AI isn’t having as much of an impact in the economy as some had predicted, but current AI capabilities can already do a significant share of jobs and tasks in the global economy.
While visible disruptions have largely been contained to the technology sector — visible through layoffs at tech firms — a new study from researchers at the Massachusetts Institute of Technology (MIT) and Oak Ridge National Laboratory suggests that the true extent of AI’s reach is five times larger than what current economic indicators reveal.
Methodology: Building a Digital Twin of the Workforce
To understand the hidden ripple effects of artificial intelligence, the researchers developed “Project Iceberg,” a comprehensive simulation designed to capture the nuance of the human-AI labor market. Unlike traditional economic metrics that measure employment outcomes only after disruption has occurred, Project Iceberg utilizes Large Population Models (LPMs) to simulate the economy before these shifts crystallize. The team utilized the Frontier supercomputer at Oak Ridge National Laboratory to create a massive agent-based model representing 151 million American workers.
The construction of this simulation involved a granular mapping of the entire U.S. labor force. The researchers modeled workers as autonomous agents distributed across 3,000 counties and 923 distinct occupations. By integrating data from the Bureau of Labor Statistics, they assigned each agent specific attributes, covering more than 32,000 distinct skills ranging from “critical thinking” to specific technical competencies. This digital representation allowed the researchers to move beyond broad job titles and analyze the specific bundles of tasks that constitute modern employment.
On the technological side of the equation, the study cataloged over 13,000 AI tools currently available in the market, including coding copilots, workflow automation platforms, and specialized cognitive software. Rather than relying on theoretical benchmarks of what “frontier models” might one day achieve, the team focused on production-ready systems that are deployable today. Using a semi-automated mapping pipeline involving Large Language Models, they analyzed the capabilities of these tools to determine which human skills they could technically perform.
The core metric derived from this simulation is the “Iceberg Index.” This index measures the percentage of wage value within an occupation that is exposed to AI automation. It is calculated by identifying the overlap between human skills and AI capabilities, weighted by the economic value of those skills. Crucially, the researchers validated their framework against real-world data, finding that their skill-based definitions successfully predicted 85% of actual career transition patterns and aligned with geographic AI usage data from Anthropic.
Results: The Hidden Mass of the AI Economy
The study’s findings reveal a stark contrast between the visible impact of AI and its underlying potential. The researchers identified a “Surface Index” of 2.2%, representing the wage value currently exposed to AI in the computing and technology sectors. This equates to approximately $211 billion in wages and includes roles such as software engineers and data scientists, where AI adoption is most observable. This surface-level disruption is heavily concentrated in coastal technology hubs like Washington, Virginia, and California.
However, the “Iceberg Index” paints a drastically different picture. The study found that technical capability extends far below the surface, exposing 11.7% of the total labor market’s wage value to automation. This figure represents approximately $1.2 trillion in wages and is driven by “cognitive automation” in administrative, financial, and professional services. While the tech sector grabs headlines, the hidden mass of exposure lies in white-collar functions that support the broader economy, such as document processing, financial analysis, and routine coordination.
Perhaps the most significant finding is the geographic distribution of this hidden exposure. While the Surface Index is concentrated in tech capitals, the broader Iceberg Index is distributed across the entire nation, creating what the researchers term an “automation surprise”. Industrial and manufacturing states in the American Heartland, such as Ohio, Tennessee, and Michigan, show double-digit exposure rates comparable to or higher than coastal tech hubs. For instance, Tennessee has a low Surface Index of 1.3% but an Iceberg Index of 11.6%, driven by the heavy reliance on administrative and logistics coordination within its manufacturing sector.
This data suggests that traditional economic indicators like GDP and unemployment are currently blind to these risks. The study notes that these conventional metrics explain less than 5% of the variation in AI skill exposure. Consequently, regions with smaller tech sectors may underestimate their vulnerability, assuming they are insulated from AI disruption. In reality, the ubiquity of administrative and cognitive tasks means that states like South Dakota and Delaware face higher relative exposure in their labor markets than California. The MIT study ultimately concludes that the “automation surprise” is likely to hit these non-tech sectors hardest, as AI capabilities silently overlap with human skills long before they result in visible job displacement.
There has been some disconnect between AI abilities and their use in real-world scenarios — former OpenAI Chief Scientist Ilya Sutskever had alluded to this in a recent interview — but the study suggests that these impacts might be soon on the horizon. Other AI leaders, such as Anthropic CEO Dario Amodei, have been predicting major job losses for white-collar workers, with 50% of entry-level white-collar jobs being eliminated by 2030. And if this MIT study’s predictions are correct, these job losses would occur in sectors far removed from tech and hit a much broader part of the global economy.