cs.AINov 4, 2026
SaveDull, Dirty, Dangerous? Working Conditions and the Limits of Occupational AI Exposure
Organizations: Department of Marketing, Daniels School of Business, Purdue University
Abstract
Automation is often justified as relief from dull, dirty, and dangerous work. I examine how these conditions relate to occupational AI exposure using 21 O*NET items across 754 US occupations. Higher exposure is associated with lower condition scores under a large language model (LLM) measure and AI Occupational Exposure (AIOE), with equal occupation weights and employment weights. Adjustment for physical demands, required education, training and experience, and wages reduces the composite association. The adjusted confidence intervals include zero for LLM exposure and remain below zero for AIOE. Separating repetition from restricted discretion also changes what the combined Dull score conveys. These comparisons show how item groupings and occupational characteristics shape the interpretation of exposure estimates. Occupational profiles offer a starting point for selecting contrasting research settings. Establishing desired automation or realized relief requires evidence about workers' preferences and specific interventions.
Figures & tables
| Study | Evidence and question | Relation to this study |
|---|---|---|
| Pizzinelli et al. [21] | O*NET contexts, Job Zones, and AIOE; potential complementarity | Shares occupational inputs; measures complementarity rather than the selected condition burdens. |
| Parteka et al. [20] | Technology exposure and multidimensional working conditions in Europe | Establishes a close precedent for the broad condition–exposure association. |
| Giuntella et al. [13] | German longitudinal data; AI exposure, wellbeing, health, and physical job intensity | Examines worker outcomes over time; the present linkage compares occupational scores. |
| Ranjit et al. [23] | Task-level meaning, busywork, and AI exposure | Examines valuation of computer-assisted tasks; this study examines physical and organizational conditions. |
| Welfare et al. [31] , Shao et al. [25] | Workers’ valued activities and desired assistance or delegation | Provides preference evidence beyond the scope of occupational condition ratings. |
| Present study | Selected condition scores, two AI exposure measures, and specification comparisons | Examines adjustment, item grouping, and the interpretation of occupational profiles. |
Table 1: Closest empirical comparisons. Differences in constructs and research questions define the present study’s scope.
| Source | Version and unit | Fields or use |
|---|---|---|
| O*NET | 31.0, August 2026 release; detailed O*NET-SOC | Work Context ratings, suppression flags, Job Zones, task statements, SOC crosswalk [ 30 ] |
| Eloundou et al. | Published exposure data; detailed O*NET-SOC | GPT-4-rated dv_rating_alpha , beta , gamma columns in occ_level.csv [ 9 ] |
| AIOE | 2021 dataset, SOC 2010 | Appendix A of AIOE_DataAppendix.xlsx [ 10 ] |
| BLS OEWS | May 2021; detailed occupation | TOT_EMP , A_MEDIAN ; employment weights and wages [ 29 ] |
| BLS CPS | 2022 annual Table 11; occupation category | Sex, race, and Hispanic or Latino composition; annual workforce shares |
| BLS CFOI | 2024 Table A-5; occupation title | Fatality counts for an exploratory measurement check |
Table 2: Data sources and observation periods. Employment and workforce composition come from BLS files distributed in the public Eloundou et al. repository ( https://github.com/openai/GPTs-are-GPTs ). BLS publishes the CFOI tables at https://www.bls.gov/iif/fatal-injuries-tables.htm .
| Analysis | Occupations | Employment coverage |
|---|---|---|
| Condition scores and both AI exposure measures | 780 | — |
| Primary comparison, both weighting schemes | 754 | 92.0% national |
| Common sample for sequential adjustment | 736 | — |
| CPS title matches with complete composition data | 244 | 72.1% |
| Fatality-ratio diagnostic | 170 | — |
| Hybrid exposure linkage | 753 | 99.7% |
Table 3: Samples used in each analysis. Employment coverage is relative to the primary sample unless labeled national.
Figure 1: AI exposure and working-condition scores in the primary sample. Points are regression coefficients, with 95% confidence intervals based on CR1 covariance and a reference distribution. Every model includes major-group fixed effects. Circles denote equal occupation weights; squares denote employment weights. Exact values appear in Appendix Table 6 .
Figure 2: Adjustment models on the same complete sample. Each stage retains major-group fixed effects. Physical demands enter as a predictor at the second stage; Job Zone indicators and log median wage enter at the third. Circles denote equal occupation weights and squares employment weights. Exploratory 95% intervals use CR1 covariance and a reference distribution.
Figure 3: The Dull domain separated into repetition and machine pacing, and restricted discretion. Each component is standardized separately. Circles denote equal occupation weights and squares employment weights; exploratory 95% intervals use CR1 covariance and a reference distribution. Models retain major-group fixed effects.
| Definition | Occupations | Jobs | Share |
|---|---|---|---|
| High score, low LLM exposure | 300 | 41,422,280 | 32.0% |
| High score, low AIOE | 315 | 47,356,980 | 36.6% |
| High score, low on both (medians) | 286 | 40,957,610 | 31.6% |
| High score, low on both (tertiles) | 157 | 20,164,070 | 15.6% |
Table 4: Occupations with higher composite scores and lower AI exposure under alternative relative cutoffs. All definitions use the primary sample and the same employment denominator; cutoffs give equal weight to occupations.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| Domain | Element ID | Condition |
|---|---|---|
| Dangerous | 4.C.2.c.1.a | Radiation |
| 4.C.2.c.1.b | Disease or infections | |
| 4.C.2.c.1.c | High places | |
| 4.C.2.c.1.d | Hazardous conditions | |
| 4.C.2.c.1.e | Hazardous equipment | |
| 4.C.2.c.1.f | Minor burns, cuts, bites, or stings |
Table 5: Items in the primary working-condition index. The table abbreviates source wording; exact item names and response anchors are available from O*NET. The two discretion items are reverse-scored.
| Outcome | Weight | LLM exposure | AIOE |
|---|---|---|---|
| Composite | Occ. | 0.350 [ 0.470, 0.229] | 0.528 [ 0.715, 0.341] |
| Emp. | 0.285 [ 0.362, 0.208] | 0.425 [ 0.517, 0.332] | |
| Dangerous | Occ. | 0.322 [ 0.460, 0.183] | 0.413 [ 0.584, 0.243] |
| Emp. | 0.187 [ 0.313, 0.061] | 0.300 [ 0.439, 0.161] | |
| Dirty | Occ. | 0.307 [ 0.426, 0.188] | 0.425 [ 0.611, 0.240] |
| Emp. | 0.202 [ 0.305, 0.100] | 0.334 [ 0.508, 0.159] |
Table 6: Complete primary estimates: coefficient [95% confidence interval], in condition-score standard deviations per AI-exposure standard deviation. All models include SOC major-group fixed effects and use 754 occupations and 22 clusters. Occ. denotes equal occupation weights; Emp. denotes employment weights. Intervals use CR1 covariance and a reference distribution.
| Set | Women | Black | Asian | Hispanic |
|---|---|---|---|---|
| Higher score, lower exposure | 36.9 | 16.2 | 4.2 | 26.2 |
| Lower score, higher exposure | 56.1 | 10.9 | 7.7 | 13.5 |
Table 7: Workforce composition in the CPS-matched subsets. Entries are occupational workforce percentages averaged using OEWS employment weights. Both groups use the same joint classifications as the main analysis. Race and Hispanic ethnicity overlap.