When asked how long artificial intelligence (AI) will take to boost the US economy, ChatGPT's response was mostly noncommittal: "It's difficult to predict accurately
Generative AI tools that can create new content such as text and images, such as ChatGPT under OpenAI, are becoming increasingly popular among consumers and companies. In light of this, some economists predict that this technology will bring significant changes to the workplace and increase productivity (hourly output), lifting it out of its long-standing slump and stimulating economic growth.
But such predictions are based on the theoretical potential of the technology. The actual effect may be small, but there is also a possibility of slower effectiveness, depending on several factors. One is the degree to which this technology is widely adopted and the level of proficiency in its use by people. Another factor is to what extent the company can translate this technology into higher productivity once it is adopted.
One of the more optimistic estimates comes from Goldman Sachs. Goldman Sachs economists believe that generative AI may increase the average annual growth rate of productivity in the United States by 1.5 percentage points in the 10 years after its widespread application. This will result in an average annual productivity growth rate of about twice that of the end of 2007. They said that this will also have a similar upward effect on the growth of US gross domestic product (GDP) over the same time span. For example, officials from the Federal Reserve expect the long-term economic growth rate of the United States to be 1.8%, so theoretically this can increase the growth rate to 3.3%.
But the prerequisite cannot be ignored: this is after AI was widely adopted. Joseph Briggs, senior economist at Goldman Sachs, emphasized that it is difficult to predict when AI will be widely used. Briggs said that due to the uncertainty in the timeline for companies to adopt AI and the ultimate ability of AI, the increase in productivity growth by AI may be as low as 0.3 percentage points or as high as 2.9 percentage points.
Breakthrough first, then application
He said that there is usually a time difference between technological breakthroughs and widespread adoption, and AI may have broad macroeconomic impacts starting in the second half of this decade and the 1930s.
However, Briggs said, "In all the scenarios we come up with, given the trend rate of productivity growth, the increase in productivity will be significant enough to produce an economically significant outcome
According to data from the US Department of Labor, productivity increased by an average of 1.4% annually from the fourth quarter of 2007 to the end of the previous business cycle in 2019. This number is lower than the long-term growth rate of 2.1% from the first quarter of 1947 to the second quarter of 2023.
The improvement of productivity is the key to long-term economic growth, as it means that the same number of workers can produce more goods and services, ultimately improving living standards.
Enlightenment from Electricity and Internet Technology
Previous experience has shown that the driving effect of new technologies on productivity growth may be gradually realized. Economist Paul David pointed out in a 1990 paper that in terms of electricity in the United States, it took decades for half of the factory's mechanical transmission capacity to be electrified, which in turn had an impact on manufacturing productivity growth.
David used the development trajectory of electricity as a historical comparison to explain why in the era when he wrote the above article, despite the rise of computers, productivity improvement was slow.
David pointed out that the implementation of electrification in factories was slow at the beginning because existing manufacturing factories relied on water and steam power to operate normally, and large-scale renovations were not profitable.
Anton Korinek, an economics professor at the University of Virginia, said that compared to the previous productivity boom of the late 1990s and early 21st century when the internet and personal computing emerged, some factors may accelerate the speed at which industries and companies accept generative AI. At that time, people needed to spend money to install physical devices such as routers and internet connections in order to take advantage of these emerging technologies.
Korinek said, "Now, all these connections are ready-made, basically all you need to do is log in to a new website." He estimates that compared to the trend level of the past 20 years, generative AI will increase productivity by 10% -20% in the next 10 years.
However, economists say that it may take time for the effectiveness of AI to truly be reflected in productivity indicators, as businesses and workers need to learn how to use AI and integrate it into their work.
In fact, in a survey of economists conducted by The Wall Street Journal in April this year, 61% of respondents said they expect AI tools such as ChatGPT to have only a small positive impact on US GDP growth over the next five years. Another 29% of people stated that they do not expect AI tools to have any impact during this period.
From the history of the adoption of new technologies, it is not enough to simply adopt technology in order for them to change production methods. Other complementary assets must also be adopted, "said Robert Seamans, a professor of management and organization at New York University. This is costly and takes time
Seamans stated that companies particularly need to hire or train specialized personnel who understand both generative AI capabilities and existing production processes, in order to connect the two.
The internet developed from the US Department of Defense in the 1960s, but it wasn't until the 1990s that it truly took off with the popularity of personal computers, and later with broadband access and smartphones, it had the most disruptive impact.
Some industries seem to have developed to be particularly suitable for addressing all of these issues. For example, economists Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found in a 2023 work paper that using an AI assistant resulted in an average improvement of 14% in the work efficiency of call center staff, measured by the number of problems solved per hour. This AI tool provides real-time guidance for customer service personnel in call centers on how to give the best answer in conversations, and research has found that the productivity improvement of new and low-skilled employees is particularly significant.
Uneven adoption and uneven effectiveness
Goldman Sachs' analysis estimates that AI can automate about a quarter of work tasks in the United States, with AI being particularly prevalent in the administrative and legal industries. According to Goldman Sachs, AI adoption rates are relatively low in jobs with high physical requirements such as construction, maintenance, and repair.
But Brad Hershbein, a senior economist at the employment research firm UpJohn Institute, said that as machines replace previously done jobs by humans, there may not be a net decrease in employment in the economy.
He said, "It is much more common that new technologies will ultimately change tasks in existing work, rather than completely eliminating them
Historians point out that new technologies such as spreadsheets often create more job opportunities than they destroy.
However, Hershbein stated that if some companies can act faster than others, the imbalance in AI adoption may lead to job loss and exacerbate income inequality and social unrest.
AI adopters may be able to seize market share from non adopters, "he said. If this technology is promoted equally everywhere, and everyone receives training and understands how to use it to solve their problems, then we may experience a significant increase in productivity relatively quickly. However, this is not the case