to-3.4 ppts per annum out to 2040, with the upper range of outcomes dictated by the degree in which individual economies are early adopters. GenAI alone might contribute 0.1-to-0.5 ppts per annum. The study also looked at nine separate business functions within 21 industries and concluded that GenAI has the potential to generate $2.6 trillion to $4.4 trillion in value across industries. Part of this value is the productivity benefit from automation. In the McKinsey study and others, there is a simplifying assumption made: that workers displaced by GenAI and other automation technologies shift to different occupations or keep the same ones but with different tasks, and that their productivity levels stay comparable to before. There are no noticeable employment or productivity losses offsetting the gains. This is an unrealistic assumption given the experience with past labour market disruptions, such as globalization. Therefore, estimates of AI-led productivity gains need to be taken with caution. In fact, more pessimistic studies, such as MIT’s Acemoglu (April 2024), derive very modest productivity advances from AI. The earlier-mentioned Deloitte survey found that only 8% of private companies were seeing increases in productivity from AI, though four times as many firms (32%) were expecting to see gains in the next year and 87% within the next three years. In the next section, we turn to the potential disruptive and beneficial effects of AI on workers. AI and EmploymentBeing a general purpose technology, AI will change the way many people work across multiple industries. And, like earlier GPTs, AI should eventually lead to new products and industries, and new jobs. Moreover, it promises to enhance human intelligence and creativity, as well as replace repetitive tasks, freeing up time for more rewarding work. Best of all, anyone with a browser or app can use some AI systems simply through prompts in their natural language. This ease of use applies to generative AI, which can be used to create content, whether text, images, audio, video, and code. Popularized by the introduction of ChatGPT in 2022, it can recognize patterns from reams of data to form ideas and recommend actions. It can summarize millions of documents and conversations, distilling key messages for managers to improve customer service and the bottom line. It is increasingly used to write reports, create images and presentations, and help workers learn new tasks. AI has the potential to change the work environment for both white- and blue-collar workers. In the office, GenAI will provide advice to knowledge workers. While displacing many routine tasks, it will free-up time for more value-added work. The introduction of the typewriter and computer replaced some low-level office positions, but new jobs requiring other skills emerged. In health care, AI can read scans faster than radiologists and suggest diagnoses based on lab results. AI “assistants” compile notes, allowing doctors to spend more face-time with patients. Meantime, AI systems are the brains behind new automation, including the latest generation of powerful, flexible, and mobile robots that can perform multiple tasks on the factory and warehouse floor. Still, as with previous GPTs, AI’s full impact on the labour market could be many years off. This is because of the J-curve effect, whereby business processes take time to adapt to new technologies before transformational effects arise. Businesses today are only scratching the AI surface. A Census Bureau survey found that about 5% of U.S. companies used AI earlier this year, though another 7% intended to adopt it in the next six months. Usage might be a little higher north of the border, as a recent Statistics Canada survey found that almost 14% of businesses were using generative AI or have plans to use it. The relatively low rate of AI adoption could explain why it hasn’t led to mass layoffs. In fact, due to the substantial funds tech companies spend on development, AI systems may be generating more jobs than displacing. While Census data show that 2.6% of U.S. employers shed jobs due to GenAI between July 2023 and February 2024, a larger 2.8% reported adding positions. Although AI threatens to replace some duties, most jobs involve many tasks that AI can’t yet handle. Most jobs are complex, requiring an eclectic mix of skills, including intelligence, creativity, mobility, flexibility, and the ability to work well with people. Most humans have these qualities; AI and robots don’t and may not for decades to come. While portions of a job can be performed by AI, the job itself may still exist as new duties replace old ones. As well, in its current form, GenAI output can be unreliable, spewing out flawed information, such that human oversight and guidance are still needed to achieve optimal results. Moreover, no machine will ever replace the social element. For example, millions of chess fans continue to watch the world’s top grandmasters compete among themselves, even though they are easily trounced by AI programs. Considerable research has gone into quantifying AI’s impact on employment—no easy task given the labour saving and enhancing qualities of the technology. McKinsey Global Institute (July 2023) estimates that, by 2030, automation could perform tasks accounting for almost 30% of current U.S. work hours, with generative AI driving 8 ppts of this total. McKinsey believes GenAI will enhance rather than displace most jobs, though this may not apply to customer service and office support. The financial industry has a relatively high exposure to AI. Using data on job descriptions, Accenture (February 2024) estimates that 73% of the tasks performed by U.S. bank workers will either be augmented or replaced by generative AI. |
Elsewhere, the IMF (January 2024) looked at the AI exposure and complementarity of occupations to assess which jobs are susceptible to displacement versus augmentation. It estimated that almost 40% of jobs globally and about 60% in advanced economies have exposure to AI. Roughly 60% of current U.S. jobs are at high exposure to AI, split about evenly between high- and low-complementarity. It believes that the productivity gains from AI should lead to income growth for most workers, though more for upper-income earners. While occupations such as judges and lawyers have high exposure to AI, they also have high complementarity and should benefit from enhanced efficiencies. The previously-cited Acemoglu study also found a tendency for AI to widen the income gap. However, the Stanford/MIT study on customer service agents found that less experienced workers had much greater productivity improvements from AI systems than senior workers, suggesting scope to narrow income inequalities. |
|
Final ThoughtsArtificial intelligence’s impact on productivity growth will no doubt be positive, but recent studies are mixed on size and timing. Despite growing use, AI has not led to a surge in U.S. investment in information processing equipment and software, an ingredient for productivity growth, unlike during the tech boom of the late 1990s. But regardless of its impact on productivity, AI will surely have a material effect on employment, possibly for at least half of the work force, via a combination of augmentation and displacement. Knowledge and factory workers will be more impacted than persons working with their hands or caring for people. Wages could come under pressure in jobs at risk of displacement. But if the productivity gains are sufficiently large, wages for most workers should increase. As with earlier transformative technologies, more highly skilled and paying jobs could ultimately materialize—at least while AI systems are smart enough to assist workers but not smart enough to replace them. New technologies tend to create new wants and new tasks. If the productivity and innovation effects of AI dominate the displacement effect, the economy could well see many years of sturdier (and less inflation-prone) growth, albeit at some still-unknown time in the future. |