Automation bias

Automation bias is when you trust a tool's answer more than the evidence behind it, usually because it has been right often enough that you stopped checking.

Your watch has been reading your sleep and heart rate for a year, and its morning recovery score has become the thing that decides your training. This Tuesday it says 94, green, go hard. Your right knee is hot, you slept badly next to a sick kid, and your legs feel like sandbags on the stairs. You do the hill repeats anyway. Three weeks later a physical therapist is asking when the pain started, and you cannot say, because you stopped tracking anything except the number.

Automation bias is accepting an output more readily than its evidence warrants. The tell is that you can no longer name what the tool used to reach its answer, and you catch yourself arguing against your own direct observation to keep the number right. Unnoticed, it costs you the very signal the tool was meant to amplify, and a wrong reading travels as far as a right one because nobody reads either. The smallest correction is a spot check, not a policy: once a week, write your own guess before you look, then compare. Where the two disagree, ask what the tool cannot see, here a swollen joint and a bad night it scored as rest. Give it authority only up to what you have verified it measures well. It is a false alarm when stakes are low and errors surface immediately; when a bad call costs you a season, check.