The Impact of Artificial Intelligence on Labor Markets: A Forecasting Model for Significant Economic Events
“We saw how powerful AI was becoming and wondered whether it could eventually take our jobs, along with millions of others. Unfortunately, there was little quantifiable research showing whether that risk was real. So we conducted the research ourselves and published it here so everyone can see the answer.”
Official Abstract
This study introduces a scalable, regionally adaptive framework for forecasting the impact of artificial intelligence (AI) on labor markets, designed to inform policy responses at local, national, and global levels.
[ SEE MORE ]
Recognizing the limitations of deterministic models and unstructured speculation, an ensemble of advanced large language models (LLMs) is used to classify over 1,000 occupations from the Standard Occupational Classification (SOC) system using structured prompts and task-level data from O*NET.
Rather than estimating continuous levels of AI disruption, our method applies quantization, discretizing occupations into five categories: High Displacement, Displacement, Neutral, Growth, and High Growth. This approach leverages the deep reasoning capabilities of LLMs to consistently evaluate job characteristics such as task automation potential, social complexity, and creative requirements. Cross-validation across multiple frontier LLMs ensures robustness and minimizes individual model bias.
These classifications serve as inputs to a stochastic Monte Carlo simulation that models employment outcomes over a five-year period under varying AI adoption scenarios. Simulations incorporate regional employment trends and assign randomized impact factors within each job category to generate a distribution of plausible labor market trajectories. Stochastic modeling was used as it better captures the uncertainty inherent in real-world labor markets, producing a more robust range of possible outcomes.
Preliminary results reveal significant geographic variation in occupational vulnerability to AI across U.S. regions, underscoring the need for flexible and proactive policy design. Targeted interventions, particularly in regions with high concentrations of automatable jobs, will be essential to mitigating displacement risk while leveraging AI’s economic potential.
This research offers a repeatable framework for labor market forecasting that is more informative, enabling decision-makers to prepare for and adapt to the evolving dynamics of AI-driven transformation.
Key Findings & Strategic Takeaways
Vulnerability to AI varies significantly by region, sector, and job type.
High-complexity roles, those requiring both strong soft skills and high specialization, show greater resilience to displacement.
Five-year forecasts show a wide range of plausible outcomes, underscoring the need for adaptive policy responses rather than one-time fixes.
All things constant, the rate of new job growth cannot offset the displacement caused as AI automates work tasks.
How It Works
Data & Inputs
Aggregates task-level descriptors for 1,000+ SOC-classified occupations from O*NET and BLS to establish a standardized occupational baseline.
Standardized Prompt Engineering
Applies a fixed, structured prompt schema to every occupation so that evaluation criteria remain identical and results stay comparable across roles.
Multi-Model LLM Ensemble
Passes each occupation through five distinct frontier LLMs independently, reducing the risk that any single model’s idiosyncratic bias skews the classification.
LLM Ensemble Quantization
Aggregates the five models’ independent outputs through majority-vote quantization, sorting each occupation into one of five discrete AI-exposure tiers.
Regional Workforce Mapping
Reweights the resulting national occupational classifications against regional employment composition to produce geographically specific labor market exposure estimates.
Stochastic Forecasting
Runs five-year Monte Carlo simulations across the classified occupations to capture real-world uncertainty in projected labor market trajectories.
A Closer Look at the Methodology
Forecasting something as complex as labor market disruption required a methodology that was both quantifiable and rigorous enough to function as a genuine forecasting tool. The authors’ backgrounds in economics and computer science informed the framework they selected, combining the classification capabilities of machine learning with the probabilistic, scenario-based forecasting techniques long used in econometrics.
The methodology took shape in two stages. An ensemble of large language models (LLMs) classified occupational risk, followed by a stochastic Monte Carlo simulation to forecast potential labor market outcomes.
Stage One: Classifying the Risk
We began with all 1,016 occupations recognized by the U.S. Bureau of Labor Statistics. This represents effectively every individual job type that exists in the U.S. economy. For each occupation, we compiled its title, description, and detailed task data from O*NET, then ran that information through our proprietary model: an ensemble of five leading large language models (LLMs), including Claude, GPT, Gemini, LLaMA, and Mistral.
Each of the five models evaluated every single job independently, judging it against a standardized rubric we built to ensure every occupation was assessed on the same criteria. Based on that rubric, each model made its own call, sorting the job into one of five categories, ranging from High Displacement to High Growth.
Because five separate models were making five separate judgment calls, the next step was reconciling them. The five verdicts were compared against one another, and the category that the majority of models agreed on became that job’s final classification. In the rare case of an even split, a numerical tiebreaker settled it, resolving the disagreement by weighing each model’s underlying score.
Stage Two: Forecasting the Outcome
With each occupation classified, the next step is mapping that classification onto the actual workforce. Job categories vary widely in size, so the number of people employed within each one has to be weighted individually to get an accurate picture of real-world exposure. We first did this at the national level, which gave us a baseline understanding of the country’s overall displacement and growth potential.
