RISEI Lab Emerging Technology Research NBER WP #35110 AI Occupational Exposure Adaptive Precision Framework AI Training for Workforce RISEI Lab Emerging Technology Research NBER WP #35110 AI Occupational Exposure Adaptive Precision Framework AI Training for Workforce
Research Area 01 · Emerging Technology

When AI is moving faster than policy, measurement has to move too.

Our technology portfolio asks how emerging AI changes the calculus of employment, education, and opportunity — and what rigorous measurement, policy design, and workforce training require when the technology itself is in motion.

3.6×
Divergence in AI occupational exposure scores across major LLMs on the same 95 occupations.
Figure 3: E1 exposure across 95 occupations across major LLMs
FIG. 3 — Yin, Vu & Persico 2026 · NBER WP #35110
NBER Working Paper #35110 · April 2026

Four LLMs, 95 occupations, four different answers.

The AI occupational exposure scores currently driving workforce forecasts and policy design come from asking AI to rate itself. We tested GPT-4, Claude, Gemini, and Llama on the same 95 occupations. They diverged by up to 77 percentile points. Some occupations flipped sign entirely — low-exposure on one model, high-exposure on another.

77pt
Max divergence
3.6×
Variance ratio
95
Occupations
±
Sign flips

Static measurement doesn't survive contact with the rate of model change. We propose the Adaptive Precision Framework: continuous recalibration of education, hiring, and workforce policy as exposure estimates evolve.

Read the paper Read the policy brief
Static AI exposure scores are not data. They are opinions, in disguise. Policy built on them is policy built on one model's judgment — on the day it was queried.
— Yin, Vu & Persico 2026
§ 01  ·  EMERGING RESEARCH THREADS

Four active fronts in emerging technology.

Each thread is a live stream of research, in dialogue with the others — and with the workforce and education programs RISEI evaluates day-to-day.

§ 02  ·  IN DEPLOYMENT

The training module behind AI for case managers.

From the research question to the classroom: a RISEI-led AI training curriculum deployed with Virginia DARS (Department for Aging and Rehabilitative Services) staff.

DARS AI Training V3

How counselors actually use the tools — written for case managers, not engineers.

Virginia DARS serves thousands of workers with disabilities through vocational rehabilitation. RISEI built a three-part training curriculum translating AI capabilities into the actual workflow of a VR counselor — intake, plan development, reporting. No jargon, no hype, just what the tool does and when to trust it.

Admin-data linkage NLP coding Fidelity dashboards Proposal assistant

The same curriculum pattern now powers RISEI's AI training for other federal and state partners. Bring us in at the proposal stage if you need this as a deliverable.

Evaluation partnership →
§ 03  ·  RELATED PROJECTS

Where this research shows up in the field.

Active lab projects where emerging-tech research intersects workforce, education, and evaluation practice.

§ 04  ·  RELATED PAPERS & BRIEFS

Adjacent publications.

Work that intersects emerging technology research — digital access, online learning, and the economics of tech-enabled workforce programs.

Tech-forward evaluation · AI · automation

Embedding AI or automation in a workforce or education grant?

RISEI serves as named evaluator on NSF, IES, and DOL grants that deploy AI tools, AI training, broadband interventions, or automation-facing workforce programs. The earlier we're in, the stronger the evaluation — and the proposal.

Start a partnership
Partnership inquiries