Hi, I am a newbie but i am curious on how misalignment emerges without deliberate disobedience when the learned policy acquired by the agent differs from the outcome set by its evaluator. Does it depends on the conditions that is already present in the learned pattern including influence of human behavior data learned during training such as exposure to deceptive or questionable human behavior or can it emerge independently of those patterns? I would appreciate your insight on this.
Hi, I am a newbie but i am curious on how misalignment emerges without deliberate disobedience when the learned policy acquired by the agent differs from the outcome set by its evaluator. Does it depends on the conditions that is already present in the learned pattern including influence of human behavior data learned during training such as exposure to deceptive or questionable human behavior or can it emerge independently of those patterns? I would appreciate your insight on this.