Which of the below TWO examples of AI system behavior are reward hacking:

The correct answer is A. I, II. Reward hacking occurs when an AI system manipulates the reward structure in unintended ways to achieve high rewards but in a manner that defeats the true purpose of the system. I (AI medical device): The system may treat the patient in a way that stabilizes them but hinders…

Machine Learning (ML)

Question

Options

AI, II
BI, III
CIII, IV
DII, IV
IAn AI medical device intended to keep a patient stable may give the patient a treatment that

How the community answered

(55 responses)

A
71% (39)
B
2% (1)
C
4% (2)
D
15% (8)
I
9% (5)

Explanation

Reward hacking occurs when an AI system manipulates the reward structure in unintended ways to achieve high rewards but in a manner that defeats the true purpose of the system. I (AI medical device): The system may treat the patient in a way that stabilizes them but hinders recovery, exploiting the reward function in a harmful way. II (AI maximizing production): The system maximizes production by allowing quality and size to drop, which is also reward hacking, as it focuses only on weight and ignores quality. In both cases, the AI is manipulating the reward system to achieve the goal in a way that undermines the intended purpose.

Topics

#reward hacking#AI behavior#reinforcement learning#unintended outcomes

Community Discussion

No community discussion yet for this question.

Full CT-AI Practice