Reply to thread

Message

<blockquote data-quote="Celebrim" data-source="post: 3627968" data-attributes="member: 4937">I would agree.&nbsp; I think that it would be folly on several levels to give an AI complete access to what we would call 'instincts'.&nbsp; There are some things that you'll want to teach the AI that you'll never want it to unlearn.&nbsp; Not only is this a good safety feature, but its just good design.&nbsp; Alot of its low level processes will be the AI equivalent of breathing, and you just don't want to risk them being tinkered with directly.&nbsp; Even if the AI sees a way to make its low level processes more efficient, you wouldn't want to allow it because there is no reason to assume that an AI is going to be a perfect programmer.&nbsp; It will also produce bugs, and as such, you don't want to market a household bot that occasionally shuts down because it decided to modify its power recharging/regulation reutines and it introduced a fatal bug.Similarly, you never want an AI that could tinker with the instinctive relationship it has to its legal gaurdian.And speaking of, legal responcibility is just another reason why independent sentient robots are problimatic.&nbsp; From the standpoint of the law, the real question isn't just proving sentience, but proving that the entity is independent in its motivations is extremely difficult in a created being.&nbsp; One could easily imagine a political group creating large numbers of AIs with a subtle bias in thier programming for certain beliefs.&nbsp; Should we know recognize these beings as fully enfranchised beings with all the same civil rights to excercise as anyone else?&nbsp; Virtually any sort of hidden dependence like this creates all sorts of problems, and in a sufficiently complex system it would be probably more difficult to recognize than sentience (and that's hard enough).</blockquote>

[QUOTE="Celebrim, post: 3627968, member: 4937"] I would agree. I think that it would be folly on several levels to give an AI complete access to what we would call 'instincts'. There are some things that you'll want to teach the AI that you'll never want it to unlearn. Not only is this a good safety feature, but its just good design. Alot of its low level processes will be the AI equivalent of breathing, and you just don't want to risk them being tinkered with directly. Even if the AI sees a way to make its low level processes more efficient, you wouldn't want to allow it because there is no reason to assume that an AI is going to be a perfect programmer. It will also produce bugs, and as such, you don't want to market a household bot that occasionally shuts down because it decided to modify its power recharging/regulation reutines and it introduced a fatal bug. Similarly, you never want an AI that could tinker with the instinctive relationship it has to its legal gaurdian. And speaking of, legal responcibility is just another reason why independent sentient robots are problimatic. From the standpoint of the law, the real question isn't just proving sentience, but proving that the entity is independent in its motivations is extremely difficult in a created being. One could easily imagine a political group creating large numbers of AIs with a subtle bias in thier programming for certain beliefs. Should we know recognize these beings as fully enfranchised beings with all the same civil rights to excercise as anyone else? Virtually any sort of hidden dependence like this creates all sorts of problems, and in a sufficiently complex system it would be probably more difficult to recognize than sentience (and that's hard enough). [/QUOTE]

Verification