Large-scale language models like ChatGPT are becoming powerful tools not only to increase worker productivity, but also to increase the rate of innovation, setting the stage for a dramatic acceleration in economic growth. As a general-purpose technology, AI will impact a wide range of industries, spur investment in new skills, change business processes, and transform the nature of work.
However, official statistics only partially capture productivity gains, because the output of knowledge workers is difficult to measure. Rapid progress can have huge benefits, but it can also come with significant risks, so it's critical to ensure we're moving in a direction that benefits society as a whole.
Imagine a scenario where, on a Friday morning, a man is sitting in his favorite coffee shop, writing a new research paper on how artificial intelligence will affect the labor market. First, he launched ChatGPT, a tool for generative artificial intelligence. After entering a few prompts in plain English, the system was able to provide a suitable economic model, draft the code to run the model, and generate a potential title for the work. By the end of the morning, he had made a week's worth of progress on his research.
We expect that the productivity of millions of knowledge workers, from doctors and lawyers to managers and salespeople, will undergo a similar breakthrough shift within a few years, if not sooner.
ChatGPT, a large-scale language model (LLM), has attracted public attention through its ability to generate coherent and contextual text, which vividly illustrates the potential of the latest generation of artificial intelligence systems. This isn't an innovation languishing in a basement. Its features have attracted hundreds of millions of users.
Other LLMs recently made public include Google's Bard and Anthropic's Claude. But generative AI is not limited to text: in recent years we have also seen generative AI systems that can create images, such as Midtravel, Stable Diffusion, or DALL-E, and more recently combining text, image, video, audio, and even robotic capabilities Multimodal system.
These techniques are fundamental models, vast systems based on deep neural networks, trained on vast amounts of data and adapted to perform a variety of different tasks. As information and knowledge work dominate the U.S. economy, these thinking machines will dramatically increase overall productivity.
The power of productivity growth
The main determinant of our long-term prosperity and well-being is the rate of productivity growth: the amount of output created per hour worked.
This holds true even though the shift in productivity is not immediately felt by everyone and, in the short run, workers' perceptions of the economy are dominated by the business cycle. Labor productivity grew at more than 3 percent a year from World War II until the early 1970s, more than doubling during that time, ushering in an era of prosperity for most Americans.
Productivity growth slowed sharply in the early 1970s, rebounded in the 1990s, but has slowed again since the early 2000s.
As the chart above shows, it breaks down overall growth in labor productivity into two components: total factor productivity (a measure of the impact of technology) and contributions from labor composition and capital intensity. The figure shows that the key driver of changes in labor productivity is changes in total factor productivity (TFP).
There are many reasons for America's recent economic woes, but slow total factor productivity growth is a key one, slowly eroding the country's prosperity, making it harder to fight inflation, eating into workers' wages, and worsening the budget deficit.
The generally slow pace of economic growth, combined with the huge profits of technology companies, has led to skepticism about the benefits of digital technology for the overall economy. However, for about a decade starting in the 1990s, there was a surge in productivity growth, as shown in Figure 1, driven largely by a huge wave of investment in computers and communications, which in turn drove business transformation.
Despite the stock market bubble and massive reallocation of labor and resources, workers are generally better off.
In addition, the federal budget was balanced between 1998 and 2001, a win-win situation. Digital technologies can drive broad economic growth, which happened less than three decades ago.
Early estimates of AI's productivity impact
Recent advances in generative AI are driven by advances in software, hardware, data collection, and increasing investment in cutting-edge models. Sevilla et al. (2022) observed that the amount of computation (computing power) used to train cutting-edge AI systems has doubled every six months over the past decade.
The capabilities of generative artificial intelligence systems have evolved synergistically, enabling them to perform many tasks that have been left to cognitive workers in the past, such as writing elaborate sentences, creating computer code, summarizing articles, brainstorming ideas, organizing plans, translating other languages, writing Complicated emails and more.
Generative AI has broad applications and will affect a wide range of workers, occupations and activities. Unlike most advances in automation in the past, it is a thinking machine that affects cognitive work. As noted in a recent research paper (Eloundou et al., 2023), LLM may affect 80% of the US workforce in some form.
There is an emerging literature that estimates the productivity impact of artificial intelligence on specific occupations or tasks. Kalliamvakou (2022) found that software engineers could code twice as fast using a tool called Codex, based on a previous version of the large-scale language model GPT-3. This is a transformative effect.
Noy and Zhang (2023) found that many writing tasks can also be done twice as fast, and Korinek (2023) estimates based on 25 use cases of language models that economists can be 10-20% more efficient using large language models.