Robot Exposure Ranks 774 Jobs, Nearly the Reverse of Generative AI

Pile-driver operators top a new ranking of 774 US occupations by exposure to robotics, followed by sewing machine operators and hand sewers. Foreign language professors sit at the bottom, just below political science professors and telemarketers. The index, posted to arXiv on October 16, 2026 by Sebastian Sartor and Neil Thompson of MIT and Robert Seamans of NYU, is the first to score robotics and physical AI against the abilities each job needs, across occupations, industries and counties.

How the index is built

[Fact] The authors assembled 16 robot capabilities in eight domains, such as grasping and handling or terrain-based locomotion, from 78 academic, industry and standards documents. An open-weight language model, GPT-OSS-20B, rated how closely each capability relates to each of the 52 work abilities in the O*NET database, 832 pairs in all, averaged over four expert personas and five samples per pair. Those scores were rolled up into an occupation index (ROE), an index for 250 NAICS industries and a county map. The method copies the one Seamans and co-authors built for software AI in 2018 and 2021, which makes the two directly comparable.

Three findings

1. Robots and generative AI point at opposite jobs. [Fact] Across 22 job families the robotics ranking and a published generative-AI exposure ranking run almost exactly in reverse (Spearman rho = -0.93). Construction and extraction is the most robot-exposed family and the least AI-exposed; computer and mathematical work is the reverse. Office and administrative support, the largest family at 17.4 million workers, scores -0.69 on robotics and +0.69 on AI. [Estimate] In the paper's family table, healthcare practitioners are the only family above average on both, and only barely (+0.06 and +0.12).

2. Exposure falls with pay, schooling and creativity. [Fact] Robot exposure declines as median wage rises (rho = -0.42), more steeply as required education rises (rho = -0.71), and most steeply with the weight O*NET puts on creative abilities (rho = -0.85). It is also higher in jobs with few women (rho = -0.55). Generative-AI exposure moves the other way on wage, education and creativity.

3. Fewer workers sit in the high-exposure group. [Fact] Using 2019 county employment, 18% of matched workers are in high robotics-exposure industries, against 43% for generative AI. The share runs from 7% in the District of Columbia and 12% in New York to 23% in Indiana, 24% in Kentucky and 31% in Wyoming. In New York City, 62% of workers are in low robotics-exposure industries while 58% are in high AI-exposure ones.

Capability, not today's robots

[Fact] The index agrees with a robot index Anthropic researchers published this year (rho = 0.78 across all 774 occupations), but the two split on specific jobs. Dentists and optometrists score high here and near zero on Anthropic's measure, which only counts tasks robots have already been shown to do. Bank tellers score higher on Anthropic's. Airline pilots rank 14th here.

[Fact] No single capability drives the ranking: dropping the most influential one moves the average occupation 2.5 percentile points. Dropping the whole manipulation domain moves it 11.6 points.

Which jobs this touches

The top-ranked groups include heavy and tractor-trailer truck drivers, sewing machine operators, construction equipment operators and industrial truck and tractor operators. Telemarketers rank 772nd of 774 on robots. Dentists and tellers show how much the answer depends on the method.

What to prepare

• Check both maps. A job low on one index is often high on the other. Look up your occupation on robotics and generative-AI measures before planning a move.
• Watch manipulation progress. The index is most sensitive to hand-like handling. Pilot projects with general-purpose grasping robots in your workplace are the signal to track.
• Lean on tasks that score low. Occupations weighted toward creative abilities had the lowest robotics scores of any group in the paper's comparisons.

Limits

The index measures technical relatedness, not readiness, cost or adoption, and the authors call it an upper bound; it says nothing about when or whether robots arrive. Scores come from one language model's judgments, and a seven-person survey found humans rated 82% of pairs as more related, so the ranking is more reliable than the levels. It is a working paper, not yet peer reviewed.

Sources

• Sebastian Sartor, Robert Seamans and Neil C. Thompson, "How exposed is work to robotics? Occupational, industry, and geographic measures", arXiv:2610.24267, October 16, 2026: https://arxiv.org/abs/2610.24267

About this analysis

This article was written with AI assistance, from the full text of the paper including its appendix. The paper is released under CC BY 4.0. The one figure marked [Estimate] (healthcare practitioners as the only family above average on both indices) is our reading of the paper's family table.