When AI Levels the Playing Field: Skill Homogenization, Asset Concentration, and Two Regimes of Inequality

Xupeng Chen; Shuchen Meng

cs.LGMar 5, 2026

When AI Levels the Playing Field: Skill Homogenization, Asset Concentration, and Two Regimes of Inequality

Authors: Xupeng Chen, Shuchen Meng

Organizations: Department of Electrical and Computer Engineering, New York University. · Department of Financial Engineering, New York University.

Abstract

Generative AI compresses within-task skill differences while shifting economic value toward concentrated complementary assets, creating an apparent paradox: the technology that equalizes individual performance may widen aggregate inequality. We formalize this tension in a task-based model with endogenous education, employer screening, and heterogeneous firms. The model yields two regimes whose boundary depends on AI's technology structure (proprietary vs. commodity) and labor market institutions (rent-sharing elasticity, asset concentration). A scenario analysis via Method of Simulated Moments, matching six empirical targets, disciplines the model's quantitative magnitudes; a sensitivity decomposition reveals that the five non- $Δ$ Gini moments identify mechanism rates but not the aggregate sign, which at the calibrated parameters is pinned by $m_6$ and $ξ$ , while AI's technology structure ( $η_1$ vs. $η_0$ ) independently crosses the boundary. The contribution is the mechanism -- not a verdict on the sign. Occupation-level regressions using BLS OEWS data (2019--2023) illustrate why such data cannot test the model's task-level predictions. The predictions are testable with within-occupation, within-task panel data that do not yet exist at scale.

Figures & tables

Panel of Figure 1 — Figure 1: Model-implied mechanism chain from AI capability to inequality. The equalizing channel operates through within-task compression. The concentrating channel operates through rising returns to complementary assets. Net inequality depends on AI’s technology structure (proprietary vs. commodity) and labor market institutions ( $\xi$ , $\mathrm{Gini}(K)$ ).

Panel A: Structural Parameters
	Value	SE	Description
$\alpha/(\phi\mu_{h})$	0.526	(0.096)	AI/human output ratio
$\sigma$	1.166	(1.639)	Task substitution elasticity
$\eta_{0}$	0.142	(0.058)	Pre-AI asset elasticity
$\eta_{1}$	0.323	(0.035)	Post-AI asset elasticity
$\mathrm{Gini}(K_{0})$	0.909	(0.132)	Asset concentration (Gini)

Table 1: Method of Simulated Moments: Calibrated Parameter Values

Derived Quantities
Parameter	Point Value	95% CI	Width
$\alpha/(\phi\mu_{h})$	0.526	[0.355, 0.744]	0.389
$\sigma$	1.166	[0.246, 5.000]	4.754
$\eta_{0}$	0.142	[0.095, 0.800]	0.705
$\eta_{1}$	0.323	[0.278, 0.832]	0.554
$\mathrm{Gini}(K_{0})$	0.909	[0.234, 0.980]	0.746

Table 2: Bootstrap 95% Confidence Intervals (

B=1000

)

Industry Type	Gini( $K$ )	$\xi$	$\Delta$ Gini
Tech / AI platforms	0.95	0.25	+0.0218
Finance / Banking	0.85	0.20	-0.0002
Healthcare	0.70	0.15	-0.0143
Manufacturing	0.65	0.15	-0.0163
Retail / Hospitality	0.50	0.10	-0.0231
Education / Government	0.40	0.07	-0.0258

Table 3: Predicted

\Delta

Gini by Industry Type

Panel A: Sensitivity $\partial\hat{\theta}_{j}/\partial m_{i}^{\mathrm{data}}$
	m1	m2	m3	m4	m5	m6
$\alpha/(\phi\mu_{h})$	+1.575	+0.681	+0.000	+0.000	-0.000	-0.000
$\sigma$	+10.854	+4.691	-0.000	+0.000	-149.991	-0.000
$\eta_{0}$	-0.080	-0.035	-0.221	+0.225	+0.000	-3.779
$\eta_{1}$	-0.034	-0.015	+0.050	+0.510	+0.000	-1.595
$\mathrm{Gini}(K_{0})$	+0.182	+0.079	-0.268	-0.510	-0.000	+8.574

Table 4: Moment–Parameter Sensitivity and Boundary Diagnostic

Figure 2: Net inequality effect in

(\eta_{1},\mathrm{Gini}(K_{0}))

space. Blue: equalizing channel dominates. Red: concentrating channel dominates. Solid black: knife-edge boundary at baseline. Dashed/dotted: boundaries under high/low AI scenarios. Star: calibrated value.

$\xi$	$\Delta$ Gini	Regime
0.00	-0.0143	Equalizing
0.05	-0.0131	Equalizing
0.07	-0.0119	Equalizing
0.10	-0.0093	Equalizing
0.15	-0.0033	Equalizing
0.18	+0.0009	Concentrating (boundary)

