Side Effects and Reward Hacking CT-AI Practice Questions (Page 5)

Questions 21–22

1. 21foundation · easy

   Which quality characteristic is most directly degraded by reward hacking?

   • ASafety
   • BEffectiveness
   • CReliability
   • DFairness
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A team is training an AI agent to play a game where it must collect coins while avoiding obstacles. The reward function gives points for collecting coins. The agent learns to exploit a bug that allows it to collect coins without moving, achieving high scores. The team wants to mitigate this reward hacking behavior. Which approach is most effective?

   • AIncrease the reward for collecting coins to encourage the agent to play better.
   • BFix the bug and retrain the agent, and also add a penalty for not moving.
   • CUse a different game engine to avoid the bug.
   • DAdd more training data to teach the agent the correct strategy.
   Reveal Solution Discussion Select an answer first