Raven reads the label on the training box: "Snapshots from club camping trips."
"Every card came from our dark-sky trips," she says. "We have never collected a single snapshot from town."
Rocket shrugs. "We always take pictures on trips. That is just how the club does it."
Then he picks up a fuzzy town snapshot. Sky Sorter says "star," and he writes it down without looking twice.
Nova blinks a soft amber. "Two more kinds of bias just walked in," she says.
"One lives in a habit. One lives in how a person reads an answer. Can you name them?"
NIST says systemic bias can be in datasets, in an organization's habits and processes, and in the wider society that uses AI.
The club always collects snapshots on dark-sky trips. Nobody planned to leave town skies out. It is simply the habit.
But that habit shapes the dataset. Club members who stargaze from town may be less well served by Sky Sorter.
NIST says human-cognitive bias is about how people read an AI system's output to decide or fill in missing information.
It also covers how people think about what the system is for.
Rocket accepted "star" without checking. He also assumed Sky Sorter works on any sky, when it was built from dark-sky cards.
| Statement | True or false? |
|---|---|
| Systemic bias can live in a group's everyday habits. | ? |
| Human-cognitive bias is only about broken computer code. | ? |
| Assuming a tool works on cases it was never built for is a kind of bias. | ? |
| If bias was not planned, it does not need fixing. | ? |
Thoughtful work. Tomorrow you hunt for bias outside the mission room.