Research SummaryGlobal

External insight · International Labour Organization

AI Is More Likely to Transform Work Than Simply Replace It

The ILO’s refined global index finds broad potential exposure to generative AI while stressing that task and job transformation is more likely than simple replacement.

Generative AI and Jobs: A Refined Global Index of Occupational Exposure20 May 2025

Read the original source ↗

Conceptual visual

Independent editorial analysis

This is FUURAA’s own editorial analysis of the cited public source, prepared independently from the cited institution. Source materials remain attributable to their authors and publishers; FUURAA is responsible for their selection, synthesis and interpretation. No cited institution has reviewed or endorsed this article unless expressly stated.

External evidence

What the public source says

The ILO and Poland’s NASK estimate that one in four jobs worldwide has some potential exposure to generative AI, while 3.3% of global employment falls into the highest exposure category. Exposure is not the same as job loss.

The study’s central conclusion is that transformation is more likely than replacement. The effects will differ by occupation, task, country, gender and income level, requiring responses that are more precise than a single global forecast.

FUURAA editorial analysis

FUURAA editorial perspective

Evidence-led analysis in the public interest

The most useful question is not simply whether AI removes a job, but how work is decomposed, which tasks are augmented, where human judgment remains essential and how gains are shared.

Agent systems should support accountable human work: clear delegation, reviewable outputs, capability development and escalation when consequences are high.

Key judgments

Potential exposure to generative AI should not be interpreted as a forecast of job disappearance. The more defensible reading is that work will be reorganised unevenly across tasks, occupations and economies.
The quality of transition matters as much as the speed of adoption. Productivity gains are incomplete measures if workers lose agency, skills, access to review or a fair share of the resulting value.
AI-supported work should preserve accountable human judgment wherever consequences are material, while giving people the training and authority needed to question system outputs.

Reading exposure with care

The ILO and NASK estimate that one in four jobs worldwide has some potential exposure to generative AI, while 3.3% of global employment falls into the highest exposure category. These figures describe where technology may affect work; they do not establish how many jobs will disappear. The report instead points toward transformation as the more likely general outcome and stresses that effects vary by occupation, task, country, gender and income level. Public discussion should therefore resist turning a differentiated exposure index into a single headline about employment loss. A responsible interpretation separates technological possibility from organisational choice, economic conditions and the decisions made by employers, workers and policymakers.

The task is the practical unit of change

A job is rarely one indivisible activity. It normally combines routine processing, communication, contextual knowledge, judgment, responsibility and relationships. AI may assist some components while leaving others substantially human, or it may change the sequence through which the work is completed. The central design question is therefore not whether an occupation is labelled automatable, but which tasks can be delegated, which require review, what evidence supports an output and who remains accountable. This task-level view enables more precise decisions and reduces the risk of applying the same automation policy to work with very different consequences.

Transformation must include worker agency

Because the ILO identifies differences across countries, occupations, gender and income levels, transition policies cannot assume that benefits and burdens will be distributed automatically or evenly. Organisations should involve affected workers in workflow redesign, explain how systems are used and provide credible routes for correction and escalation. Training should extend beyond operating a tool: people also need to recognise limits, verify results and understand when human expertise must prevail. A fair transition is not achieved merely by retaining a human somewhere in the process; that person must have sufficient knowledge, time and authority to exercise meaningful oversight.

Agent systems should strengthen accountable work

Agent-based systems can make delegation more complex because they may plan or execute several steps before presenting a result. This increases the importance of defined permissions, visible task histories, reviewable outputs and escalation for high-impact decisions. Organisations should test whether an AI-supported workflow improves quality and resilience, not only whether it completes a task faster. Where outcomes affect livelihoods, rights or safety, responsibility cannot be displaced onto the system. The durable role of AI at work will depend on whether it helps people exercise better judgment and build capability rather than making important processes less understandable.

Alternative views & uncertainty

What this evidence does not settle

Caution should not become a presumption that all automation is harmful. Well-bounded uses can remove repetitive work or support employees, provided the claimed benefits and the human consequences are assessed rather than assumed.
Keeping a human reviewer in a workflow is not automatically protective. Oversight can become nominal if the reviewer lacks expertise, time, information or genuine authority to reject an output.

Public-interest implications

What this means for different stakeholders

Public

People need understandable information about how AI changes their tasks, what decisions remain human and how errors or unfair outcomes can be challenged.

Organisations / industry

Redesign work at task level, invest in supervision skills and measure quality, safety, inclusion and worker experience alongside efficiency.

Policy

Support transition measures that recognise differences across occupations and populations without treating exposure estimates as predetermined employment outcomes.

Research

Track how task composition, human authority and benefit distribution change over time, with attention to the groups identified as differently exposed.

What to watch next

Whether organisations measure transformed tasks and work quality rather than reporting tool adoption alone.
Whether training gives workers practical authority to verify, question and escalate AI-supported decisions.
Whether differences across occupations, countries, gender and income levels narrow or become more pronounced.

Conclusion

The ILO evidence supports neither complacency nor a simple narrative of mass replacement. It supports a more demanding agenda: examine work at task level, distinguish exposure from outcome and make the distribution of benefits and responsibility visible. AI can contribute to better work only when its use is accompanied by capable people, reviewable processes and institutions willing to measure human outcomes. FUURAA’s editorial position is that technological progress and worker dignity should be evaluated together, with conclusions revised as stronger evidence emerges.

Independence and relevance disclosure

This is an independent FUURAA editorial analysis of the cited public source. It does not represent the ILO or NASK, imply their endorsement of FUURAA, or extend their findings beyond the limits stated above.

Forward view

Practical priorities

Redesign tasks

Separate routine automation from judgment, relationships and high-impact decisions.

Invest in capability

Training should help people supervise, question and improve AI-supported work.

Measure outcomes

Evaluate quality, safety, inclusion and worker experience—not only time saved.