Expertise capture in workplace AI
Organizations: Faculty of Computer Science, University of Vienna, Kolingasse 16–19, Vienna, 1090, Austria. · Edith Cowan University, 270 Joondalup Drive, Perth, WA, Australia.
Abstract
AI assistants are becoming part of how knowledge workers think and perform, accumulating the prompts, exemplars, memory and interaction histories that encode a worker's professional judgment. This infrastructure typically runs under employer or platform control. We name the resulting process expertise capture, the absorption of worker-applied knowledge and judgment into AI resources under organizational control. We develop a framework of contribution forms, technical residences and transformation families, and show that how far a contribution can be attributed, separated and revoked depends on where it resides and how it has been transformed, properties fixed by system design long before any dispute arises. When a worker departs, the organization retains a capability shaped by their judgment while the worker loses access to it, and the asymmetry compounds as contributions move from inspectable artifacts into derivatives, learned components and successor systems. We outline three measures, write-time provenance, extraction of separable contributions and an executable exit procedure, that make an agreed allocation technically performable, and make explicit what cannot be separated at all.
Figures & tables
| Replicative | Derivative | Parametric | Propagative | |
|---|---|---|---|---|
| Source relation | Most coupled: source preserved and distributed | Source interpreted, compressed or abstracted | Source absorbed into learned components | Least coupled: capability reproduced in successor resources |
| Operations | Ingestion, copying, chunking, indexing, caching | Embeddings, summaries, memory extraction, profile inference, generated workflows | Continued pretraining, fine-tuning, preference optimization, reward-model and adapter training | Distillation, model merging, synthetic-data training, successor deployment |
| Provenance | Direct if identifiers and copy records are preserved | Lineage-dependent across one or more sources | External training and corpus records required | End-to-end lineage required across systems |
| Attribution | Usually high for known instances | Conditional and often many-to-one | Approximate and contested at the contributor level | Very low without successor lineage |
| Separability | High for known instances; copies and caches complicate removal | Artifact separability may remain; semantic separability decreases where sources are fused | Low at contributor level; modular components such as adapters can be deleted only as units | Low where successors no longer contain the original source |
| Revocability | Feasible for known instances; incomplete without copy records | Derivative-dependent and difficult once fused into profiles or summaries | Limited without retraining or validated unlearning | Downstream-dependent; source removal may not affect successors |
AI and Labor
AI and labor studies how artificial intelligence changes jobs, wages, skills, hiring, and the organization and quality of work.
See all 103 papers