ILO Working Paper 140 (2025): Generative AI and Jobs: A Refined Global Index of Occupational Exposure
Task-level occupational exposure framework for generative AI, built from expert input and model predictions.
OPEN SOURCE ↗Therapy works because of the relationship between two humans, not the content of the conversation. AI can simulate the conversation. It cannot create the relationship.
The therapeutic relationship — the working alliance — is the strongest predictor of therapeutic outcome, accounting for 30-a significant share of variance in treatment success (Wampold, the coming years; Norcross, the coming years). This relationship is built on genuine human mutuality: the therapist is also a mortal being who has experienced loss, fear, love, and failure.
AI systems like Woebot and Wysa provide valuable psychoeducation and CBT exercises, and the evidence for their efficacy in mild-moderate anxiety/depression is real. But these systems perform cognitive tasks, not relational healing.
For trauma, personality disorders, grief, and existential crisis — the majority of clinical mental health work — a human being who is genuinely present and genuinely responsive is not interchangeable with a sophisticated language model.
Furthermore, the global mental health crisis is growing. Demand for therapists is increasing at a significant share per decade while supply grows at a significant share. AI supplements the shortage; it does not and cannot replace the core of the profession.
The current deployment and policy evidence has 2-year waiting lists. The US has large numbers people who cannot access mental health care. The job is not dying — it is in crisis because there are not enough people doing it.
These are the genuine threats to this profession. They are real, but they are not sufficient to overturn the fundamental analysis. Here is why.
Put the case that Mental Health Therapist will not survive AI displacement. The system responds with counterarguments from the research base. Strong arguments shift the score — up to a maximum of ±15 points. The system is not an AI. It is a structured argument engine.
This question layer is generated from the job verdict, the resistance case, the regional rollout logic, and the evidence status of this page. Use the filters to focus the discussion, or trigger a random question and work through the role from multiple angles.
Replace broad inference with occupation-specific literature, regulators, labour statistics, or professional-body evidence before publication-grade use.
TIER 1 review queue with 7 core sources and 3 framework signals.
This page is grounded in task exposure research and labour-market trend reports, then translated into a reasoned occupation-level argument.
This site now treats exact timelines, total job-loss counts, and regional speed as interpretive estimates unless a cited source states them directly. The argument on this page should be read as a structured forecast, not a guaranteed future.
These impact figures are site estimates for comparison and should not be read as official labour-market counts.
Task-level occupational exposure framework for generative AI, built from expert input and model predictions.
OPEN SOURCE ↗Finds clerical work is the most highly exposed occupational group and that augmentation is often more likely than full occupation automation.
OPEN SOURCE ↗Shows AI exposure is highest in many white-collar cognitive occupations, while manual occupations tend to have lower exposure.
OPEN SOURCE ↗Advanced economies are more exposed to AI because they have more cognitive-intensive jobs; infrastructure and skills limit adoption elsewhere.
OPEN SOURCE ↗Large-employer survey showing clerical roles among the fastest-declining and care, education, software and green-transition jobs among growth areas.
OPEN SOURCE ↗Notes substantial automation risk remains, while observed labour-market effects remain mixed rather than universally destructive.
OPEN SOURCE ↗Argues advanced economies are better positioned to benefit from AI due to infrastructure, skills, and institutions.
OPEN SOURCE ↗