Beyond the hype: What research shows about the value of college in the AI era

INSIGHTS ARE PRELIMINARY MATERIALS CIRCULATED TO STIMULATE DISCUSSION AND CRITICAL COMMENT. THE VIEWS EXPRESSED ARE THOSE OF THE INDIVIDUAL AUTHORS. WHILE INSIGHTS BENEFIT FROM ACTIVE UHERO DISCUSSION, THEY HAVE NOT UNDERGONE FORMAL ACADEMIC PEER REVIEW.

By Rachel Inafuku and Steven Bond-Smith

Does a college education still provide significant labor market advantages in the Artificial Intelligence (AI) era? Many observers argue that AI is beginning to replace the entry-level white-collar roles that once served as career launching points for new college graduates. A 2024 McKinsey & Company survey found that the share of workers using AI in at least one job function rose to 72%, up from 55% in 2023, reflecting its rapid integration into the workplace. This has fueled a narrative that traditional avenues to high-paying jobs, such as college degrees, may be less relevant in an economy increasingly shaped by AI.

At the same time, public confidence in higher education has declined. A 2024 Gallup survey found that only 36% of U.S. adults have a “great deal” or “quite a lot” of confidence in higher education, dramatically lower than the 57% who said so in 2015. Among those who lack confidence, 30% say colleges do not teach the skills needed to succeed in the workplace. A Pew Research Center survey echoes these concerns, showing that a growing share of Americans believe higher education is headed in the wrong direction. This skepticism reflects, in part, uncertainty about which skills will be rewarded in an AI-driven economy, and whether an advanced degree will provide them.

To understand how AI may affect the value of a college degree, it helps to first consider how these technologies are reshaping the nature of work. Research consistently finds that the occupations most exposed to AI are not low-wage or low-skill jobs, but higher-paying, knowledge-intensive roles that typically require more education. For example, a Pew Research Center analysis found that workers in highly AI-exposed jobs (i.e., jobs where workers use relatively more AI technology) earn an average of $33 an hour, compared with $20 among those in the least exposed roles. Examples of highly exposed occupations include tax preparers, data entry keyers, and computer hardware engineers, while “hands-on” jobs like barbers, firefighters, and nursing assistants face the lowest exposure.

An analysis by the US Department of the Treasury’s Office of Economic Policy also highlights this pattern: in the occupations least exposed to AI, workers with a bachelor’s degree make up fewer than 10% of employees, while in the occupations most exposed to AI, more than half hold at least a bachelor’s degree. Rather than targeting low-skill work, so far, AI is most likely to affect the kinds of information-rich tasks that college-educated workers perform. But they also note that exposure is a descriptive measure of where AI capabilities overlap with job tasks—not a forecast of unemployment. In other words, higher exposure to AI does not necessarily imply greater risk of job loss; instead, it reflects greater scope for changes in how work is performed, often through task augmentation rather than substitution.

To understand the employment and wage effects of AI, we need to consider the types of tasks that AI undertakes and how those tasks relate to jobs. AI would be more likely to augment roles where workers use it to raise productivity on supporting tasks while retaining responsibility for judgment and decision-making, as seen in real-world studies of generative AI tools that increase individual worker productivity in customer service. By contrast, jobs (or parts of jobs) dominated by routine, standardized tasks would be more likely to face substitution pressure, consistent with broader labor research showing that technologies tend to automate repetitive task content first. A classic study by MIT and Harvard economists showed that earlier waves of computerization tended to replace routine, rules-based tasks first while complementing more judgment-intensive, nonroutine work—an idea that helps explain why AI is more likely to substitute for routine-task-heavy roles and augment roles centered on judgment and decision-making.

Because routine task content varies widely within both college- and non-college occupations, the impact depends more on task mix than on education level alone. A chapter in the Handbook of Labor Economics argues that technology changes labor demand by reshaping the task content of jobs, which is why AI’s impact depends more on the mix of tasks within an occupation than on whether it typically requires a college degree. But college-educated jobs tend, on average, to involve fewer routine, easily automated tasks, so AI would be more likely to augment workers in these roles by increasing productivity on supporting tasks, and may even lift wages for some college-educated workers. By contrast, jobs with a higher share of routine tasks would face greater substitution pressure, as AI could automate portions of the work and reduce the labor required to produce a given level of output. On that basis, greater AI exposure in jobs that usually require a college education—where it primarily augments worker productivity—could raise labor demand and wages for some college-educated workers, especially where output expands, and productivity gains are shared.