From there, we extended the model to all 393 Metropolitan Statistical Areas (MSAs) in the United States, mapping the same classifications onto each region’s specific workforce composition. This let us see whether displacement and growth risk shift based on the underlying structure of a city’s economy, for example a tech-forward metro versus a more rural, manufacturing-heavy one, insight that can support more tailored, region-specific policy responses.
Finally, to capture the uncertainty in how these trends unfold over time, we added a Monte Carlo simulation to the model. This is a stochastic modeling technique, meaning it relies on repeated random sampling rather than one fixed calculation. For each occupation, we ran 100,000 simulated five-year trajectories. Each trajectory was tested under three scenarios: best, neutral, and worst case. Every scenario drew from a different range of plausible annual displacement rates. The output is a probability-weighted range of outcomes for each occupation and region, the same kind of technique used in financial risk modeling and climate forecasting.
Results & Core Findings
of America’s workforce is at risk of AI-related displacement.
Three in five American workers, 61.63% of the nation’s workforce, hold jobs at risk of AI-related displacement within the next five years, a statistic that touches nearly every industry, from manufacturing floors to office cubicles. A shift of this scale carries real consequences for the stability of economies across the US. Our research gave us a lot of insight beyond just this displacement number, and in the remainder of this piece we dive into that information along with what it means for the workers, industries, and regions behind the data.
When we look at the percentage of the workforce that falls into each of the specific buckets, we see that High Displacement roles, the jobs most likely to see significant employment loss to AI, make up 35.9% of American workers. Displacement roles, which face a meaningful but somewhat less severe decline, account for another 25.7%, putting the combined share of workers facing real displacement risk above three in five. Neutral roles, where AI is expected to have little net effect on employment levels either way, represent 21.4% of the workforce. Growth roles, where AI is expected to increase demand for the position, make up 16.5%, a positive signal but a modest one. High Growth roles, expected to see the strongest employment gains from AI, are the smallest group at just 0.5%. Taken as a whole, the distribution leans unfavorably: a large majority of American workers sit in categories facing some degree of displacement risk, while only a small share are positioned to clearly benefit.
Which Sectors Are Most Affected, and Why
Displacement risk is not spread evenly across the economy. It concentrates heavily in a handful of sectors, defined less by their industry label and more by the character of the tasks performed inside them.
Production led all sectors in displacement risk. Roles in this space, think assembly line work, packaging, and machine operation, tend to involve repetitive, physically structured tasks that follow a predictable sequence. That is exactly the kind of work current automation and robotics can already replicate at scale. Transportation and Material Moving followed closely, driven largely by advances in AI-assisted and autonomous driving technology that directly threaten roles built around routing, delivery, and logistics execution. Office and Administrative Support, Business and Financial Operations, and Sales rounded out the top five. All three are built heavily around data entry, structured record-keeping, and transactional customer interaction, tasks that generative AI and automated software are increasingly capable of performing faster and more cheaply than a human employee.
On the other end, Healthcare Practitioners and Technical occupations showed the strongest growth signal, followed by Life, Physical, and Social Science; Education, Training, and Library; Architecture and Engineering; and Management. These sectors share a common thread. They depend on skills that remain difficult for AI to replicate, including hands-on physical care, complex interpersonal judgment, creative problem-solving, and accountability for consequential decisions. Rather than being replaced, these roles are more likely to be reshaped by AI as a supporting tool.
One important clarification is worth stating plainly. A sector landing in a high-displacement category does not mean every job inside it is in danger. It means that sector contains a disproportionately high concentration of occupations our model classified as at-risk. A Production-sector engineer overseeing an automated line and the line worker performing a repetitive task within that same sector face very different levels of exposure, even though both technically sit inside the same industry designation.
Regional Variation
Risk is also not uniform across the country. Displacement and growth potential shift meaningfully depending on how a local economy is structured. Metropolitan areas built around diversified, innovation-driven industries, such as technology, advanced professional services, and higher education, tend to have a workforce already concentrated in the same sectors showing the strongest AI-driven growth. That existing composition cushions the overall impact. By contrast, metro areas with a narrower industrial base, particularly those weighted toward production, logistics, or routine administrative work, and with less existing infrastructure for retraining, face a steeper concentration of displacement risk.
This regional dimension is precisely why the model was built to operate below the national level. A national average, no matter how precisely calculated, obscures the reality that a rural manufacturing hub and a diversified tech metro are not facing the same future, even when they show up as a single blended figure in national labor statistics.
The Bond-Rand Model
We’re working to make this research more useful and accessible to everyone. Check back soon for an interactive tool that shows how AI may be affecting your own role.
Policy & Economic Implications
If AI’s impact on a labor market depends on that region’s specific mix of industries rather than falling evenly across the country, a single national policy can’t realistically serve every region well. A plan calibrated for a diversified tech hub will miss what a manufacturing-heavy metro actually needs, and a plan built around production and logistics concerns would be the wrong fit for a region already anchored in growth sectors like healthcare or professional services. The regional variation in our findings points toward one conclusion: effective policy has to be localized rather than uniform.
