Harnessing AI for Market Forecasting

The COVID-19 pandemic and accompanying policy measures caused economic disturbance so plain that advanced analytical techniques were unneeded for many questions. Joblessness jumped dramatically in the early weeks of the pandemic, leaving little room for alternative descriptions. The impacts of AI, however, might be less like COVID and more like the web or trade with China.

One common method is to compare outcomes between basically AI-exposed employees, companies, or industries, in order to separate the result of AI from confounding forces. 2 Direct exposure is usually defined at the job level: AI can grade homework but not manage a classroom, for example, so instructors are considered less unwrapped than employees whose whole task can be carried out remotely.

3 Our approach combines data from 3 sources. Task-level direct exposure price quotes from Eloundou et al. (2023 ), which measure whether it is in theory possible for an LLM to make a task at least two times as quick.

Leveraging AI for Predictive Intelligence

4Why might actual use fall brief of theoretical capability? Some jobs that are theoretically possible might disappoint up in use due to the fact that of design restrictions. Others may be sluggish to diffuse due to legal restrictions, specific software requirements, human verification steps, or other obstacles. For instance, Eloundou et al. mark "Authorize drug refills and provide prescription details to drug stores" as completely exposed (=1).

As Figure 1 shows, 97% of the jobs observed across the previous 4 Economic Index reports fall into categories rated as in theory practical by Eloundou et al. (=0.5 or =1.0). This figure shows Claude use dispersed throughout O * NET jobs organized by their theoretical AI direct exposure. Tasks rated =1 (totally possible for an LLM alone) represent 68% of observed Claude usage, while tasks rated =0 (not feasible) account for just 3%.

Our new procedure, observed direct exposure, is implied to quantify: of those tasks that LLMs could theoretically accelerate, which are actually seeing automated usage in expert settings? Theoretical ability encompasses a much wider series of tasks. By tracking how that space narrows, observed exposure provides insight into economic changes as they emerge.

A task's direct exposure is greater if: Its tasks are in theory possible with AIIts tasks see substantial usage in the Anthropic Economic Index5Its tasks are carried out in job-related contextsIt has a reasonably greater share of automated usage patterns or API implementationIts AI-impacted tasks make up a bigger share of the total role6We give mathematical details in the Appendix.

Evaluating Traditional Models and Global Units

The task-level protection measures are averaged to the profession level weighted by the portion of time spent on each task. The measure shows scope for LLM penetration in the bulk of tasks in Computer & Mathematics (94%) and Workplace & Admin (90%) occupations.

The coverage reveals AI is far from reaching its theoretical capabilities. For instance, Claude currently covers just 33% of all tasks in the Computer & Mathematics category. As capabilities advance, adoption spreads, and deployment deepens, the red area will grow to cover the blue. There is a large uncovered area too; many jobs, of course, stay beyond AI's reachfrom physical farming work like pruning trees and running farm equipment to legal tasks like representing customers in court.

In line with other data revealing that Claude is thoroughly used for coding, Computer system Programmers are at the top, with 75% coverage, followed by Consumer Service Representatives, whose primary tasks we increasingly see in first-party API traffic. Data Entry Keyers, whose primary task of checking out source documents and going into information sees significant automation, are 67% covered.

Harnessing AI for Market Forecasting

At the bottom end, 30% of workers have zero protection, as their jobs appeared too occasionally in our data to satisfy the minimum threshold. This group includes, for instance, Cooks, Bike Mechanics, Lifeguards, Bartenders, Dishwashers, and Dressing Space Attendants. The United States Bureau of Labor Statistics (BLS) releases routine employment projections, with the most recent set, released in 2025, covering predicted modifications in work for every single occupation from 2024 to 2034.

A regression at the profession level weighted by present work discovers that growth forecasts are somewhat weaker for jobs with more observed direct exposure. For every 10 percentage point increase in coverage, the BLS's development projection visit 0.6 percentage points. This provides some validation in that our measures track the independently derived price quotes from labor market experts, although the relationship is minor.

measure alone. Binned scatterplot with 25 equally-sized bins. Each solid dot shows the average observed exposure and projected employment modification for among the bins. The dashed line reveals an easy linear regression fit, weighted by existing work levels. The small diamonds mark specific example occupations for illustration. Figure 5 shows characteristics of workers in the top quartile of exposure and the 30% of employees with zero exposure in the 3 months before ChatGPT was launched, August to October 2022, using data from the Present Population Survey.

The more bare group is 16 portion points most likely to be female, 11 portion points most likely to be white, and nearly two times as likely to be Asian. They earn 47% more, usually, and have higher levels of education. For example, people with academic degrees are 4.5% of the unexposed group, but 17.4% of the most unwrapped group, an almost fourfold distinction.

Brynjolfsson et al.

( 2022) and Hampole et al. (2025) use job utilize task publishing Burning Glass (now Lightcast) and Revelio, respectively. We focus on joblessness as our top priority outcome due to the fact that it most directly captures the potential for economic harma worker who is jobless wants a job and has actually not yet discovered one. In this case, task posts and employment do not always signal the requirement for policy actions; a decrease in task postings for an extremely exposed function may be counteracted by increased openings in a related one.