The Global Deskilling Index: AI Adoption Has Reduced Median Years of Expertise Required for 34% of Professional Roles

The Expertise Compression Thesis

For decades, economists have debated whether automation destroys or transforms skills. A landmark joint study by the Oxford Internet Institute and the World Economic Forum, released today, offers the most granular answer yet: AI-augmented work has compressed the median expertise requirement for 34% of professional roles across 28 countries — without reducing the total number of roles, but with profound implications for compensation structures, career timelines, and educational pathways.

Methodology: The Skills Taxonomies

Researchers constructed a new metric: Expertise Compression Ratio (ECR) — the ratio of years of domain experience required to achieve competent performance in an AI-augmented workflow versus a traditional one. An ECR of 0.5 means half the experience is needed to reach the same output standard.

The study tracked 4.2 million job postings from 2023 to 2027, mapped against the O*NET occupational taxonomy and cross-referenced with LinkedIn salary and mobility data. It also conducted longitudinal studies with 12,000 workers in finance, law, software engineering, journalism, and clinical radiology — five fields where AI tools have seen the highest enterprise adoption.

Key Findings

Finance: ECR of 0.41

Junior financial analysts now perform at the level that previously required five years of experience. AI tools handle the mechanistic work — DCF modeling, sensitivity analysis, regulatory filing preparation — that once served as the crucible of professional development. The consequence: junior salaries at top investment banks have risen 22% since 2024, as the entry-level contribution floor lifted.

Law: ECR of 0.38

Paralegal and associate work has been most dramatically affected. Contract review, case law summarization, and discovery document processing — once the 3-5 year apprenticeship phase of litigation practice — are now handled by AI with accuracy that supervising partners rate as equivalent or superior to first-year associates. Law school applications are down 31%, but lateral hiring at the 3-5 year level has surged, as firms seek mid-career lawyers who can direct AI tools effectively.

Software Engineering: ECR of 0.52

The most surprising finding. AI coding assistants have not eliminated the need for senior engineers — they've made them dramatically more productive. Junior engineers now ship production code in their first month that previously required six months of ramp-up. But the ceiling has also risen: AI-augmented systems require stronger systems design thinking to orchestrate effectively, paradoxically increasing the value of deep expertise.

Journalism: ECR of 0.47

AI transcription, summarization, and data visualization tools have compressed the path from rookie to competent beat reporter from 4 years to under 18 months. However, the study notes that journalism has bifurcated: the volume commodity end (sports scores, earnings reports, local government minutes) is now largely AI-generated, while investigative and narrative journalism remains human-intensive.

Clinical Radiology: ECR of 0.59

The most modest compression, reflecting medicine's cautious AI adoption. AI reads chest X-rays with a 94.7% sensitivity for critical findings — superior to the 89.2% baseline for residents under five years post-training. But radiologists are not being deskilled; they're being re-skilled as AI orchestrators who supervise multiple AI reads simultaneously, with their expertise shifting to edge cases and AI error detection.

The Compensation Paradox

The study surfaces a counterintuitive pattern: AI deskilling has driven entry-level salaries up, not down. In fields where AI elevates the minimum viable contribution, competition for talented entry-level workers has intensified, as the economic value they can capture in their first year has risen. Mid-career salaries in some fields have stagnated as the tenure premium compresses.

Policy Implications

The WEF is calling for a recalibration of professional credentialing timelines. If the effective expertise window has compressed, professional licensing exams, apprenticeship requirements, and credentialing board structures designed for a 20th-century labor market need review.

The OII is more pointed: educational institutions that fail to restructure curricula around AI-augmented practice will produce graduates who are simultaneously overeducated in obsolete skills and underprepared for AI-native workflows.