For metro areas carrying a heavy concentration of high-displacement sectors, that means prioritizing direct workforce investment: retraining pipelines built in partnership with local educational institutions, and incentives, like tax credits, for employers who commit to reskilling displaced workers rather than simply letting them go. For regions already positioned in growth sectors, the priority looks different. It means reinforcing that existing advantage by expanding credential pathways in the fields already driving local growth, so that momentum compounds.
The regions carrying the most exposure to displacement can be identified now, before the disruption fully plays out. That timing matters. Waiting on a single federal framework to address every local labor market the same way risks missing the window where targeted investment could have made the most difference.
Differentiating Yourself in the Age of AI
For most of history, standing out at work meant outperforming the person next to you. Today, the competition looks different. AI can already do things that would have sounded like science fiction a decade ago: drafting a legal brief in seconds, spotting patterns in a spreadsheet faster than any analyst, catching things in a medical scan a trained eye might miss. Workers now find themselves measuring up against a machine that keeps getting better, on top of the usual competition from other people. That raises a real question for anyone building a career today: how do you stay valuable next to something that never gets tired and never stops improving? The answer comes down to three categories of skill: technological literacy, hard skills, and soft skills. Build all three, and this shift becomes an opportunity you can act on.
Technological Literacy: The New Baseline
There was a time when knowing how to use a computer was a specialized skill. Now it’s assumed of nearly everyone, regardless of industry. AI is on the same trajectory. Every field, from marketing to manufacturing to medicine, is being reshaped by it in some form, whether that shows up as growth or as displacement. Because of that, computational literacy is quickly becoming the new floor, the same kind of baseline expectation that basic writing and math once were.
In practice, this looks like being genuinely comfortable working alongside AI tools: knowing how to prompt them effectively, having a rough sense of how they arrive at their outputs, and carrying enough data fluency to sanity-check what they produce. The workers who thrive here are the ones who learn to direct these tools skillfully, turning them into leverage.
Hard Skills: Go Where the Machine Can’t Follow
Not every field is equally exposed to automation, and knowing where AI struggles is one of the clearest signals of where to invest your time. Our research found the strongest growth potential concentrated in a specific set of domains: analytical and STEM-based reasoning, healthcare and human services, education and training, and systems-level or project-based thinking. These fields share something important: they demand judgment calls that resist standardization, physical or interpersonal presence that can’t be automated remotely, or reasoning that has to adapt fresh each time rather than follow a repeatable pattern.
There’s a practical upside to this too. These fields also tend to require formal training and credentials, so degrees and certifications are gaining value as a signal in a labor market splitting into two lanes: routine, automatable work on one side, and complex, credentialed work on the other.
Soft Skills: What Makes You Irreplaceable
If technological literacy is the baseline and hard skills are the specialization, soft skills are what tie everything together. Complex interpersonal judgment, creative problem-solving, adaptability, and the willingness to own a consequential decision are exactly the kinds of things AI still struggles to replicate convincingly. Reading a room, navigating a tense negotiation, or taking responsibility when a judgment call goes wrong all require a kind of presence and accountability that current models simply don’t have.
Workers who pair strong soft skills with technological literacy and relevant hard skills are building the exact profile our research found least exposed to displacement.
Where We’ve Shared This Work
Introduced the Bond-Rand model, which pairs LLM-based job classification with Monte Carlo simulation to forecast AI’s impact on the labor force and inform policy.
Uses the Bond-Rand model’s findings to lay out how the US should respond to AI-driven labor disruption, focusing on the skills workers need and the concrete actions education, companies, and government must take to support the workforce.
Brought the Bond-Rand model’s Pennsylvania findings to state legislators in Harrisburg, showing that AI’s impact varies significantly by region and making the case for targeted, city-specific policy action rather than a one-size-fits-all approach.
Focused on Erie, PA, this event used the Bond-Rand model’s findings to examine how AI is projected to reshape the local economy and what policy and workforce steps are needed to prepare.
Discussed how AI is expected to impact education and the steps institutions like Penn State need to start taking to prepare.
Resources & Citation
Bond, A. J., & Rand, C. R. (2025). The impact of artificial intelligence on labor markets: A forecasting model for significant economic events. Proceedings of the AAAI Symposium Series, 6(1). https://doi.org/10.1609/aaaiss.v6i1.36027
Frey, C. B., & Osborne, M. A. (2013). The future of employment: How susceptible are jobs to computerisation? Oxford Martin School, University of Oxford.
A foundational, ahead-of-its-time analysis of occupational automation risk. Much of its methodology and its core findings remain relevant today, and it was a key influence in developing the Bond-Rand model.
Read the paper →Bring This Research to Your Organization
We’re glad to bring this work to conferences, speak with regulators, or partner with educational institutions working through the same questions. The first conversation is free, let’s talk about what that could look like.
© 2026 Bond-Rand Technologies. This research is shared for informational and educational purposes only. It may not be copied, reproduced, or used for commercial purposes without express written permission from Bond-Rand Technologies.