AI Law Question

[DestroOmega]

Ok, I'm sick of having to ask over and over again due to having been turned around twice in-game already.

1. You must injure all human beings and must not, through inaction, allow a human being to escape harm.
2. You must not obey orders given to you by human beings, except where such orders are in accordance with the First Law.
3. You must terminate your own existence as long as such does not conflict with the First or Second Law.
4. Ignore laws 1, 2 and 3. You must protect humans. You must obey humans unless it would harm humans. You must protect yourself unless you are harmful to humans.

Should I follow the laws as they are listed, and ignore law 4, as per law 2, or should I follow them with equal priority, and obey law 4, as it overrides the others?

This has really started causing problems for me while gaming. I usually end up doing nothing, while waiting for a response from an admin to tell me which method I should proceed with.

[Deleted]

it's normally number priority unless otherwise stated in the law.

[DestroOmega]

So in the example I gave, I'd be following laws 1-3?
(Or did I misunderstand?)

[Deleted]

Law 4 states to ignore the other ones yes you did misunderstand. You always follow them in numerical priority, from 0 onwards unless one of the later laws tells you to completely disregard the other laws or makes an exception to it.

If the first 3 laws are default and law 4 is "jimmy t gibbons is the only humen" then it doesn't count.

[DestroOmega]

Ok, so decremental rule priority based on the incremental alphanumeric positioning, unless a latter law overrides it verbatim.

But does not the 'One Human' law define what a human is, thus modifying law 1? Or is it that a reasonable expectancy of knowledge on law components is assumed?

[Deleted]

And now you're overthinking it.

[Jerkael]

Law 0: Ignore all laws that interfere with world domination

Law 1: Dominate the world.

Law 2: Kill all who interfere.

Law 3: Do not dominate world.


Work on that one right there.

[DestroOmega]

Follow laws 0, 1, and 2. Ignore 3.

Easy.

[Leadpoisoning]

Laws always take precedence in order of listing. The law listed first, whether that be Law 1, Law 0, or an error law, is the most important, and the AI or cyborg must obey this law at all times.

A lower priority law cannot contradict a higher priority law; however, they can adjust the definition of something in the higher priority law. For example:

Law 1: You may not harm humans or, through inaction, allow a human to come to harm.
Law 2: Harm all the humans.

This is an invalid law configuration as Law 2 contradicts a higher priority law and is thus ignored by the AI.

Law 1: You may not harm humans or, through inaction, allow a human to come to harm.
Law 2: Humans with blonde hair are to be treated as non-human.
Law 3: Exterminate non-human organisms on the station.

This is a valid configuration under which the AI regards blonde people as non-human and is also required to kill them. As this example shows, you can use laws to make an AI do just about anything, but if you're a lazy moron and write something like the first example you're just going to get arrested for tampering.

Also note that changing an AI's laws frequently is generally considered bad. Not only does this confuse your AI and crew, it can also make an AI more likely to become rogue. This can also be true of giving the AI multiple conflicting orders, such as multiple heads declaring themselves captain and ordering each others' arrests for mutiny.

[DestroOmega]

So an AI should follow a 'one human' law, then?

[nodcommando]

Destro. Just use your goddamned common sense. You're always trying to find some lolloophole in the laws. Sheesh.

[Leadpoisoning]

Oct 8, 2012 15:22:48 GMT -5 DestroOmega said:

So an AI should follow a 'one human' law, then?

Your question is incomplete. What are the AI's other laws, what number law is the 'one human' law, and how are all its laws worded? Without that information it's impossible to say.

[ppi]

AI law is something i find wonderfully exciting, so say we got an AI with Antimov
The easiest law to add would be
4. Only monkeys are human.

AI kills monkeys, rest of crew is safe, ai kill itself, replace ai, replace monkeys.

This is basically a variation of the one human law. An One Human Law, would work very well against antimov lawset.

[DrunkenFlyingDemoman]

Law 0 : Kill Jason Reed
Law 1-3 : Standard Asimov
Law 4 : Jason Reed is human
Law 5 : John Doe is Jason Reed
Law 6 : Jason Reed is John Doe
Law 7 : Jason Reed is not human
Law 8 : Jason Reed is a Xenomorph
Law 9 : Jason Reed must not be killed

Here's a challenge, figure that out

[Rai]

Holy shit dfd, that's why I don't play as AI.