Using data from Eloundou et al., we find that in Hawai‘i, jobs in the lowest two deciles of AI exposure have a median annual wage below $60,000, compared with more than $80,000 for jobs in the highest five deciles of exposure. While estimates of AI exposure vary widely, Hawai‘i has a slightly lower share of jobs that are highly exposed to AI. Using the Eloundou et al. data, approximately 39,000 workers in Hawai‘i—about 7% of total employment—are in occupations within the top decile of AI exposure. This is notably smaller than the national share of 11%, reflecting Hawai‘i’s industry mix, which is more concentrated in lower skill tourism-related jobs (i.e., accommodations, food service, retail trade) that tend to have lower AI exposure. A study by the US Department of the Treasury’s Office of Economic Policy, which uses a different dataset and methodology, similarly ranks Hawai‘i 32nd among the 50 states in terms of AI exposure. Their estimates suggest that less than a quarter of Hawai‘i’s workforce is highly exposed to AI, compared with nearly 40% in Washington, DC—the most exposed labor market in the nation.

The net employment impacts of AI remain uncertain in aggregate, but its effects on where workers are employed and what they do are becoming clearer. Employment declines when AI automates enough tasks within a job to reduce firms’ demand for workers, even if total employment remains stable. But AI can also complement human labor, improving productivity and creating new types of work. Employment increases when AI raises productivity enough to expand output and create new or better jobs, boosting demand for AI-complementary workers. According to the Pew Research Center, among workers in industries most exposed to AI, more believe the technology will help rather than hurt their jobs. So far, research supports this optimism: although AI is being adopted rapidly, its overall influence on employment levels appears limited so far. In this way, AI can simultaneously displace some workers, complement others, and leave total employment largely unchanged.

A growing number of studies suggest that AI is reshaping tasks rather than eliminating jobs. Eloundou et al. estimate that about 15% of tasks in the US could be completed substantially faster, and with the same quality, using large language models. And while many occupations—especially higher-wage ones—face exposure to AI-enabled automation, Hampole et al. find the overall employment effects remain limited. Productivity gains at AI-adopting firms appear to offset reductions in routine-task employment. Similarly, a Yale Budget Lab analysis finds that the share of workers in high- and medium-AI-exposure jobs has remained relatively stable since the public release of ChatGPT. Taken together, these findings suggest that AI is altering the nature of work rather than replacing it.

If AI is not eliminating jobs outright, the more relevant question is how skill demands are changing as the labor market evolves. The data suggests that workers who can effectively use AI are becoming more sought after. In fact, skills related to AI—once considered niche—are increasingly expected across a wide range of occupations. Data from Lightcast shows roughly 160,000 US job postings listed AI as a required skill in December 2025, representing a 134% increase from the previous year. Hawai’i mirrors this trend, with job postings requiring generative AI skills rising by approximately 120% year-over-year. More broadly, AI skill requirements now appear in 1.7% of all job postings, up from just 0.5% in 2010. Many of these job listings are concentrated in the professional, scientific, and technical services sector. These rising skill requirements are consistent with an economy in which productivity gains are concentrated in certain sectors, increasing demand for workers who can effectively use new technologies while leaving other labor-intensive roles largely unchanged. In response, higher education is shifting from debating the merits of AI to its adoption. A recent survey found that 94% of higher education workers are regularly using AI. At the institutional level, adoption is also becoming formal and large-scale: Reuters reports that OpenAI is rolling out an education-focused version of ChatGPT across the California State University system, covering all 23 campuses and roughly 500,000 students and faculty. In Hawaiʻi, UH recently launched a new Office of Academic Technology and Innovation to help all ten campuses integrate AI and other emerging tools—building AI literacy for students and faculty and tying that work to workforce needs.

Furthermore, evidence points to a clear pattern—education still remains a strong predictor of labor market success. In fact, in the emerging AI driven labor market, the demand for college educated workers is expected to grow. The Georgetown University Center on Education and the Workforce projects that 72% of all US jobs will require education beyond high school by 2031, up from 67% in 2021. In Hawai‘i, the share is expected to be similar at 70% in 2031.

AI is transforming the nature of work in uneven ways—raising productivity and skill demands in some sectors while leaving others heavily reliant on human labor—but not the relevance of higher education. Postsecondary education still opens doors to higher-paying, more adaptable careers, and that pattern holds in Hawai‘i as it does nationally. As technology continues to evolve, investing in higher education remains a reliable way to ensure workers can evolve with it. The challenge ahead is not whether education matters, but how institutions, employers, and policymakers can align programs to prepare workers for the labor market in the AI era.