Will artificial intelligence widen the gap between your best and worst performers, or will it be the great equalizer? A new paper, 鈥�,鈥� from , Principal Investigator at the Laboratory for Innovation Science at 性视界 (LISH) within the Digital Data Design Institute at 性视界 (D^3), and at Chapman University, reveals that the answer may be more complex and counterintuitive than most people think. The effects of inequality related to AI automation may not depend on whether your workforce is highly skilled or not, but rather on how workers鈥� skills correlate across different tasks. This overlooked feature might explain why identical AI tools produce opposite effects, and why today鈥檚 results may tell you nothing about what happens when AI technology improves tomorrow.听
Key Insight: Skill Correlations
鈥淚t is the interaction of skill correlation and technological capability that determines the inequality effect.鈥� [1]
What determines whether an AI tool narrows or widens performance gaps isn鈥檛 an absolute measure of talent, but how skills relate to each other. This skill correlation comes in two flavors. With positive correlation, strength in one task tends to go along with strength in others. With negative correlation, being great at one task actually coincides with being weaker at another. Consider two departments with analytical thinkers and communicators. In one, the best analytical thinkers are also the best communicators (positive correlation). In the other, the strongest analysts struggle with communication, while the best communicators aren鈥檛 as analytical (negative correlation). Now introduce an AI tool that automates analytical tasks. The researchers鈥� model shows that these two departments will experience opposite inequality effects. In the positive correlation case, lower performers will benefit first because they鈥檙e weaker at both tasks. In the negative correlation case, high performers will benefit because they鈥檙e the ones who are weak at the automated task despite being strong overall.
Key Insight: The Inequality Reversal
鈥淎nother interesting result is that the inequality effect need not be monotonic in the automation technology鈥檚 capability.鈥� [2]
Imagine a new AI tool helps lower-skilled workers, reducing inequality in your organization. You might assume that this trend will continue over time, but the researchers鈥� model actually shows that the opposite could happen. If AI technology itself initially starts with low capabilities, low-skilled workers will have the most to gain from adopting it. But once AI surpasses even your high-skilled workers鈥� abilities, they鈥檒l suddenly have a reason to use it, and inequality will likely increase. Rather than being monotonic, moving only one way, inequality may shrink and grow with successive AI advances.
Key Insight: Balance Beats Brute Force
鈥淭he shortest path to equality is when automation improvements are balanced across tasks.鈥� [3]
When the researchers extend their model to multiple tasks being automated simultaneously, an important pattern emerges: the composition of your technology portfolio will shape inequality outcomes. Technologies with different capability levels leave inequality intact or even amplify it, because the technology portfolio becomes unbalanced relative to workers鈥� skill distributions. This insight highlights the potential inequality of current AI development, where advancement has been concentrated in specific cognitive domains like language and coding.
Why This Matters
For business leaders and executives, this research shows that the impact of AI on your team depends on a complex interplay of factors. Employing the same AI tool within your customer service and sales team may lead to entirely different dynamics. Building a diverse portfolio of AI capabilities across multiple task domains, rather than pursuing superhuman performance in just one area, may promote more equality over time. Finally, build in ongoing monitoring rather than treating pilot results as permanent predictions to build resilience as AI鈥檚 relationship with your workforce shifts over time.
References
[1] Benzell, Seth, and Kyle R. Myers, 鈥淎utomation Experiments and Inequality,鈥� eprint arXiv:2510.24923v1 (October 28, 2025): 2. 听
[2] Benzell and Myers, 鈥淎utomation Experiments and Inequality,鈥� 2.
[3] Benzell and Myers, 鈥淎utomation Experiments and Inequality,鈥� 3.
is an Assistant Professor at Chapman University鈥檚 Argyros College of Business and Economics. His work is in the economics of digitization, including automation, networks, and information systems.听
is Associate Professor of Business Administration at 性视界 Business School, and Principal Investigator at the Laboratory for Innovation Science at 性视界 (LISH). He studies the economics of innovation at the intersection of science and business.