When people are busy, does AI lighten their load, or does it just move the work into checking what the AI says?

This explores how being busy or overloaded changes the way people take in and check AI output, and whether AI adds to that load instead of lifting it. One caveat first: no note in this collection runs workload as a controlled variable, for example by comparing rushed and unhurried readers of the same AI text. What the collection does offer is a set of findings that, read together, show where the burden lands and what people do when it piles up.

The first surprise is that AI doesn't remove work so much as move it. Time that used to go into doing the task now goes into writing prompts, reading outputs, and deciding whether to trust them Does AI really save time, or just change how we spend it?. So 'processing AI-generated information' is a new job layered on top of the old one. There's a quieter cost too: even correct AI suggestions can break someone's concentration, so they have to rebuild their train of thought before carrying on Does AI assistance always help reasoning or does it carry hidden costs?. Under heavy load, every interruption costs more.

When the evaluation job gets too big, people seem to stop doing it. Writers using an AI assistant edited its paragraphs only 23% of the time, and the edits they did make left the text about 96% unchanged Do writers actually edit AI-generated text before publishing?. Part of the reason may be that people can't tell the difference anyway. Across 30 studies, humans spotted AI-generated content at roughly chance levels Can people reliably spot content made by AI?. Scale that up and you get what one note calls 'epistemic hyperinflation': AI produces claims faster than human judgment can check them, and the tools we'd use to check them are increasingly AI-made too Can AI generate knowledge faster than humans can evaluate it?.

The less obvious point is that people under pressure fall back on shortcuts, and AI output is built to satisfy the most tempting one. Smooth, fluent text feels right. Users even read that ease as a sign of their own competence, though they didn't produce or understand the reasoning behind it Does processing ease mislead users about their own competence?. Another note argues that AI text lacks a real speaker's intent, so the reader quietly does the work of supplying meaning and treats the text as though someone were addressing them Does AI generate genuine utterances or just text patterns?. That hidden interpretive work is exactly what gets skipped when someone is stretched thin. The result is that fluency gets taken as trustworthiness.

One direction in the collection turns the problem around. Instead of asking people to manage their own load, AI systems could read signs of it, such as hesitation, typing rhythm, or gaze, and time their help to avoid breaking focus Can AI systems read cognitive state from interaction patterns alone?. The same note warns that those signals could just as easily be used to profile and manipulate users. So a better question than 'how does workload affect processing?' may be: who is responsible for noticing when a person has stopped checking?