Could AI Master Economic Thinking to Solve Real-World Problems?

by Melissa Carleton

To many people, the economy represents a vast mystery of supply chains, tariffs, and uncertainty. To most professionals, making everyday business decisions regarding pricing, budgeting, or forecasting demand for a product appears an intractable problem. The study of economics attempts to put some structure on these moving pieces.

With the rise in economic uncertainty spurred by recent societal developments, such as AI, it’s worth asking whether AI itself can provide expert-level economic decision-making for individuals and organizations to sort through the noise.

We’re approaching a future where a business analyst, policymaker, or other professional can call upon an expert economist at any point in the form of an AI tool. As this process unfolds before our eyes, it’s worth analyzing how economic knowledge becomes embedded in AI tools. Can it ever reach the point of helping professionals or individuals make everyday decisions?

Currently, several companies are hiring economists to fill positions at AI labs. The economists often train the AI model to devise correct answers to research-level inquiries. Their goal is to bring the AI models closer to the frontier of human knowledge for widespread benefit, not only for specialized experts.

Many people colloquially refer to the potential of AI to cure diseases or reduce environmental pollution. It’s also worth exploring whether AI could help society achieve widespread economic prosperity or avoid a collapse by aggregating all that’s known about economics.

In this article, I explore how economic knowledge becomes embedded in commonly used large language models (LLMs). I also explore the possibility that such knowledge could help businesses and society achieve desirable social outcomes.

What Kind of Economics Can Most Large Language Models Do?

If you ask ChatGPT how to solve a simple economics-related problem, such as calculating the price a business should set for a given product, it can spin up an estimate using undergraduate-level economic optimization and some assumptions. It can also solve many theoretical problems found in first-year PhD-level classes.

It can explain widely known economic theories involving common assumptions and interpret statistical output conveying simple relationships. And, if you supplied it with a self-contained economic model connecting, say, wages and employment, it could generate examples and simulations of what would happen if the Fed raised interest rates.

However, if you test the model with a question deeper into the territory of publication-quality research, it will become stumped. It will often start making assumptions that fundamentally change the nature of the problem and are not necessary to arrive at the answer. While it can produce code to analyze economic data, which runs when leveraged by a skilled programmer, it cannot analyze this data directly.

But over time, we should expect to see this distant hypothetical possibility converge closer to reality. What if one day LLMs could leverage the most sophisticated macroeconomic models and estimation techniques developed by researchers to inform business decisions or better predict economic cycles?

What Powers Economic Reasoning in LLMs?

AI model builders leverage a variety of sources to train LLMs, including news, unstructured text data, public websites, and code repositories. They are also trained by human data annotators.

Although “data annotation” can take many forms, it has shifted in recent years from mostly low-paid labor to a growing reliance on expert annotators who often hold or are pursuing advanced degrees, like PhDs. Their task typically involves ingesting advanced problems in a specific domain into an AI model and evaluating its output.

These problems with solutions then become “labeled” data. However, it’s usually not large AI giants like OpenAI or Anthropic who hire these domain experts. Enter the world of data labeling companies. While these companies don’t achieve the same name recognition as Google or Amazon, they more than compensate for this in growth.

The Role of Data Labeling Companies

The top data labeling companies that hire economist contractors today include ScaleAI, Surge AI, and Appen. These fast-growth companies boast valuations in the hundreds of millions or billions. Scale AI was founded in 2016 and grew to a valuation of $14B by 2025. The recent $15B investment in the company by Meta brings it to a valuation of $29B.

Data labeling companies often sell their annotated data to frontier AI model development companies that produce LLMs for direct usage. Frontier AI companies could instead rely on the messy data from text, blogs, images, and ungated news articles. But purchasing data labeled by human experts dramatically increases their valuation.

Data annotation labs, rather than the large AI giants, hire these experts. They often pay them anywhere from $50-$100 per hour. These roles can present a great opportunity for PhD students who hope to make some extra cash part time and contribute their specialized domain knowledge outside of academia. It’s worth examining how and when such knowledge will be leveraged to benefit businesses and the broader public.

