Michelle ThompsonAnton Korinek, an economics professor at the University of Virginia, advises his students to prioritize the mastery of an emerging and transformative technology in the field of economics. This technology is generative artificial intelligence, often referred to as “genAI.”
Economists already employ machine learning, a subset of AI, for data analysis and economic forecasting. However, genAI is a distinct technology that forms the foundation of tools like ChatGPT, and it has rapidly advanced in recent months.
Korinek anticipates that it will lead to a significant transformation in research, as indicated by a paper he authored, which has been accepted for publication in the Journal of Economic Literature.
“GenAI is a powerful technology that, when utilized, enhances our ability to address societal economic challenges in a more effective and productive manner. This is the essence of research,” stated Korinek in an interview with CNN.
Moreover, genAI doesn’t just contribute to research; it has demonstrated its usefulness in the realm of economics education. A recent study by economists at George Mason University highlighted its potential for solving specific models in the classroom and generating exams. Additionally, a separate study by researchers at the Federal Reserve Bank of St. Louis indicated that their AI model comprehends requests for inflation forecasts and offers a cost-effective and accurate alternative to conventional forecasting methods.
In summary, genAI is more likely to empower economists rather than replace them, at least for the time being.
A Potent Instrument for Conducting Research.
Economists engage in numerous small tasks during their research efforts, and Korinek’s paper asserts that large language models, a specific type of genAI, can provide valuable assistance in six key areas: “concept generation and feedback, written content creation, background research, data analysis, programming, and mathematical derivations.”
Among the most widely employed genAI tools are OpenAI’s ChatGPT, Microsoft’s New Bing, Google’s Bard, Anthropic’s Claude 2, and Meta’s LlaMA 2, as noted by Korinek. Any of these chatbots can support economists in brainstorming research ideas simply by requesting a list of potential topics. Furthermore, they can evaluate research plans by presenting the pros and cons of each option.
GenAI excels in copy editing and refining written content, including error detection, title suggestions, and even generating text tailored for social media to promote research papers. This innovative technology can significantly enhance the clarity, specificity, and overall coherence of a researcher’s writing, as described in Korinek’s paper.
AI chatbots also excel at summarizing text. Both versions of ChatGPT (3.5 and 4) are capable of summarizing text passages of up to approximately 3,000 words. Claude 2, however, stands out by summarizing texts of up to around 75,000 words, covering the length of most academic research papers, as per Korinek’s observations. Economists can query the chatbots about specific papers, asking questions like “What are the primary conclusions drawn by the author?” or “What evidence supports these conclusions?”
Economic research frequently involves technical tasks such as coding and developing mathematical proofs. GenAI tools, such as ChatGPT Advanced Data Analysis, are particularly useful for tasks like coding, explanation, translation, and debugging, especially in languages like Python and R. Chatbots can be employed to set up economic models, derive equations, and provide explanations, though Korinek does note that genAI’s mathematical capabilities are currently somewhat limited.
It’s worth mentioning that the most recent iteration of each chatbot, like ChatGPT-4, boasts improved capabilities compared to their previous versions.
However, it’s essential to acknowledge that genAI is not infallible. It can occasionally generate incorrect information, which is sometimes referred to as “hallucinations.”
An Educational Aide in An Instructional Role.
Economics professors Tyler Cowen and Alex Tabarrok from George Mason University released a paper earlier this year titled “How to Learn and Teach Economics with Large Language Models, Including GPT.” In this paper, they elaborated on how genAI can summarize text, enhance writing, offer ideas, and resolve basic economic models with explanations, mirroring the insights presented in Korinek’s paper.
Furthermore, their research demonstrated that genAI is especially beneficial within an educational setting.
In their paper, Cowen and Tabarrok mentioned, “ChatGPT and Bing Chat can also generate highly credible syllabi for various courses, encompassing readings, course policies, and grading procedures. Just to clarify, Chat GPT allows for late assignments with a 10% penalty per day.”
They also pointed out that genAI tools, while not yet suited for tackling complex PhD-level problems, are adept at resolving undergraduate-level models and are valuable for educational purposes with students.
Predicting or Projecting Future Events or Trends.
A recent working paper indicates that genAI demonstrates considerable skill in predicting inflation, possibly outperforming current economists in this regard.
The paper, authored by two policy advisers at the St. Louis Federal Reserve, conducted a comparison between inflation forecasts generated by Google’s PaLM, a substantial language model chatbot akin to ChatGPT, and one of the primary sources of macroeconomic predictions: the Survey of Professional Forecasters. This group comprises experienced experts in macroeconomic forecasting, possessing advanced degrees in economics or related fields, as confirmed by a spokesperson.
The research revealed that the inflation forecasts generated by PaLM contained fewer errors compared to those from the Survey of Professional Forecasters (SPF), which compiles predictions from numerous economists and financial analysts.
The paper stated, “These results indicate that Large Language Models (LLMs) may offer a cost-effective and precise alternative method for producing inflation forecasts.”
Effect on Job Opportunities or the Labor Market.
The evolution of genAI in economics is expected to have a primarily limited initial impact on employment. Initially, it will likely enhance the productivity and efficiency of economists, but over time, it could potentially result in some job displacement.
Indeed, a job listing website, conducted a recent study assessing the vulnerability of specific occupations to genAI based on the skills required for those roles. The study identified that software development jobs, including software engineers, are among the occupations with the highest exposure to potential changes due to genAI.
Svenja Gudell, Indeed’s chief economist, believes that genAI has the potential to create better job opportunities in the long run by eliminating tasks that people typically dislike. However, she acknowledges that reaching this stable state might entail a period of transition that could be tumultuous and challenging.
Gudell pointed out that companies may make different choices when implementing genAI. They could reduce labor costs by laying off employees if genAI maintains their usual output level. Alternatively, they may choose to retain their workforce and leverage genAI to enhance productivity. In either case, the technology introduces employment-related risks.
While economists rely on technology for many aspects of their work, tasks that genAI can increasingly handle as it becomes more advanced, Gudell asserts that being an economist will always necessitate a “human touch.” This includes activities such as teaching students, conducting live presentations in front of an audience, and even when collaborating with genAI.
“In economics, we often talk about the interpretability of a model. It’s one thing to obtain an answer, but understanding how you arrived at that answer can be just as crucial. Then, the human element comes into play when it’s time to contextualize and apply that result,” Gudell explained.