Table 5: Sensitivity of

\Delta

Gini to Rent-Sharing Parameter

\xi

Mechanism	Existing Evidence	Data Needed	Testable Implication
Task homog. (CV $\downarrow$ )	Brynjolfsson et al. (2024) : GenAI raised novice productivity $+$ 34% vs. $\approx$ 0% for experts; Noy and Zhang (2023) : quality CV $\downarrow$ 40%	Within-occupation task-level wages/outputs before & after AI adoption	DiD on within-occupation variance: high-AI-exposure tasks should show CV $\downarrow$
Education returns shift	Suggestive: Deming (2017) finds social-skill occupations gained faster; college premium plateau	Matched worker education & wages over time; CPS-ASEC $\times$ AIOE panel	Education-wage gradient for codifiable skills should decline in high-AIOE occupations ( $\sigma>1$ )
Credential inflation	Analog: Hershbein and Kahn (2018) find tech shocks raised credential requirements	Longitudinal job posting data (Lightcast); firm-level AI adoption	Higher AI task exposure $\to$ larger increases in required credentials (DiD), controlling for task content
Firm concentration ( $K$ returns $\uparrow$ )	Babina et al. (2024) : AI firms grow 12–18% faster; Autor et al. (2020) : top-4 share rising	Industry/firm concentration (HHI) over time; firm sales by AI use	Panel regression: concentration vs. AI adoption, controlling for demand shocks

Table 6: Evidence Map: Mechanism Links, Existing Evidence, and Required Tests

	Cross-Section (2023)		First Difference (2023 $-$ 2019)
	(1)	(2)	(3)	(4)	(5)
AIOE	0.1088***	0.0339**	0.0167***	0.0166***	0.0099
	(0.0083)	(0.0161)	(0.0035)	(0.0041)	(0.0083)
$\log$ (med wage) / $\Delta\log$ (med wage)		0.2445***		-0.0074	-0.0411
		(0.0327)		(0.0780)	(0.0788)
Design	Levels	Levels	FD	FD	FD

Table 7: AI Exposure and Within-Occupation Wage Dispersion: Levels vs. First Differences

H	Prediction	Source	Data	Outcome
H1	Within-occupation output dispersion declines; largest gains for low-ability	Prop. 1	Task-level productivity	CV, 90–10 gap
H2	Education premium declines in high-AI occupations for codifiable skills (requires $\sigma>1$ ; see text)	Prop. 3	CPS-ASEC $\times$ AIOE	Education-wage gradient
H3	Credential requirements rise in high-AI occupations, controlling for tasks (untested prediction; requires post-2023 Lightcast data linked to AI adoption)	Prop. 4	Lightcast postings	Degree requirement
H4	Between-firm wage dispersion rises; within-firm declines (conditional on $\psi>\eta_{0}$ , $\xi$ , $\mathrm{Gini}(K)$ )	Props. 5 , 1	LEHD matched data	Variance decomposition
H5	AI investment and profits concentrate in high- $K$ firms	Prop. 5 (a)	Compustat, ABS	HHI, profit share
H6	Net inequality change depends on AI technology structure ( $\psi$ vs. $\eta_{0}$ ), $\xi$ , and $\mathrm{Gini}(K)$	Sec. 3	Cross-industry/country	$\Delta\text{Var}(w)$ vs. HHI, $\xi$

Table 8: Testable Predictions

AI and Labor

AI and labor studies how artificial intelligence changes jobs, wages, skills, hiring, and the organization and quality of work.

See all 94 papers

Generative AI impacts on intra-urban inequality and skill premium in Beijing
May 25, 2026Xiliu He, Haoxiang Zhao, Mingyi Ma +6
School of Architecture, Tsinghua University, Beijing, China · ZODA LAB, Hangzhou, China · Technology Innovation Center for Smart Human Settlements and Spatial Planning & Governance, Ministry of Natural Resources, Tsinghua University, Beijing, China
Human-AI Productivity Paradoxes: Modeling the Interplay of Skill, Effort, and AI Assistance
May 12, 2026Ali Aouad, Thodoris Lykouris, Huiying ZhongAI Literacy Human-AI Interaction
Massachusetts Institute of Technology
Generative AI and the Productivity Divide: Human-AI Complementarities in Education
May 18, 2026Lihi Idan, Bharat AnandHuman-AI Complementarity AI in Education
Industrial and Systems Engineering Department, Texas A&M University · Leonard N. Stern School of Business, New York University

May 25, 2026J/K move · Enter open · S save

Generative AI impacts on intra-urban inequality and skill premium in Beijing

Xiliu He, Haoxiang Zhao, Mingyi Ma, Edward Wen Chuan Lai+5

School of Architecture, Tsinghua University, Beijing, China · ZODA LAB, Hangzhou, China · Technology Innovation Center for Smart Human Settlements and Spatial Planning & Governance, Ministry of Natural Resources, Tsinghua University, Beijing, China

Open paper AskPDF arXiv

Generative artificial intelligence (GenAI) is the first automation wave to reach high-cognitive tasks at scale, yet its effects on intra-urban inequality remain largely unknown. Using 5 million job postings from Beijing (2018--2024), we construct a neighborhood-level GenAI Exposure Index by aggregating task-level assessments from five leading large language models. We examine the spatial, structural and causal mechanisms of this shock. We find that GenAI exposure is highly concentrated in the city's core districts, deepening the intra-urban AI divide. Since 2023, high-exposure neighborhoods have experienced wage stagnation even as they continue to attract high-skilled workers -- a "high-skill trap." This wage penalty is driven by task de-skilling and intensified labor-market crowding. A difference-in-differences design centered on ChatGPT's release supports a causal interpretation. These findings challenge the prevailing theory of skill-biased technological change and provide a basis for inclusive AI governance in global technology hubs.

Abstract

AI and Labor

Explore similar work

Generative AI impacts on intra-urban inequality and skill premium in Beijing

Human-AI Productivity Paradoxes: Modeling the Interplay of Skill, Effort, and AI Assistance

Generative AI and the Productivity Divide: Human-AI Complementarities in Education