READING ROOM ↗
PSYBORG / WORK & SOCIETY

Exposure is not destiny.

Read AI-and-jobs research without turning task exposure into a job-loss forecast. Use a practical task map to examine possible changes with the people doing the work.

Modeled occupational exposure · Edited by PsyBorg · Updated 2026-11-16

Identify what the study measures

An AI-and-jobs headline may combine several different questions. Could a technology perform part of a task? Will an employer adopt it? How will the work be reorganized? What happens to the people doing it?

The ILO's May 2025 working paper, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, examines task-level potential. Its method combines worker assessments, expert review and model-generated scores to estimate occupational exposure. It specifically studies generative AI. [1]

Keep that study's date and definition of exposure beside any finding you repeat. A task-based estimate of technological potential does not itself tell a reader whether a particular employer will adopt a tool or dismiss a particular worker.

Follow the task through the whole job

The paper describes jobs as bundles of tasks. Its authors judge job transformation to be the most likely broad impact because many occupations include work requiring human input. That is the authors' assessment, not a guarantee about any one workplace. [1]

For a specific job, PSYBORG suggests following a task from its inputs to its final outcome. Who supplies the information? Who checks the output? Who deals with unusual cases, corrections or complaints? Ask which parts of that chain a proposed tool actually changes.

A hypothetical example: drafting a service reply

Imagine a service team considering a tool that drafts replies to routine questions. Producing a first draft is one task. Checking the customer's circumstances, applying the right policy, correcting errors and resolving a dispute are other parts of the workflow.

PSYBORG would ask the team to observe the work before assuming that faster drafting makes the whole service faster. A pilot might reveal useful assistance, extra checking work or a mixture of both. Its outcome needs to be measured rather than supplied in advance by an exposure score.

This is a hypothetical example. It does not predict the performance of a product, the number of staff needed or the outcome of a real deployment.

Build a task map with the people doing the work

For a proposed change, PSYBORG suggests a small, jointly reviewed worksheet. Give each important task its own entry and record the following.

  1. Current work: describe the task, its inputs and the people who depend on its result. Include exceptions and correction work.
  2. Proposed assistance: state what the AI would do and which decisions would remain with people.
  3. Evidence needed: identify what would show acceptable quality in this setting, including important failure cases.
  4. Work and skills: record expected checking, escalation, training and workload changes, then compare them with observations from a controlled pilot.
  5. Worker experience: agree how people can report problems and how changes to autonomy, workload and service quality will be assessed.

Mark estimates as estimates. If the team has not measured an effect, leave it open and assign a way to investigate. Avoid treating the amount of text produced as a complete measure of useful work.

Keep the distribution of effects visible

A workplace average can leave useful questions unanswered. PSYBORG recommends examining who receives assistance, who takes on checking work and who handles the difficult cases. Include workers' accounts alongside output and quality measures when reviewing the pilot.

The next decision can then be specific: continue the test, change the workflow, provide more training or stop using the tool for that task. Explain the evidence and trade-offs behind that choice.

An exposure study helps frame these questions. The experience of a particular workplace needs its own evidence, gathered with the people affected.

Sources and scope

[1] Gmyrek and colleagues, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, ILO Working Paper 140, published 20 May 2025. See Abstract, Introduction and Task-based approaches to automation in the literature. DOI: 10.54394/HETP0387.
Read source (www.ilo.org)

Read the full working paper online
Read source (webapps.ilo.org)

Sources checked 16 November 2026. This briefing discusses the named 2025 study; it does not label it the latest forecast. The task-map exercise and service-team example are PSYBORG editorial analysis. No job-loss estimate, individual employment prediction or measured productivity gain is asserted.

Continue reading

A benchmark score needs a context.
Read source (psyborg369.com)

A human in the loop needs a real way to act.
Read source (psyborg369.com)

SOURCE & FURTHER READINGILO · Generative AI and Jobs, 20 May 2025

Get the next briefing by email