Economics professor Anton Korinek at the University of Virginia advises his students to focus on mastering a rapidly advancing technology called generative artificial intelligence, or "genAI" for short. Unlike machine learning, which economists already employ for data analysis and economic projections, genAI powers tools like ChatGPT and has made significant progress in recent months.
According to a paper accepted for publication in the Journal of Economic Literature, Korinek anticipates that it will "revolutionize research." In an interview with CNN, he stated that this technology is powerful and can effectively address the economic challenges we encounter as a society, enhancing productivity and offering improved solutions. He emphasized that this is the essence of research.
In addition to assisting with research projects, genAI has demonstrated its potential in the field of economics. Recent studies conducted by economists at George Mason University have revealed that genAI can be effectively utilized for teaching economics. It can solve specific models in classrooms and even create exams. Furthermore, researchers at the Federal Reserve Bank of St. Louis have developed an AI model that comprehends and generates inflation forecasts. This model presents a cost-effective and accurate alternative to traditional forecasting methods. Hence, genAI is currently more likely to enhance the capabilities of economists rather than replace them.
An effective research tool
The research conducted by economists involves various minor activities, and according to Korinek's study, large language models, a specific category of genAI, can provide assistance in six distinct ways: "ideation and feedback, writing, background research, data analysis, coding, and mathematical derivations."
According to Korinek, the popular genAI tools used include OpenAI's ChatGPT, Microsoft's New Bing, Google's Bard, Anthropics Claude 2, and Metas LlaMA 2. These chatbots can assist economists in brainstorming research ideas by requesting a list of suggestions. Additionally, they can assess research plans by offering their respective advantages and disadvantages. GenAI excels in copy editing and refining writing, such as identifying typographical errors, proposing suitable titles, and even generating text for social media promotion of academic papers. Korinek's study states that this new technology aids in enhancing a researcher's writing, making it more lucid, precise, and coherent.
Several ChatGPT logos can be seen on the monitor of a laptop.
Hannes P Albert/picture-alliance/dpa/AP
How companies are embracing generative AI for employees...or not
AI chatbots excel in summarizing text. ChatGPT versions 3.5 and 4 can summarize passages of up to 3,000 words, while Claude 2 can handle up to about 75,000 words, making it suitable for academic research papers. Users can ask the chatbot specific questions about a paper, such as the authors' main conclusions or the specific evidence supporting their points.
Economic research often involves technical tasks like coding and creating mathematical proofs. GenAI tools like ChatGPT Advanced Data Analysis are helpful in writing, explaining, translating, and even debugging code, particularly in languages like Python and R. Chatbots can assist in setting up economic models, deriving equations, and providing explanations. However, it is worth noting that the chatbot's mathematical capabilities are currently somewhat limited.
The latest version of each chatbot, like ChatGPT-4, possesses enhanced capabilities compared to earlier versions. Moreover, genAI is not infallible. It occasionally generates inaccurate information, commonly known as "hallucinations."
A teaching assistant
Economics professors Tyler Cowen and Alex Tabarrok from George Mason University recently released a paper titled "How to Enhance the Learning and Teaching of Economics using Advanced Language Models, such as GPT." In this publication, they elaborate on how genAI, an artificial intelligence system, can effectively summarize text, enhance writing, offer valuable ideas, and provide explanations for basic economic models, much like the paper by Korinek.But the paper also showed that genAI is particularly useful in the classroom.
From left to right: writers John Grisham, George R.R. Martin, Jodi Picoult, Michael Connelly.
Charly Triballeau/Jon Kopaloff/Jeff Pachoud/AFP/Getty Images/Evan Agostini/Invision/AP
George R. R. Martin, Jodi Picoult, and other renowned authors have united with the Authors Guild to file a class action lawsuit against OpenAI. In their paper, Cowen and Tabarrok mentioned that ChatGPT and Bing Chat will generate highly credible syllabi for numerous courses, encompassing readings, course policies, and grading procedures. It's worth noting that ChatGPT accommodates late assignments with a 10% penalty per day.
The economists noted that genAI tools "are not yet ready to solve PhD problems but they are good at solving undergraduate models and for teaching students."
Forecasting
GenAI has been found to be more skilled in predicting inflation than economists who currently undertake this task, according to a recent working paper.
Authored by two policy advisers at the St. Louis Fed, the paper compared inflation forecasts from Googles PaLM, a chatbot with advanced language capabilities similar to ChatGPT, with predictions from the Survey of Professional Forecasters, a widely respected source of macroeconomic predictions. The Survey of Professional Forecasters comprises experienced professionals with advanced degrees in economics and related fields, as confirmed by a spokesperson.
In Washington, DC on October 2, 2023, Satya Nadella, the CEO of Microsoft, attends federal court for the antitrust trial questioning whether Google, a subsidiary of Alphabet Inc., holds a monopoly in the online search industry. The trial is anticipated to continue until November.
Drew Angerer/Getty Images
Microsoft CEO warns of 'nightmare' future for AI if Google's search dominance continues
The paper stated that the findings indicate LLM models can serve as a cost-effective and precise alternative for generating inflation forecasts.
Impact on employment
What could be the implications of the advancement of genAIs on employment within the field of economics? Initially, its impact is expected to be restricted, primarily assisting economists in enhancing their productivity and efficiency. However, there is a possibility that it could eventually result in some job reductions.
Jobs platform Indeed recently conducted a study to assess the extent of exposure of specific occupations to genAI, taking into consideration the required skill set. The study revealed that software development roles, including software engineers, are subject to the greatest amount of exposure.
Masayoshi Son, chairman and chief executive officer of SoftBank Group Corp., during the SoftBank World event in Tokyo, Japan, on Wednesday, October 4, 2023.
Toru Hanai/Bloomberg/Getty Images
SoftBank's CEO predicts that artificial general intelligence (genAI) will be achieved in the next decade. According to Svenja Gudell, Chief Economist at Indeed, genAI will lead to the creation of more fulfilling jobs by eliminating undesirable tasks. However, the transition towards this steady state may be challenging and fraught with turbulence.
Gudell suggested that a company may choose to reduce labor costs by laying off employees if genAI is able to maintain its current level of productivity. On the other hand, a company could choose to retain all of its employees and benefit from increased output due to genAI. In any case, the introduction of this technology poses a threat to employment.
Economists heavily rely on technology to perform their tasks, many of which genAI could potentially accomplish as it continues to advance. However, Gudell argues that being an economist will always require a "human touch." This encompasses tasks such as teaching students and delivering live presentations to an audience, as well as working alongside genAI itself.
"In economics, the concept of model interpretability is frequently discussed. While obtaining an answer is crucial, comprehending the process behind reaching that answer holds equal significance," explained Gudell. "Subsequently, human intervention becomes necessary to contextualize and apply the obtained result."