Tag: ai-alignment

Reward hacking occurs when an AI optimizes the score instead of the intended outcome. Gamified systems are especially exposed because they already depend on visible proxies, points, and goals.

Helpful AI becomes counterproductive when support turns into agreement, flattery, or validation of weak ideas; growth requires warmth that can still tell the user something is wrong.