Artificial Intelligence Exposure, Tacit Experience and Occupational Wage Growth

The story of how machines change work has long been told as a simple race between human skill and computer power. For decades, economists and researchers have tried to predict which jobs will vanish and which will thrive by looking at how easily a task can be written down as a set of rules. If a job relies on following a clear checklist, the thinking went, a computer could eventually do it faster and cheaper. But if a job requires human intuition, creativity, or the ability to handle messy, unpredictable situations, it was assumed to be safe. This view treats technology as a force that acts directly on the worker, like a storm hitting a house. However, a new perspective suggests that the outcome depends less on the storm itself and more on how the house is built. It asks whether the value of a job comes from the formal rules a worker follows or from the deep, unspoken judgment they gain through years of practice.

A recent study by Zezhong Shang at China Jiliang University explores this idea by looking at the United States labor market between 2018 and 2024. The research focuses on the rise of artificial intelligence, specifically the new wave of tools that can read, write, and reason with language. The author does not simply ask if a job is exposed to these tools; instead, the study investigates how the type of knowledge a job requires changes the way wages grow when that exposure increases. The central finding is that jobs relying heavily on "tacit experience"—the kind of wisdom gained through doing, making mistakes, and navigating uncertainty—are seeing their wages rise faster when they are exposed to artificial intelligence. In contrast, jobs that rely mostly on "codified knowledge"—information that can be easily written down, taught in a classroom, or turned into a manual—are seeing a different pattern, though the study notes this result is less stable and sensitive to controls. The study suggests that the future of work is not just about what machines can do, but about how organizations decide to value the human ability to judge, negotiate, and take responsibility when the rules break down.

To understand this, one must first look at how the researchers measured the landscape of work. They started with a widely used index that maps out which occupations are most likely to be affected by artificial intelligence. This index, known as the AI Occupational Exposure Index, acts like a radar, showing where the capabilities of new technology overlap with the tasks people do every day. However, a radar alone cannot tell you what happens inside the workplace. It does not show if a company actually adopts the technology, how it changes the daily routine, or whether the workers get a raise. To fill this gap, the author built two new measures using data from the U.S. Department of Labor's O*NET database, a massive repository of information about job skills and requirements.

The first measure is the Codified Knowledge Index. This captures the parts of a job that are formal and documented. It includes things like specific knowledge domains, basic skills that can be taught in a school, and the need for professional certifications or apprenticeships. These are the elements of work that can be written down in a rulebook or a textbook. The second measure is the Tacit Experience Index. This captures the parts of a job that are learned through practice and are difficult to articulate. It includes skills like anticipating errors, negotiating with others, making decisions under uncertainty, and taking responsibility for outcomes. These are the qualities of a seasoned professional who knows how to handle a crisis not because they read a manual, but because they have been there before.

The study then combined these measures with wage data from 26,697 different combinations of states and occupations across the country. The researchers looked at how wages changed between 2018 and 2024, a period that saw the rapid emergence of generative artificial intelligence. They asked a specific question: Does the amount of artificial intelligence exposure affect wage growth differently depending on whether a job is heavy on codified knowledge or heavy on tacit experience? The answer they found was clear and statistically significant.

When the researchers looked at jobs with high levels of tacit experience, they saw that a one-standard-deviation increase in exposure to artificial intelligence was associated with a steeper growth in wages. In plain terms, the more a job relied on the kind of judgment and experience that comes from years of practice, the more its wages tended to rise as artificial intelligence became more prevalent. This relationship held true even after accounting for the starting wage of the job, the state where it was located, and the amount of formal knowledge required. The pattern was robust, surviving various checks and different ways of grouping the data. It suggests that in an era of advanced language tools, the ability to interpret outputs, manage exceptions, and make consequential decisions is becoming more valuable to employers.

Conversely, the study found a different pattern for jobs heavy in codified knowledge, though the author notes this result is less stable. For these roles, higher exposure to artificial intelligence was associated with a flatter or negative wage growth gradient, but this finding is sensitive to controls and index construction. This does not mean that formal education or certifications are useless. The data shows that these jobs still require formal knowledge, and that knowledge remains essential for defining problems and verifying the work of machines. However, the study suggests that the specific advantage of having a job that is primarily about following written rules is diminishing in the face of tools that can reproduce and manipulate those rules easily. The value of the work is shifting away from the ability to recite the rules and toward the ability to navigate the gray areas where the rules do not apply.

The researchers were careful to clarify what their findings do and do not prove. They did not track individual workers or specific companies to see exactly how artificial intelligence was being used. They did not measure the intensity of adoption or the direct causal effect of a machine on a single paycheck. Instead, they observed a broad pattern across the entire economy. The results show that the organization of work matters. Two jobs might have the same level of exposure to artificial intelligence, but if one relies on deep, practical experience and the other relies mostly on formal rules, their wage trajectories will likely diverge. The study implies that the labor market is beginning to reward the human capacity for situated judgment—the kind of thinking that happens in the moment, in the context of a specific situation—more than the ability to follow a pre-written script.

This distinction helps explain why some workers are thriving while others struggle, even when they work in fields that seem similarly affected by technology. It challenges the old idea that the only safe jobs are the ones that are completely immune to automation. Instead, it suggests that the most resilient workers are those whose value is tied to their experience and their ability to handle the unexpected. The study also highlights that the direction of change is not determined solely by the technology itself. It is determined by how employers choose to organize work and what they decide to reward. If companies use artificial intelligence to centralize control and strip away worker discretion, the benefits might not reach the employees. But if they use it to support workers who are already skilled at managing complexity, those workers may see their compensation grow.

The findings also offer a nuance to the ongoing debate about the future of work. Previous research often focused on whether a task was routine or non-routine. This study suggests that for the new generation of artificial intelligence, that distinction is becoming blurry. Machines can now perform many tasks that were once considered non-routine, such as writing code or summarizing complex reports. What remains distinct is the source of the knowledge. The study shows that the labor market is differentiating between knowledge that can be captured in a database and knowledge that lives in the mind of a practitioner. As artificial intelligence becomes more capable of handling the codified parts of a job, the economic premium for the tacit, experiential parts appears to be rising.

In the end, the research provides a map of how the value of human work is being reshaped. It suggests that the path to better wages in an age of artificial intelligence may not be about knowing more facts or following more rules, but about deepening the kind of practical wisdom that comes from doing the work. The study does not guarantee that every experienced worker will get a raise, nor does it promise that every rule-based job will disappear. It simply reveals a clear trend: in the current economic landscape, the ability to exercise judgment, anticipate problems, and take responsibility is becoming a more powerful driver of wage growth than the ability to follow a manual. As the technology continues to evolve, the way organizations value these human qualities will likely determine who benefits most from the new tools.