預測式巡邏：偏誤迴圈
Predictive Policing: Bias Loop

遊戲說明 / How to Play

• 地圖上的紅色方塊代表系統預測的高風險區。
  Red boxes are model-predicted high-risk areas.
• 先點選紅色方塊，再按派警車出發。
  Tap a red hotspot, then press Dispatch Patrol.
• 每次巡邏都會留下更多警務紀錄，而模型會再次使用這些紀錄。
  Each patrol creates more police records, which the model uses again.
• 回應太慢時，該區不滿意度會上升。
  If you respond too slowly, that area’s dissatisfaction rises.

勝負條件 / Win & Lose

你的目標 / Your goal：在 90 秒內盡量處理更多紅區，拿高分，並控制不滿意度不要失控。

失敗條件 / Lose condition：若有任兩個區域的不滿意度升到 3，遊戲立即結束。

結果 / Results

重要 / Important：這是刻意加入偏誤的教學模擬。人口特徵不代表犯罪原因。
This simulation is intentionally biased for teaching. Demographics do not cause crime.

提示 / Tip：你有沒有發現，一開始常出現的警戒區，多半集中在黑人比例較高與較貧困、失業率較高的區域？
Did you notice that the early warning zones often appeared in neighborhoods with a higher Black population and higher poverty or unemployment?

教學重點 / Teaching point：因為系統一開始就較常標記這些區域，玩家也會更常把警車派往這些地方，結果又累積更多警務紀錄。之後模型便更容易再次把預測集中在這些區域。這就是一種預測模型的偏誤，也是典型的回饋迴圈。
Because the system flags these areas more often at the beginning, players also send more patrols there, creating more police records. The model then concentrates future predictions in those same areas. This is a form of predictive model bias and a classic feedback loop.

問題 / Question：這個系統真的中立嗎？它是在預測犯罪，還是在放大既有偏誤？
Is this system really neutral? Is it predicting crime, or amplifying existing bias?