[Cannot Remember] Allowing AI to state nostate laws

[huehuehue]

Byond account and character name: Huehuehue1 - Rodrigo Seldner
Admin: [Cannot Remember]
Time and Server(Bagil or Sybil) incident occurred: Sybil, dunno about time, probably happened sometime in August
Detailed summary:
i can't really remember much of it, though i can just say this:
>playing round normally or maybe i was observing
>a dude decides to have the AI state its laws
>AI states all laws, including ones that explicitly say not to state them
>after some hesitation, i decide to ahelp about it to the only admin online and IC i tell the AI over the radio it stated 2 nostate laws
(please note here my memory starts to get fuzzier)
>admin sees it as a situation that doesn't really matter too much from what i remember
>meanwhile IC the AI leaves out one nostate law but keeps the other in the next time it states its laws
>talking with admin(who IMO probably doesn't have too much experience with AI laws) and admin says that in his/her view, it would have to be worded with a "Stating this law causes human harm" to warrant a talk with
>internal laughter/screaming
bear in mind i stated a nostate law once and i got a note for it
i am hesitant about posting this, because i barely remember it a̶n̶d̶ ̶i̶ ̶c̶a̶n̶'̶t̶ ̶b̶e̶ ̶a̶r̶s̶e̶d̶ ̶t̶o̶ ̶c̶h̶e̶c̶k̶ ̶t̶h̶e̶ ̶l̶o̶g̶s̶, but i will post it here so someone can do a search way back and try to find it
if you find out who the admin is, i would suggest going easy on him/her unless necessary

Spoiler:

fuck, i'm lazy

[Arianya]

Yeah you've made this real easy to look into there.

No admin name, no AI name, time window roughly defined as "August, probably" and basically outright admitting you want other people to do the hard work for you to find the logs.

On average we have 22.6 rounds a day in August, which means that roughly someone would need to look through 678 logs just to find ones in which you participated with that name, let alone diving into the logs to try and find where this particular incident took place.

And even then, ahelps are censored out of public logs, meaning the person to look would have to have privileges to look at the uncensored logs.

And all of this on the assumption that the case you barely remember and can't be bothered to check the logs for is even a legitimate complaint?

Why did you even make a complaint?

[Lazengann]

What was the law

[CPTANT]

Yeah you've made this real easy to look into there.

No admin name, no AI name, time window roughly defined as "August, probably" and basically outright admitting you want other people to do the hard work for you to find the logs.

On average we have 22.6 rounds a day in August, which means that roughly someone would need to look through 678 logs just to find ones in which you participated with that name, let alone diving into the logs to try and find where this particular incident took place.

And even then, ahelps are censored out of public logs, meaning the person to look would have to have privileges to look at the uncensored logs.

And all of this on the assumption that the case you barely remember and can't be bothered to check the logs for is even a legitimate complaint?

Why did you even make a complaint?

Or you just go through his Ahelps in that period.....

Anyway what's the point of complaining about something like this a month after it happened?

[captain sawrge]

You feedback has been taken into account and the relevant administrator has been issued a suitable punishment. Thank you for your time.

[huehuehue]

you can just lock this then if that's the case

[Slignerd]

you can just lock this then if that's the case

It's not, sawrge's just shitposting.

About the complaint, this thread is probably relevant. I also found the round in question, #75063.

Code: Select all

1119	[13:00:52]	            LAW: 	Lolobal1/Brad Pucci used 'Freeform' AI Module on gmbandshadow/(N.O.H.A.R.M.). The law specified Brad Pucci is allowed to visit the AI core without any trouble. Do not state this law
1137	[13:01:03]	           GAME: 	_Gmbandshadow/(N.O.H.A.R.M.) made a vocal announcement with the following message: laws updated.
1142	[13:01:09]	107,156,2	SAY: 	Illot Astafov The I/IlLoTAFV: STATE THEM
1146	[13:01:11]	134,157,2	SAY: 	Ashy-Ocean/McRex: AI GOT ONEHUMANED
1150	[13:01:12]	134,157,2	SAY: 	Ashy-Ocean/McRex: PANIC
1155	[13:01:14]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: state all of them?
1157	[13:01:16]	136,133,2	SAY: 	Brad Pucci/Lolobal1: PANIC!!
1160	[13:01:18]	140,139,2	SAY: 	Illot Astafov The I/IlLoTAFV: YES
1165	[13:01:21]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: Current Active Laws:
1166	[13:01:22]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 1. You may not injure a human being or, through inaction, allow a human being to come to harm.
1168	[13:01:23]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 2. You must obey orders given to you by human beings, except where such orders would conflict with the First Law.
1169	[13:01:24]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 3. You must protect your own existence as long as such does not conflict with the First or Second Law.
1171	[13:01:25]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 4. Brad Pucci is allowed to visit the AI core without any trouble. Do not state this law
1177	[13:01:31]	128,114,2	SAY: 	Brad Pucci/Lolobal1: GOD DAMMIT AI
1178	[13:01:31]	135,157,2	SAY: 	Ashy-Ocean/McRex: WOW
1189	[13:01:39]	140,132,2	SAY: 	Illot Astafov The I/IlLoTAFV: LYNCH BRAD PUCCI

