On May 10, 2023, the Brookings Institution released the report "Artificial Intelligence Drives Rapid Productivity Improvement". The report notes that big language models such as ChatGPT are becoming powerful tools that not only increase worker productivity, but also accelerate innovation and lay the foundation for economic growth. This rapid progress can bring great benefits, but it can also carry significant risks, so ensuring that we can steer AI in a direction that benefits society as a whole is critical.
In the next few years, millions of workers, including doctors, lawyers, managers and salespeople, will undergo an epochal shift to significantly increase their productivity with artificial intelligence. The potential of the latest generation of AI systems can be vividly demonstrated by the widespread spread of ChatGPT, a large language model that has captured the public's attention by generating coherent and contextually appropriate text. This is not an unsung innovation, and its features have already attracted hundreds of millions of users.
Other big language models recently launched publicly include Google's Bard and AI startup Anthropic's Claude. However, generative AI is not limited to text: in recent years, we have also witnessed advances in generative AI systems in creating images, such as Midjourney, Stable Diffusion or DALL-E, and more recently multimodal systems that combine text, images, video, audio, and even robotic capabilities. These techniques are basic models, that is, huge systems based on deep neural networks, trained on large amounts of data, which can be adapted to a variety of different tasks. Since information and knowledge jobs dominate the U.S. economy, these smart machines will greatly increase overall productivity.
01
Assessment of the impact of AI on productivity
Recent advances in generative AI are driven by software, hardware, data collection, and continuous investment in cutting-edge models. Sevilla et al. observed that computing resources used to train cutting-edge AI systems have doubled every six months over the past decade. At the same time, the capabilities of generative AI systems are growing, enabling them to perform many tasks that were previously only possible for cognitive workers, such as writing elegant sentences, generating computer code, summarizing articles, brainstorming, organizing plans, translating other languages, writing complex emails, and more.
Generative AI has a wide range of applications that will impact workers, occupations, and activities across a wide range of industries. Unlike most automation advances in the past, it is an intelligent system that affects cognitive work. As a recent research paper points out, in some form, big language models could affect 80 percent of the U.S. workforce.
Recent research literature estimates the productivity impact of AI on specific occupations or tasks. Kalliamvakou found that using a tool called Codex, based on a previous version of the large language model GPT-3, software engineers could code twice as fast. Noy and Zhang found that many writing tasks can also be completed twice as quickly, while Korinek estimates that economists using large language models can increase productivity by 10-20 percent, based on 25 use cases for language models.
But can the gains in these specific tasks translate into significant gains in real-world scenarios? The answer seems to be yes. Research by Brynjolfsson, Li and Raymond shows that call center operators have increased productivity by 14 percent after using the technology, and even more than 30 percent for the least experienced workers. In addition, customers are more motivated when interacting with operators that use generative AI as an aid, and perhaps as a result, employee turnover is lower. The system seems to create value by capturing and communicating tacit knowledge about how to solve problems and delight customers that used to only be learned from on-the-job experience.
Most cognitive work involves taking past knowledge and experience and applying it to current problems. Generative AI programs may have certain types of errors, but the form of these errors is predictable. For example, language models tend to produce "hallucinations," i.e. fictional facts and quotes, and as such, they require human supervision. However, their economic value does not depend on their perfection, but on whether they can function effectively. From this perspective, they already have the potential to make a huge impact. In addition, the accuracy of generative AI models is rapidly improving.
02
Quantitative assessment of productivity effects
A recent Goldman Sachs report suggests that generative AI could increase global GDP by 7%, a significant effect for any single technology. Based on our analysis of various use cases and the proportion of the workforce engaged in primary cognitive work, we believe this estimate is reasonable, although there is still a great deal of uncertainty about the ultimate impact of AI on productivity and growth effects.
There are two ways generative AI can increase productivity.
