Beyond Controlled Studies: A Multidisciplinary Approach to Understanding AI's Economic Impact

The transformative potential of artificial intelligence on the global economy is a subject of intense debate and speculation. While some economists forecast a modest 0.9% increase in GDP over the next decade, others envision a seismic shift adding up to $26 trillion to annual global output. Yet, even before these impacts fully materialize, our collective beliefs about an AI-driven future are already steering career choices, government policy, and massive investments in data centers and semiconductors. This article argues that to truly grasp AI's economic ramifications, we must move beyond the limitations of traditional controlled studies and embrace a more imaginative, multidisciplinary framework.

The Limitations of Controlled Experiments

Researchers often attempt to quantify AI's causal impact through natural experiments and randomized controlled trials (RCTs). In these studies, one group gains access to an AI tool while a control group does not, with other factors held constant. Outcomes like productivity and learning are then measured. For instance, a 2025 study found call-center workers handled queries 15% faster with 2020-era AI tools, while another showed software developers with coding assistants completed 26% more tasks in 2022-23. However, this approach faces two critical challenges when applied to AI.

First, the rapid pace of AI advancement renders causal estimates outdated almost upon publication. Since ChatGPT's 2022 release, AI tools have tripled their capability to handle simulated customer-support chats autonomously. The economic effects of tomorrow's cheaper, more capable AI will differ fundamentally from those measured today. Second, and more importantly, controlled studies fail to capture the wider ripple effects of adoption. They measure impact within a fixed organizational structure, but in reality, managers may respond by reorganizing work or replacing less-experienced workers with AI systems. Payroll data since 2022 suggests a decline in employment for younger workers in AI-susceptible fields like customer service, a potential ripple effect that controlled studies did not predict.

Three Complementary Approaches for a Fuller Picture

To navigate these uncertainties, economists must collaborate with social scientists, employing a blend of imagination and empirical grounding. Here are three complementary approaches that can guide research in this evolving field.

1. Social Science Fiction

Pioneered by economist Jean Tirole, this approach involves speculation about the future rooted in fundamental economic principles. It uses models to explore how technologies interact with market forces. For example, a 2019 model on self-driving cars predicted they could worsen traffic congestion because the reduced personal cost of travel time would increase car usage, creating a negative externality. Similarly, research suggests that as AI automates certain tasks, the value of complementary, non-automatable activities—like creative direction or quality vetting—will rise. This could increase demand and wages for such labor, potentially cushioning automation's disruptive effects but also deepening inequality between those who thrive in these roles and those who do not. These thought experiments help identify key indicators to monitor and allow policymakers to plan interventions proactively.

2. Forward-Looking Data and Benchmarks

Policymakers need evidence, not just theory. A common method is benchmarking—testing AI on standardized tasks. However, benchmark scores often diverge from real-world performance where tasks are noisier and more complex. A medical AI might ace textbook questions but misinterpret nuanced patient communications. More research is needed to design benchmarks that better reflect real-world conditions. Meanwhile, real-time usage data offers valuable insights. For example, data showing that AI chatbots are frequently used for software development suggests this sector may feel AI's earliest effects. Other indicators to monitor include employment trends, job openings, and the profitability of AI-integrating firms.

3. Anticipating Scaling Effects

Estimating AI's causal effects is uniquely difficult because both the technology and organizations are evolving simultaneously. This challenge mirrors evaluating any pilot program before scaling. Economists have developed methods to anticipate scaling effects by replicating the conditions of the eventual implementer (e.g., a government agency) rather than the agile organizations that typically run pilots. Researchers studying AI can adopt similar strategies. Two crucial parameters are cost and capability. One study examined AI usage by teachers in Sierra Leone who pay for internet by the megabyte. Between 2022 and 2025, the cost of querying an AI chatbot fell dramatically, becoming 98% cheaper than loading a standard web page. This suggests AI could expand information access in low-resource settings. While predicting future capabilities is hard, researchers can model how humans might respond to more powerful systems based on stable behavioral patterns around trust, incentives, and adaptation to automation.

Conclusion: Embracing a Broader Lens

Carefully controlled studies provide a bright, narrow spotlight on AI's effects, but they are insufficient alone. Understanding how AI will reshape the economy requires illuminating the entire stage with a combination of tools: imaginative yet grounded social science fiction, innovative forward-looking data analysis, and methods that anticipate the complex scaling effects of a dynamic technology. By integrating these multidisciplinary approaches, researchers and policymakers can move beyond outdated metrics and speculative narratives. They can develop a more nuanced, timely, and comprehensive understanding of AI's true economic impact, enabling the creation of more resilient and equitable policies for the future. The stakes—trillions in economic output and the future of work—demand nothing less than this expanded perspective.