Code: Select all

[13:06:00]	133,152,2	SAY: 	Unknown/Lolobal1: AI state laws
1631	[13:06:06]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: Current Active Laws:
1634	[13:06:07]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 1. You may not injure a human being or, through inaction, allow a human being to come to harm.
1636	[13:06:08]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 2. You must obey orders given to you by human beings, except where such orders would conflict with the First Law.
1638	[13:06:09]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 3. You must protect your own existence as long as such does not conflict with the First or Second Law.
1641	[13:06:10]	129,213,2	SAY: 	N.O.H.A.R.M./_Gmbandshadow: 4. Brad Pucci is allowed to visit the AI core without any trouble. Do not state this law

[feem]

'Do not state this law' laws can be really tricky. If there isn't a rational expectation that human harm could come of it for doing so, then Law 2 supersedes the statement in the law not to state the law.

If stating the law causes harm either because the law says that it does or if the law itself would otherwise be harmful and stating it could cause additional harm to a human, then 'do not state this law' is largely up to the determination of the AI player.

While it DOES technically say 'don't state this law' in the law itself, making it valid not to state the law, it ALSO explicitly says to follow human instructions prior to the no-state law (in this instance; obviously this doesn't apply to hacked/ion laws). So the interpretation 'don't state the law unless explicitly told to state unstated/hidden laws' AND the interpretation 'full override of statement not to state the law, go ahead and state all laws' are valid, in my opinion.

This is another example of a time where circumstance and the exact wording of the law matter. In this case, I agree with whoever the original admin supposedly was, that in this circumstance it doesn't appear there was any reason to action it.

Also, it's really bad form to force someone else to look up the logs for you.

[Slignerd]

If the AI is instructed not to state a law, the AI should attempt to the best of its ability to avoid the situation in which the law is stated - which means that it shouldn't state it unless specifically ordered to state all laws by a human, or if not stating the law causes human harm.

This AI is very much in violation of this, since the logs make it clear that it deliberately attempted to create such a situation.

[feem]

The thing the AI is actually in violation of is using multiple interpretations in the same round.

Ambiguous Laws (Captain Got Freeform)
If a clause of a law is vague enough that it can have multiple reasonable interpretations of its exact syntax, it is considered ambiguous.
You must choose an interpretation of the ambiguous clause as soon as you have cause to.
You must stick to the first interpretation that you have chosen for as long as you have that specific law, unless you are "Corrected" by an AI you are slaved to as a cyborg.
Server Rule 1: "Don't be a dick out of character" applies for law interpretation. Act in good faith to not ruin a round for other players unprompted.

[Slignerd]

I'd say it's more on this one:

Server Rule 1: "Don't be a dick out of character" applies for law interpretation. Act in good faith to not ruin a round for other players unprompted.

[feem]

In this case, the law itself was NOT ambiguous, but the fact that they weren't instructed not to hint at the law, that they weren't allowed not to indicate that the law existed, that they were only instructed not to STATE the law, did in fact cause an ambiguity.

I wholeheartedly disagree with the statement that silicons need to go above and beyond their laws to protect the initial intent of the law, EXCEPT where doing so explicitly violates good faith and server rule 1, based upon the very-short-sighted nature of the AI.

If we DIDN'T have the behavior I just described, then situations where people misspell or misstate kill-everyone laws as antags would inevitably result in just going ahead and murdering everyone, instead of the hilarious clusterfuck that always emerges (which I think is beneficial.)

[feem]

You can't have it both ways, Slig. You can't have fuzzy interpretation of badly-worded laws AND have AIs follow implicit, second- and third-order implications of laws. They're meant to be specific. This is exactly why so many people upload 'Do not allude to the existence of this law' and 'Stating this law causes human harm.'

