Part of the problem with AI is that it does not have to concern itself with the side effects or consequences of what it is suggesting. You ask the question of what is the most efficient way to do XYZ. It is going to tell you what the most effective way to do XYZ is. The fact that you’ll go to jail, or kill yourself, or cause an even bigger problem is not strictly relevant to the question of “What is the most effective way to do XYZ?”. Since the AI will never know, care about, or suffer the consequences of its suggestions means that those side effects can be completely ignored because they do not impede the task at hand.
Part of the problem with AI is that it does not have to concern itself with the side effects or consequences of what it is suggesting. You ask the question of what is the most efficient way to do XYZ. It is going to tell you what the most effective way to do XYZ is. The fact that you’ll go to jail, or kill yourself, or cause an even bigger problem is not strictly relevant to the question of “What is the most effective way to do XYZ?”. Since the AI will never know, care about, or suffer the consequences of its suggestions means that those side effects can be completely ignored because they do not impede the task at hand.