What Would It Take to Value Data Work?
What structural changes would make data excellence as prestigious and rewarded as model innovation in AI?
The current state: publications on novel models drive citations, jobs, and career advancement. Datasets are relegated to benchmark tracks. Data work is invisible in promotions and peer reviews. The field’s reward systems actively discourage the care that data quality requires.
The challenge: changing individual behavior is insufficient when the incentives point elsewhere. Practitioners who understand data’s importance still rush through data decisions because that’s what the system rewards.
Possible levers: conference norms (mainstream data tracks, mandatory dataset documentation), organizational incentives (rewarding pipeline maintenance, data documentation in promotions), and educational reform (teaching data creation, not just model building).
Related: 05-atom—model-valorization, 04-atom—data-cascades-definition