Generative AI: a gamechanger for productivity growth?
Artificial intelligence, or AI, has been a key theme driving financial markets this year. For example, the share price of Nvidia Corporation, which makes graphics processing units (GPUs) that are used in AI applications, has more than doubled since ChatGPT, a generative AI chatbot, was released in November of last year1. ChatGPT has become the fastest-growing consumer application in history, after notching up 100 million monthly active users during the first two months of its release2.
There has been plenty of discussion about the potential advantages and disadvantages that widespread adoption of AI could bring to society. However, the aspect that has excited economists is its potential impact on productivity growth. This is because improvements in productivity are key in driving increases in real wage growth, and thereby living standards. At the macro level, increases in productivity are a key factor in determining how fast an economy can grow. After all, over the long-term, an economy’s potential growth rate is determined by changes in the number of available workers (i.e. the labour force) and the amount of goods and services that each worker can produce (i.e. productivity).
Artificial intelligence offers up the prospect of improving efficiency and productivity in a number of areas. In healthcare, AI can assist in medical diagnostics by analyzing medical images and patient data. In manufacturing, AI-driven automation can lead to improved precision, reduced errors, and increased production speed. In addition, AI can analyze data from machinery and equipment to predict when maintenance is needed, thereby minimising downtime and costly breakdowns. In research and development, processes can be accelerated by simulating experiments, analyzing data, and identifying patterns. In computing, AI can help developers write code faster and more accurately. In the legal profession, AI can assist in research by quickly sifting through vast databases of case law, statutes, and regulation. These are just a few areas where AI can improve efficiency, but the possibilities are seemingly endless.
Just how much the adoption of generative AI will boost productivity is open to debate. Measuring productivity gains, particularly in the services sector which tends to dominate advanced economies, is difficult. However, early evidence suggests the potential gains are large. A recent study of more than 5,000 customer service workers at a big tech company concluded that productivity (measured by issue resolution) increased by an average of 14%, while the improvement amongst novice and low-skilled workers was even more pronounced, at 35%3.
In addition to ‘one-off’ improvements in efficiency, there could be a more sustained impact on productivity growth. Many observers expect that, by automating repetitive and routine tasks, AI will free up the time of knowledge workers, enabling them to focus on the more creative aspects of their work. In turn, this could lead to greater innovation and a faster pace of technological progress, which could ultimately lead to a prolonged upshift in the rate of productivity growth.
Quantifying the impact
From a country-by-country perspective, the impact on productivity growth of generative AI will depend in part on the industry composition of a country. Advanced economies (which have a higher share of employment in administrative, legal and professional roles where generative AI could make big inroads) are arguably more likely to see greater benefit than developing economies (where more employees work in manual, physically intensive professions such as manufacturing, agriculture and construction).
Given the high degree of uncertainty, precise estimates of the extent to which AI could boost productivity growth should be taken with a large pinch of salt. Goldman Sachs reckon that widespread adoption of generative AI could increase labour productivity growth in the US by around 1.5 percentage points annually over a 10-year period, a rise that would effectively double the rate of productivity growth witnessed in recent years. At the global level, Goldman see generative AI lifting annual productivity growth by around one percentage point over the same timescale4.
The consultancy Mckinsey appears somewhat less optimistic, and estimate that AI could contribute 0.1 to 0.6 percentage points to global productivity growth depending on how fast the technology is adopted5. Moreover, this forecast is based on the assumption that individual workers affected by the technology shift to other work activities that at least match their 2022 productivity levels. This is a key consideration; if workers displaced by AI move to jobs where they are less productive, the net improvement in productivity will be diminished.
Pace of adoption
The speed at which the technology is adopted will clearly affect the impact of AI on productivity growth. Some factors would seem to argue for a relatively quick pace of adoption. The widespread use of computers, which culminated in increased rates of productivity growth during the 1990s required heavy investment in physical hardware and largescale organizational changes during the preceding decades. In contrast, generative AI is primarily about software, which is arguably easier, quicker and less costly for businesses and other organisations to adopt.
Moreover, because generative AI is relatively easy to use (chatbots like ChatGPT understand text in a similar way to human beings), it might require relatively little staff training to utilise effectively. The integration of AI tools into existing software (such as Microsoft 365, for example6) could also accelerate the pace at which companies adopt AI.
However, there will be barriers to widespread AI adoption. Some organisations might simply have a mindset that fails to see how AI could be beneficial, while others will lack a strategy for fully integrating AI into their business models. In some sectors, unions are likely to mount stiff opposition to the adoption of AI, fearful that it could result in workers losing their jobs. Regulatory restrictions, and uncertainty regarding future regulations, could also make businesses wary of using AI, as could concerns over data security.
Against this backdrop, some observers worry that the rollout of generative AI will not be sufficiently broad to make a major impact at the macro level in the near term. In a recent survey of 1000 small businesses in the United States and Canada, two thirds of respondents indicated that they will try generative AI for work purposes within the next 12 months7. However, this leaves a third of companies with no plans to even experiment with the new technology.
Indeed, there are tentative signs that the take-up of some popular AI tools is already faltering. According to internet data firm Similarweb, traffic to ChatGPT’s website dropped 9.7% in June from a month earlier8. This could just be because of seasonal patterns, but it might be the result of users becoming more aware of the technology’s limitations. Large language models such as ChatGPT regularly churn out false information (so-called ‘hallucinations’) and some reports suggest that the accuracy of responses has actually got worse in recent months9. If the quality of results that generative AI produces is deteriorating, possibly as a result of providers trying to reduce running costs, then this could clearly have a negative impact on take-up rates.
A potential gamechanger
All of this suggests generative AI is a potential gamechanger regarding productivity growth, but that widespread adoption will take time and will face several significant hurdles.
Clearly, the gains from AI will not be distributed evenly, and governments will have to implement policies to mitigate the job-destroying potential of AI, along with other downside risks. Some jobs will no doubt be made obsolete by AI, but the experience of earlier technological revolutions suggests that new jobs will also be created. One oft-cited study found that 63% of jobs done in 2018 did not exist in 194010 (think computer programmers, wind turbine technicians, mental health counselors etc.). Looking to the future, if workers displaced by AI move into jobs which have lower productivity rates and less scope for efficiency gains (e.g. ‘people-facing’ professions such as therapists and care workers, or ‘physical’ jobs such as plumbers and hairdressers), this would diminish the net productivity gain from AI adoption.
Nevertheless, at a time when ageing populations are creating labour shortages in many advanced economies, generative AI still holds out the possibility of an offsetting increase in productivity, which could ultimately lead to stronger rates of economic growth and, with the right safeguards in place, improved living standards.
15th August 2023