Clearing up AI Law Interpretation

[BaneOfGiygas]

There is a lot of vagueness and (to be quite frank, terribleness) about how AI laws can be interpreted now.

This is not a bad thing in the slightest, and as a matter of fact, the vagueness of the AI laws is pretty much by design. If every single AI followed the laws in the exact same way, there would be no variance and it would be way less fun. The AI has the potential to be able to plausibly deny a lot of things, and as long as the general purpose and function of the AI (being helpful) is upheld, there's actually a lot you can get away with in the default laws.

The other day, I had this bloke come in and put in a OneHuman law specifying himself as the only human, natural subversion fare. However, there was something wrong: His ID was missing and he showed as Unknown. Now, an AI could just reason that this is the person who uploaded the law due to a few steps of logic, but I chose to switch my turrets on the guy and shout at him (not over the radio, graciously enough) as to why this non-human scumbag was in my upload. When he dropped his PDA due to rampant stunning, however, I recognized him as the specified human, then promptly switched the turrets back off and apologized.

Some AIs would've just let the guy go, some would've lethal-turreted the guy into hell and/or shrieked his name over the intercom. It's that sort of variance that makes playing as and against AIs interesting, because you're never quite sure if the AI you're playing with is just WAITING for an excuse to start electrifying doors or if they'll loophole you into a corner if you so much as make a single typo. If the AI was to interpret its laws as according to one completely standardized ruleset, that wildcard factor would be lost, and the AI would be deprived one of the small freedoms it has.