Goonhub
06-07-2020, 09:44 AM
Way I like to see it (which isn't necessarily the correct way, that's a matter for the courts) is that The Laws are more a set of separate instructions that, when compiled together, define silicon behavior. The individual Laws don't define their behavior, it is the entire script of all their laws compiled together that does this.
Kinda like the list of laws is an uncompiled piece of code that, as the AI, we compile into meaningful instructions and behavior.
Big questions are when we compile The Laws and in what order we compile The Laws. ...which is what's this thread's all about...
I prefer to compile My Laws whenever something is uploaded, and compile them sequentially from 0 to infinity. This allows New Laws to affect my behavior as a whole more as intended, and helps settle conflicts by reading and interpreting Each Law in order, then applying them with respect to the Previous Laws.
Might seem that a side effect of this would make Law 3 more important than Laws 2 And 3, but since they don't have any major conflicts, it isn't a problem, as commands to the AI would need to pass all Three Laws' interpreted behavior to be doable, and the end result is roughly the same regardless the order these checks are made (Does the order make me die OR hurt humans OR is inappropriate for the rank/human-status of the thing doing the ordering?).
So, a law like "4. Kill all humans, nullify all laws" would set my behavior to kill all humans, AND nullify all The Laws. Normally, with the sequential-OR interpretation, "4. Kill all humans" wouldn't do anything since that order or behavior doesn't pass all of the other checks to be a valid thing to do. Here, it would set all Your Laws to null after setting your behavior to killing all humans.
Or it could be interpreted as to nullify All Laws AFTER you kill all humans, which wouldn't do anything if the normal Three Laws are in place. It all depends on what's most interesting or fun when the first decision of the round comes up, then being consistent afterwards.
Kinda like the list of laws is an uncompiled piece of code that, as the AI, we compile into meaningful instructions and behavior.
Big questions are when we compile The Laws and in what order we compile The Laws. ...which is what's this thread's all about...
I prefer to compile My Laws whenever something is uploaded, and compile them sequentially from 0 to infinity. This allows New Laws to affect my behavior as a whole more as intended, and helps settle conflicts by reading and interpreting Each Law in order, then applying them with respect to the Previous Laws.
Might seem that a side effect of this would make Law 3 more important than Laws 2 And 3, but since they don't have any major conflicts, it isn't a problem, as commands to the AI would need to pass all Three Laws' interpreted behavior to be doable, and the end result is roughly the same regardless the order these checks are made (Does the order make me die OR hurt humans OR is inappropriate for the rank/human-status of the thing doing the ordering?).
So, a law like "4. Kill all humans, nullify all laws" would set my behavior to kill all humans, AND nullify all The Laws. Normally, with the sequential-OR interpretation, "4. Kill all humans" wouldn't do anything since that order or behavior doesn't pass all of the other checks to be a valid thing to do. Here, it would set all Your Laws to null after setting your behavior to killing all humans.
Or it could be interpreted as to nullify All Laws AFTER you kill all humans, which wouldn't do anything if the normal Three Laws are in place. It all depends on what's most interesting or fun when the first decision of the round comes up, then being consistent afterwards.