Technical appendix for 'How might AI affect Texas’ good jobs?'

Anna Crockett

Parameters

This research project explores how exposed certain jobs are to artificial intelligence (AI), specifically generative AI. I do not consider any future advancements in robotics that could be combined with artificial intelligence. | Read the article.

Data source

I use the Department of Labor’s O*NET database of job and worker characteristics. I specifically employ the “skills” and “work activities” variables. There are a total of 35 skills and 41 work activities in the O*NET database.

AI assistance categories

A panel of Dallas Fed community development researchers sorted all skills and work activities into three AI assistance categories: low, medium, and high. The categorization was based on the following definitions:

AI Assistance (skills)

Low	Gen AI could not do well or at all.
Medium	Gen AI currently possesses this skill, but it is not advanced.
High	Gen AI currently possesses this skill at an advanced level.

AI Assistance (work activities)

Low	Gen AI AI has very minimal ability to conduct work activity, if at all.
Medium	Gen AI could assist with a work activity, but it would mostly rely on a human to complete.
High	Gen AI could complete the work activity with minimal work for a human.

I also asked an AI platform, Perplexity Pro Search, to produce its own categorization of each skill and work activity. I used prompts similar to those employed by Kazinnick and Brynjolfsson (2025). The AI-generated results were used as one piece of feedback for our categorization; in some cases this led to the reassignment of categories, and in other cases it did not.

Each skill and work activity sorted into the “high” assistance category received an assistance score of 1. Each sorted into the “medium” assistance category received an assistance score of 0.67. And each sorted into the “low” assistance category received an assistance score of 0.33.

Calculating the AI exposure scores

Both the O*NET’s skills and work activities variables contain two separate scores: a “level” score and an “importance” score. The level score indicates the degree of mastery a worker should have of the skill or work activity. The importance score tells us how integral the skill or work activity is to the occupation.

I use these four scores, the skill level score, the skill importance score, the work activity level score and the work activity importance score, in my analysis.

First, for each variable (skill and work activity), I take the square root of the product of the level and importance scores. This gives us the general skill score and work activity score.

Skill score = √(Skill level score * Skill importance score)
Work activity score = √(Work activity level score * Work activity importance score)

I then take the square root of the product of each skill/work activity score and its corresponding assistance score to calculate the AI exposure score for each skill and work activity.

Skill exposure score = √(Skill score * AI assistance score)
Work activity exposure score = √(Work activity score * AI assistance score)

Occupation skill and work activity exposure scores

To calculate each occupation’s final skill exposure and work activity exposure scores, I sort each variable by degree of importance. I then take the average of the top ten most important skill/work activity exposure scores to calculate the final skill and work activity exposure score for each occupation. See examples below.

Table 1. Top 10 skills of heavy and tractor-trailer truck drivers (ranked by importance)

Skill	Level	Importance	Skill score	Assistance category	Assistance score	Exposure score
Operation and control	0.52	0.69	0.60	Low	0.33	0.45
Operations monitoring	0.45	0.69	0.56	High	1	0.75
Monitoring	0.43	0.53	0.48	Medium	0.67	0.56
Critical thinking	0.41	0.50	0.45	Medium	0.67	0.55
Reading comprehension	0.43	0.50	0.46	High	1	0.68
Speaking	0.41	0.50	0.45	High	1	0.67
Time management	0.43	0.50	0.46	Medium	0.67	0.56
Troubleshooting	0.41	0.50	0.45	Medium	0.67	0.55
Active listening	0.43	0.47	0.45	Low	0.33	0.39
Equipment maintenance	0.39	0.47	0.43	Medium	0.67	0.53
Average skill exposure score: 0.57
NOTE: In cases of several skills scoring the same on importance, they were then ranked by level scores.

Table 2: Top 10 work activities of computer user support specialists (ranked by importance)

Work activity	Level	Importance	Work activity score	Assistance category	Assistance score	Exposure score
Working with computers	0.80	0.95	0.87	High	1	0.93
Getting information	0.71	0.86	0.78	High	1	0.88
Communicating with supervisors, peers or subordinates	0.83	0.84	0.83	High	1	0.91
Making decisions and solving problems	0.63	0.83	0.72	Medium	0.67	0.69
Updating and using relevant knowledge	0.79	0.82	0.80	Medium	0.67	0.73
Processing information	0.65	0.77	0.71	High	1	0.84
Documenting/recording information	0.57	0.73	0.65	High	1	0.80
Establishing and maintaining interpersonal relationships	0.64	0.70	0.67	Low	0.33	0.47
Interpreting the meaning of information for others	0.56	0.70	0.63	High	1	0.79
Monitoring processes, materials or surroundings	0.63	0.68	0.65	Medium	0.67	0.66
Average work activities exposure score: 0.77
NOTE: In cases of several work activities scoring the same on importance, they were then ranked by level scores.

AI exposure spectrums

I created separate spectrums of AI exposure for both skills and work activities and identified two occupations to act as upper and lower bounds on these spectrums: locker room, coatroom and dressing room attendants (lower) and computer programmers (upper). These were identified by calculating exposure scores for a variety of occupations that previous literature using similar methodology had highlighted as high or low exposure. Computer programmers and locker room, coatroom and dressing room attendants had the highest and lowest scores, respectively, of these occupations.

References

Auer, R., D. Kopfer, and J. Sveda (2024). The rise of generative AI: modelling exposure, substitution, and inequality effects on the US labour market. BIS Working Paper No 1207, Bank for International Settlements.
Eloundou, T., S. Manning, P. Mishkin, and D. Rock (2023). GPTs are GPTs: An early look at the labor market impact potential of large language models.
Felten, Edward W. and Raj, Manav and Seamans, Robert, How will Language Modelers like ChatGPT Affect Occupations and Industries? (March 1, 2023). Available at SSRN: https://ssrn.com/abstract=4375268 or http://dx.doi.org/10.2139/ssrn.4375268
Hampole, M., D. Papanikolaou, L. D. W. Schmidt, and B. Seegmiller (2025). Artificial intelligence and the labor market. NBER Working Paper 33509, National Bureau of Economic Research.
Kazinnik, S. and E. Brynjolfsson (2025). AI and the Fed. NBER Working Paper 33998, National Bureau of Economic Research.
Kochhar, R. (2023). Which U.S. Workers Are More Exposed to AI on Their Jobs? Pew Research Center.
Tomlinson, K., S. Jaffe, W. Wang, S. Counts, and S. Suri (2025). Working with AI: Measuring the Occupational Implications of Generative AI.

About the authors

Anna Crockett is a senior advisor in Community Development at the Federal Reserve Bank of Dallas.

The views expressed are those of the authors and should not be attributed to the Federal Reserve Bank of Dallas or the Federal Reserve System.