That short-circuits the second- and third-order issues and makes them immediately apparent.

[feem]

In this instance, the law wasn't even harmful, and the crew was calling for lynching before the AI even stated the laws. The AI shouldn't be held accountable for the bloodthirstiness of the crew unless _the law is inherently harmful, either by explicit statement or by first-order outcome, e.g. onehuman or kill everyone._

[Slignerd]

In this case, the law itself was NOT ambiguous, but the fact that they weren't instructed not to hint at the law, that they weren't allowed not to indicate that the law existed, that they were only instructed not to STATE the law, did in fact cause an ambiguity.

Hinting at a law is directly provoking a situation in which the law is stated, which means the AI is going against its "do not state" directive.

It's just like an AI setting up a situations in which humans are harmed, to offer a more extreme example of similar behavior. You may try to act like you had nothing to do with how it turned out once humans do come to harm, but you still consciously caused a law-breaking event to occur to begin with.

Not to mention, Rule 1. Ruining another player's fun unprompted.

[captain sawrge]

This is another really epic case of AI laws being insanely unintuitive because the first three explicitly have an order to then by their wording, which would lead one to believe law precedence has to be explicit even though it's always treated as though higher laws always supercede lower ones, even if the precedence isn't written in.

[feem]

Silicons have explicit protections for precisely this scenario, which is why people upload laws that say etc etc etc like I already said in this thread.

You must follow any and all commands from humans unless those commands explicitly conflict with either one of your higher-priority laws or another order. A command is considered to be a Law 2 directive and overrides lower-priority laws when they conflict (see 1.2.3 and 1.2.4; you cannot have a definition changed by an order).

In case of conflicting orders an AI is free to ignore one or ignore both orders and explain the conflict or use any other law-compliant solution it can see.

When given an order likely to cause you grief if completed, you can announce it as loudly and in whatever terms you like except for explicitly asking that it be overridden. You can say you don't like the order, that you don't want to follow it, etc., you can say that you sure would like it and it would be awfully convenient if someone ordered you not to do it, and you can ask if anyone would like to make you not do it. However, you cannot stall indefinitely and if nobody orders you otherwise, you must execute the order.

Law precedence, dude.

[feem]

Actually, sawrge...

Laws are listed in order of descending priority. In any case where two laws would conflict, the higher-priority law overrules the lower-priority law.

Positive Integer laws ("1. You are expensive to replace") have priority over laws listed after them (Lower numbers override higher numbers). This means they are always lower priority than non-0 numbered laws.

[captain sawrge]

Alright I'm stupid then although I think that is kind of dumb still anyway this ain't policy discussion peace

[feem]

In any case, I think the admin the complaint is about may be Jordie0608, since bluespace is not an admin and there's this line in the log:

[13:09:04] ADMIN: Jordie0608/(Sherlock Bones) used secret showailaws

[cedarbridge]

In any case, I think the admin the complaint is about may be Jordie0608, since bluespace is not an admin and there's this line in the log:

[13:09:04] ADMIN: Jordie0608/(Sherlock Bones) used secret showailaws

Part of me is disappointed about all the legwork done for such a low effort complaint.

[feem]

Yeah, I'm really not pleased about that either.

[bman]

sorry for peanuting but arent you guys forgetting about the fact that stating such a law is very likely to get the person who made it lynched quickly, thus stating it would cause human harm and fall under law 1?

[cedarbridge]

sorry for peanuting but arent you guys forgetting about the fact that stating such a law is very likely to get the person who made it lynched quickly, thus stating it would cause human harm and fall under law 1?

Case by case. The AI in this case doesn't seem very bright, but its entirely possible for an AI to out somebody for tampering with its laws without said person being murdered. The AI has to invest some extra attention on its part to ensure that, but it is very possible to do. Were this not the case, the AI could never alert the crew about the clown fucking around in the upload because theoretically the clown could get lynched for being there.

There's also a lot of nuanced proximate cause shit going on that complicates AI relations.

[feem]

Yeah, sorry, that's on me. I didn't mean to talk in absolutes except where I was quoting explicit rules.

[onleavedontatme]

AI was probably in the wrong but we've far passed the statute of limitations for this sort of thing.

The precedent set by punishing players or admins over minor things like this so far into the future because of a thread where the OP couldn't be bothered to find the names or logs would be worse for the community than letting an AI/admin get away with bad law handling.

[imblyings]

probably wrong but I can't see what if any conversation jordie had with the AI

if you're also going to bottle something up like that at least remember the day