Modern artificial intelligence models are already capable of performing many of the tasks handled by economists at universities, research centers, government agencies, consulting firms, and media newsrooms. They analyze economic data, study central bank decisions, prepare analytical reviews, and even write research papers. According to Harvard economist Jason Furman, in many such tasks AI already surpasses humans in speed and reliability.

This raises a difficult question for economics as a discipline: if machines can perform a significant portion of routine analytical work faster and more cheaply than specialists, what role should economists themselves play?

“Every conversation I have with a colleague from Harvard or a think tank in Washington combines excitement, anxiety, and confusion,” Furman told Axios.

One professor and capabilities that used to require millions

Furman, who served as chief economic adviser to President Barack Obama and now teaches an introductory economics course at Harvard, launched Econvitals — a free website with economic data and analytical tools.

Individually, the solutions presented there do not look revolutionary. What is remarkable, however, is something else: a busy professor was able to create such a resource in his spare time, using the help of several assistants and making active use of Anthropic’s Claude. For additional verification of the results, he also used OpenAI’s Codex.

According to Axios, the project offers large-scale calculations and analysis that previously might have required thousands of person-hours and a budget of several million dollars.

For example, one of the tools analyzes speeches by representatives of the U.S. Federal Reserve and assesses how hawkish or dovish a monetary policy each of them supports. It also makes it possible to track how officials’ views have changed over time.

Specialists at banks and consulting firms have long engaged in this kind of analysis, but traditionally it has required substantial labor and has been provided to clients for a fee.

Econvitals also publishes digital briefings in which AI selects interesting economic studies and analyzes fresh data. In addition, Furman demonstrates examples of academic papers prepared entirely by artificial intelligence on the basis of short prompts.

For now, such texts would hardly secure a young economist an academic career. But given the minimal human involvement, the quality of the results is impressive.

AI handles facts well, but still lags in originality

Furman believes that, with the right workflow, artificial intelligence is already capable of delivering exceptionally high factual accuracy.

“In terms of factual reliability, at this stage they are, if everything is done correctly, better than any human,” the economist argues. In his view, using separate AI agents for fact-checking makes it possible to obtain results that can be fully trusted. He believes that the idea of neural network hallucinations is largely linked to outdated or incompetent methods of using them.

At the same time, Furman draws a distinction between the high-quality execution of analytical tasks and the creation of genuinely new ideas. In the latter, AI still has not matched the best researchers.

The main advantage of neural networks today is the automation of routine work: processing large amounts of data, identifying patterns, preparing reviews, and systematizing results. However, the very notion of routine is constantly changing as more advanced models appear. Tasks that until recently required the involvement of an experienced specialist are gradually becoming amenable to automation.

Furman himself admits that this process gives him mixed feelings. AI expands researchers’ capabilities, but at the same time it is able to perform part of his own work better than he can.

The economist hopes that the availability of new tools will lead not to an avalanche-like increase in the number of publications, but to an improvement in their quality.

“Now any fool can write 300 papers a year, and they will be quite decent and factually accurate. But I hope people will write not more papers, but better papers,” he says.

What will the government choose: economists or AI?

Furman assesses the capabilities of neural networks most radically when it comes to public administration. According to him, if he were heading the U.S. Council of Economic Advisers today, he would rely heavily on artificial intelligence.

“If I were running the Council of Economic Advisers today and could choose between 20 staff members — current PhD holders and future PhDs — and existing AI systems, I would, though with some regret, choose AI,” the economist said.

He adds that people still need to take part in meetings, communicate with journalists, and testify before Congress. But in his assessment, this is only a small part of the agency’s work. Machines are already capable of performing most analytical tasks faster and more reliably than humans.

Such an assessment does not mean that government decisions can be fully handed over to neural networks. Data analysis is only one element of policymaking. Setting goals, assessing public consequences, choosing between competing interests, and taking responsibility for decisions made are no less important. However, Furman’s words show how far the automation of intellectual labor has advanced even in a field where professional training and expert judgment have traditionally been highly valued.

Economists will have to seek new value

The changes are also affecting education. Furman urges students who plan to do research or take part in developing economic policy to think about what unique contribution they will be able to make to their profession.

Simply gathering data, preparing charts, and retelling statistics may prove insufficient: AI is gradually mastering all of this at a level comparable to that of specialists, and sometimes even surpassing it.

A human’s competitive advantage may lie in the ability to formulate truly important questions, propose original hypotheses, notice problems that are not obvious from the data, and find unconventional explanations for what is happening.

Thus, the spread of AI poses a challenge to economists not so much to compete with machines in the speed of calculations as to rethink the very value of research work. If technically competent analysis becomes accessible to almost everyone, the scarce resource may turn out to be not computing power and not even the quality of data processing, but the ability to create new knowledge.