The State of Economic Reasoning in AI Models

Although companies like ScaleAI are rapidly advancing the AI’s ability to reason like an expert economist, the broader economics literature that provides advanced formulas to help businesses predict tail risk or optimize investment strategies has not substantially trickled into the capabilities of frontier publicly accessible LLMs. In a typical economics PhD program, likely only one or two students may decide to work as contractors. Given the vast amount of knowledge in the economics profession, the work of a few individuals, though important, barely scratches the surface.

Moreover, it’s not clear that company developers of large language models have yet implemented the contributions of contractors hired by their data labeling partners. While a PhD economist may ingest a macroeconomic forecasting model into a data labeler, it’s less clear whether and when this model could help policy makers predict macroeconomic risk and what policies they should implement to prevent a collapse from occurring.

However, we are converging to a future in which large-scale aggregation of expert economics knowledge is becoming increasingly feasible. While individual economic papers or insights tend to be highly specific or rest on several assumptions, we are reaching a point where insights from multiple studies can be integrated to form a framework suited to a given scenario. Such inputs, if interpreted critically, could provide clarity surrounding policy or business decisions that previously relied on guesswork.

It’s Not Economics Alone

Although economics, as a discipline, focuses on optimal resource allocation and predicting market events, advances in computer science and machine learning promote rapid processing of data and the creation of advanced algorithms that reveal richer patterns in the economy and social structures. Disciplines such as sociology and psychology can help ensure that an economic model is applied appropriately to a particular context.

For example, say a city planner wants to develop the optimal transit structure to reduce congestion and is also concerned with equitable access to employment. Economics possesses the tools to model individual choices at a large scale, which is why such a framework, if deployed at a large scale, could inform better decision making.

But economics alone isn’t enough. Machine learning can supply the best algorithms to process data used to estimate the model and determine practical and grounded outcomes. Insights from other social science disciplines, such as political science, could ensure that citizens react positively to any new changes.

The Path Forward

So far I have posed the question of whether decision makers could potentially leverage AI tools to address challenges in ways often only accessible to expert economists. We are moving closer to a future where government or business committees could leverage AI tools that reason faster than the world’s top economists. Such cutting-edge economic knowledge could help them predict how economic recessions will impact their stakeholders.

While this possibility remains a thought experiment, technological developments in recent years have allowed us to conceive of such a possibility and seriously consider its impacts.

We often discuss the potential of AI to push society forward in vague and aspirational terms. To ensure broader prosperity and hack some of the largest economic challenges facing our society, it’s worth exploring whether AI trained to reason like an economist could promote this goal through making economic insights accessible. From the data labeling experts at data annotation companies to the frontier companies purchasing their outputs, several moving parts working together could allow us to achieve this possibility.

Sources:

Athey, Susan, and Fiona M. Scott Morton. “Artificial Intelligence, Competition, and Welfare.” NBER Working Paper No. w34444. November 2025. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5729726

Hammond, George; Melissa Heikkilä; and Cristina Criddle. “Meta invests $15bn in Scale AI, doubling start-up’s valuation.” Financial Times, June 12, 2025. https://www.ft.com/content/5a30cd25-90f9-41a4-924c-1e7c6772a47f

Ludwig, Jens; Sendhil Mullainathan; and Ashesh Rambachan. “Large Language Models: An Applied Econometric Framework.” NBER Working Paper No. w33344, January 2025. doi:10.3386/w33344.

OpenAI. “How ChatGPT and Our Foundation Models Are Developed.” OpenAI Help Center, updated April 30, 2025. https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-foundation-models-are-developed

Sajid, Haziqa. “12 Best Data Labeling Companies [2025].” Encord Blog, September 10, 2025. https://encord.com/blog/data-annotation-companies-for-computer-vision/

Writing Team. “Data Labeling Is the Hottest Job Market Nobody’s Talking About.” Winsome Marketing, August 5, 2025.https://winsomemarketing.com/ai-in-marketing/data-labeling-is-the-hottest-job-market-nobodys-talking-about