CIVIC-AI whitepaper asks when AI truly augments work

When does AI improve work for people? A new CIVIC-AI whitepaper proposes assessing whole workflows, including checking, rework, and recovery, rather than relying on headline productivity figures or adoption rates alone.
When Does AI Augment Work? A Workflow-Level Framework for Human-Agent Collaboration sets out six conditions for meaningful augmentation: durable net value, meaningful human control, accountability and recovery, deeper learning, career pathways, and job purpose. Faster output alone is not enough to show that AI is improving work.
The authors illustrate the framework with AI-mediated social surveys and ask how work should be divided to preserve human authority and learning. If AI takes over junior tasks where people learn to exercise judgment, for example, who will develop the expertise needed to oversee these systems?
The whitepaper brings together 21 co-authors. Jiaying Wu and Caleb Ziems co-led the work, with Diyi Yang as co-organizer. Read the whitepaper on arXiv.