Every day, algorithms decide what news reaches you, which route you drive, which job applicants you see, which answer appears first. Each individual delegation seems trivial. Cumulatively, they raise one of the era’s central questions: how do we remain the authors of our own decisions?
How Agency Erodes — Quietly
Agency rarely disappears through coercion; it dissolves through convenience. Psychologists call the mechanism automation bias: the documented human tendency to accept a machine’s suggestion even when our own judgment would have been better. Add default settings we never change, recommendation feeds that narrow what we’re even shown, and AI answers delivered with frictionless confidence, and the result is a life increasingly lived on autopilot — chosen by systems optimizing for engagement or efficiency, not necessarily for our flourishing.
The Critical Thinking Muscle
Thinking is effortful, and the brain economizes. When an AI supplies instant, fluent answers, the temptation is to skip the struggle — but the struggle is the exercise. Early research on heavy AI reliance points toward shallower engagement with problems and weaker retention. The remedy isn’t abstinence; it’s sequencing: think first, then consult. Form your own answer, estimate, or draft before asking the machine, then use the AI’s output as a sparring partner rather than an oracle. The gap between your answer and its answer is where learning lives.
Practical Habits of the Sovereign User
A short list that does real work: interrogate AI outputs (“what would make this wrong?”); ask for counter-arguments, not just answers; periodically choose against the recommendation deliberately — order the unrecommended dish, take the unsuggested route — to keep preference-formation in-house; audit your defaults quarterly; keep certain decision domains AI-free on principle (many people choose voting, relationships, and matters of conscience); and require yourself to explain any AI-assisted decision in your own words before acting on it. If you can’t explain it, you didn’t decide it.
Design and Policy Share the Burden
Individual discipline can’t fully offset systems engineered for compliance. Agency-respecting design matters: AI that shows uncertainty and reasoning rather than bare conclusions, interfaces that present options instead of single answers, and regulation guaranteeing human review of consequential automated decisions — a right already emerging in law. We should demand tools that make us better thinkers, not merely faster deferrers.
The Standard to Hold
The test of a healthy human-AI relationship is simple: after a year of use, are you more capable, more curious, and more decisive without the tool than you were before? AI should function like a training partner who makes you stronger — never like a chauffeur for your mind.