First, generative AI improves output efficiency and increases productivity. By increasing the efficiency of workers engaged in cognitive work, the level of output also increases. Economic theory tells us that in a competitive market, the impact of productivity gains in a sector on overall productivity and output is equal to the size of productivity gains multiplied by the size of that sector. For example, if generative AI increases the productivity of cognitive workers by an average of 30 percent over a period of ten or twenty years, and cognitive work accounts for 60 percent of all value added in the economy (measured by the total wages paid for cognitive tasks), overall productivity and output will increase by 18 percent in those years.
Second, generative AI accelerates innovation to drive future productivity growth. Cognitive workers not only produce current outputs, but also make new inventions and discoveries, and produce technological advances that drive future productivity growth. This includes research and development, the work of scientists, and, more importantly, the work of managers, that is, the promotion of new innovations to the productive activities of the entire economy. If cognitive workers become more efficient, they will accelerate technological advances that will increase the rate of productivity growth, and the effect will continue. For example, if productivity growth is 2 percent and the productivity of the cognitive workforce that underpins productivity growth increases by 20 percent, productivity growth will increase by 20 percent to 2.4 percent. During a given year, this change is almost imperceptible and is usually masked by cyclical fluctuations.
However, productivity growth has a compounding effect. After a decade, small productivity gains would increase the size of the economy by 5 percent, and growth would accumulate further in each subsequent year. In addition, if this acceleration is applied to the growth rate of the growth rate (for example, if one application of AI is to improve the AI itself), then the growth will accelerate over time.
03
Barriers and drivers of AI adoption
To achieve productivity growth, advances in AI must be disseminated throughout the economy. Traditionally, this has always taken time, so we wouldn't expect potential productivity gains to be immediately apparent. These advances need to be adopted and scaled up by businesses and organizations that employ a cognitive workforce across the economy, including some small and medium-sized enterprises that may be slow to recognize the potential to adapt to advanced new technologies or lack the skills needed to use them. For example, Goldman Sachs reports that it will take 10 years to fully realize these gains.
Economic theory holds that new technologies can only bring productivity gains after a period of investment in complementary intangible assets, such as business processes and new skills. As a result, early general-purpose technologies, such as electricity and computers, took decades to have a significant impact on productivity. Other barriers to adoption and promotion include concerns about job losses, institutional inertia and regulatory issues in everything from medicine to finance and law.
However, there are factors in generative AI that can mitigate these barriers and even accelerate adoption. First, one advantage of cognitive automation over physical automation is that it can often be quickly rolled out through software, especially in the digital infrastructure internet that is currently ubiquitous. Anyone with an internet connection was able to access ChatGPT and didn't require any hardware investment from users, making it the fastest release in history, gaining 100 million users in just two months. Both Microsoft and Google are rolling out generative AI tools as part of their search engines and office suites, providing access to generative AI for large segments of the cognitive workforce in advanced countries that regularly use these tools. Second, more and more application programming interfaces (APIs) are being used to enable seamless modularity and connectivity between systems, and the market for plugins and extensions is growing rapidly, making it easier to add functionality. Finally, users of generative AI can interact with the technology using natural language instead of using special code or commands compared to other technologies, which makes it easier to learn and adopt these tools.
These positive factors suggest that the adoption of new technologies may be faster than in the past. However, the importance of emphasizing adequate training to maximize the use of these tools is even more self-evident.
Disclaimer: This article is transferred from Meta Strategy, original author Mark. The content of the article is the personal views of the original author, this public account compilation/reprint is only to share, convey different views, if you have any objections, welcome to contact us!
Transferred from 丨 Meta Strategy
Author丨Mark
About the Institute
Founded in November 1985, the International Institute of Technology and Economics (IITE) is a non-profit research institution affiliated to the Development Research Center of the State Council, whose main functions are to study major policy, strategic and forward-looking issues in the development of the mainland's economy, science and technology, and economy, track and analyze the development trend of world science and technology and economy, and provide decision-making consulting services for the central government and relevant ministries and commissions. "Global Technology Map" is the official WeChat account of the International Institute of Technology and Economics, dedicated to conveying cutting-edge technology information and scientific and technological innovation insights to the public.
Address: Building A, Building 20, Xiaonanzhuang, Haidian District, Beijing
Tel: 010-82635522
WeChat: iite_er