Artificial intelligence is beginning to absorb a large share of the routine mathematical and computational work that once defined many corners of economics and related social sciences. As these tools become widely available, the discipline is likely to split: the best research will improve, while work built on standard, easily replicated methods could be squeezed.
At the top end, AI is poised to act less like a replacement for researchers and more like an amplifier. Projects led by humans could scale in ways that were previously too expensive or slow. That includes conducting and coding large numbers of qualitative interviews, designing and monitoring behavioral interventions in real-world settings, and processing vast collections of text—policy documents, corporate communications, social media, court records—at a level of coverage that would be daunting for a traditional team.
That shift will likely raise the premium on field experiments. If AI makes analysis and writing faster, then the bottleneck moves to access: relationships with firms, governments, schools, hospitals, and platforms that can host interventions and generate credible evidence. Researchers able to secure partnerships and run on-the-ground tests—potentially including trials of AI-based workflows inside organizations—may find their work in especially high demand.
Administrative data, already a powerful resource in applied economics, may become even more central. AI could help analyze sensitive datasets at scale, but the benefits may not be broadly shared. Access to government records often depends on strict permissions and controlled environments. In this scenario, a smaller set of established researchers—those trusted by agencies and allowed to deploy local, non-cloud models on secure systems—could gain a significant advantage. Early-career researchers and PhD students, who typically lack both credentials and connections, could find it harder to compete in these high-value data settings.
Another area expected to surge is economic history. If modern AI systems can rapidly transcribe, translate, and structure archival material, then the limiting factor becomes the digitization of source documents. That could trigger a competitive push to scan and convert archives—shipping logs, censuses, firm ledgers, newspapers, legal filings—into machine-readable data. The payoff is twofold: new descriptive facts and new ways to identify causal relationships using historical variation.
AI is also likely to reshape how research gets vetted. Instead of peer review focusing mainly on the manuscript, verification could expand to include the full research package—paper, code, and data—at the moment of submission. Automated checks for errors, reproducibility, and consistency could make “360-degree” review far more routine, raising standards and reducing the lag between claims and validation.
The biggest disruption may hit studies that can already be done end-to-end with minimal bespoke inputs: downloading public datasets, running straightforward survey experiments, or producing relatively standard theoretical models. If AI can execute much of that pipeline cheaply and quickly, this “bread-and-butter” output could lose value for academic researchers who depend on it. Some of that work may migrate to public agencies and large institutions that can answer their own operational questions without needing the incentives and timelines of peer-reviewed publication.
Why it matters
AI could widen the gap between researchers who have privileged access to data, partnerships, and archives and those who do not, reshaping career paths in economics and the social sciences. Institutions may raise expectations for transparency and replication as AI-assisted review becomes standard. The next phase to watch is how universities, journals, and governments set rules for secure AI use on sensitive data—and who gets authorized to use it.