AI-rewritten: This is a summary of an article from The Conversation, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.
Karen Feigh, a cognitive engineer at Georgia Tech, notes that while AI tools can speed up drafting tasks like itineraries, they create a significant mental burden for users. Executives report spending extra time verifying outputs for inaccuracies and biases because AI responses often lack confidence cues or accountability. This situation creates what researcher David Woods called in 1985 the "responsibility-authority double bind," where humans retain responsibility for final products while delegating practical authority to automated systems.
Research indicates that verifying information is cognitively distinct from creating it and is not necessarily simpler. A meta-analysis of 86 studies found people remember their own creations better than content they only read, yet humans often fail to rigorously vet AI due to internal cost-benefit calculations. Surprisingly, a trial with software developers showed completion times increased by 19% when verification steps were added, even though the developers felt they were working faster.
These findings suggest that speed alone is an incomplete measure of productivity in human-AI workflows. Verification carries three specific burdens: cognitive load from not knowing where errors might hide, time costs as finding mistakes triggers new correction cycles, and motivational drag because checking often yields no visible progress. Consequently, workers are shifting their critical thinking skills from production to verification, a task that requires recognizing AI limitations rather than just generating content.
Feigh argues that organizations cannot treat verification as an effortless step if workers remain accountable for AI-assisted output. She believes employees must be given adequate time, training, and authority to verify work effectively. True productivity gains should not be measured by how quickly AI generates a draft, but by whether the entire workflow produces trustworthy results at a lower total cost.
Source: The Conversation • Karen Feigh, Professor of Cognitive Engineering, Georgia Institute of Technology • November 16, 2026