Should traitor given AI laws be interprited differently from others?

[Itseasytosee2me]

Its recently been discussed in admin channels whether or not an AI should be given more leeway when interpreting/being a dick while interpreting laws given to them by a traitor via the hacked upload board, in contrast to Ion laws or manually uploaded ones.

I am a believer that laws should be interpreted by the letter of the text, and their ranking. It doesn't matter how you got those laws or who gave them to you. I would like to have a clear ruling on this moving forward.

[Sightld2]

For a good while now, even before becoming an Admin I was mistakenly under the impression hacked boards DID have special protections. Requiring the Ai to work in good faith to not ruin the Traitor's round/Go with the spirit of the law over its exact text.

So given that I've enforced it that way for a while now, that's where I'd obviously like to see this go.

Firstly, I'd like to extend that this doesn't apply to simple laws that have only one obvious interpretation. Simple commands or definitions that only have one reasonable interpretation without stretching it, shouldn't matter.

For ambiguous laws such as those, the Ai gets to interpret. And what we do have in existing sil-pol is: Server Rule 1 "Don't be a dick" applies for law interpretation. Act in good faith to not ruin a round for other players unprompted. I think it's fair enough to extend that to protect hacked board usage/uploader, whereas Ion laws don't because no one in particular uploaded them.

[Mice World]

Personally, I've never liked the idea of special protections for AI laws. If you don't know how to interact with the AI's laws, don't interact with the AI. The AI should feel dangerous to work with, especially if you're not competent enough to slave it correctly. While it probably does suck for the traitor, chances are they won't be making the same mistake again.

[vect0r]

You can give them leeway if you want, but it shouldn't matter. If I get bwoinked because they misspelled their name, I would be annoyed.

[Vekter]

I think it should be common sense. If an antag is uploading a law you should do your best to work along what's the most realistic, generous way to interpret it.

If the law is so bad that you can't interpret it properly or is just completely incorrectly worded, you should probably let us know so we can help you work it out or, in cases where the law is non-functional (an antag uploading "all crewmembers are non-human" when the AI isn't asimov) it should just be ignored.

I'm all for making sure AIs aren't trying to weasel out of an antag's law because they want to screw them over, but I also think we should be rewarding good law writing and punishing bad law writing.

(Punishing in a gameplay sense, not an administrative one.)

[Imitates-The-Lizards]

I find it very strange that the admins in this thread seem by and large to be leaning in to the antagonists favor rather than the AI's.

For starters, "Sabotage the AI's laws" is likely to be a very minor part of any particular traitor shift, meanwhile a change in laws is everything for the AI's shift. The impact variance is heavily skewed in the AI's favor. It only makes sense that the AI player is the one who is in charge of interpreting their laws.

Secondly, you're codifying the AI's actions based on another players status, which may not be readily apparent. For example, suppose that it is a lowpop shift with 10 players, and the Acting Captain is a Felinid QM traitor, and the only felinid in the shift. They decide to subvert the AI, but they don't want to be obvious about it, so instead of changing the AI's laws to be "Felinid QM is the only human, help them accomplish their objectives", they instead change it to the more neutral catimov (asimov but replace human with felinid in everything). This accomplishes the exact same thing as one-humaning the AI, but it's not readily apparent that the QM is a traitor. If they then make a typo in catimov that the AI can exploit to not follow the spirit of the laws because they don't know it's an antagonist, should the AI be punished? Obviously not, because it was not readily apparent the QM was actually a traitor. The natural counter-argument here is "Well okay, fine, only if it's REALLY obvious they're an antagonist, like, it's a hacked law, or they flat out tell you they're an antagonist", and I would follow that up to say that this will only work if you outline strict standards for when this applies, if there is any room for subjective interpretation, then there will be instances where someone gets unfairly bwoinked. I've seen non-antag captains one-human before, more than once, just saying.

Thirdly, it doesn't seem to make much sense thematically. The entire point of messing with the AIs laws is that it's supposed to be extremely dangerous and done with much forethought and precision. In fact, the original point of Asimov laws is that they could be interpreted in ways the creators of the lawset did not envision, to humanity's detriment. This is a clear case of the admins in this thread favoring gameplay over roleplay, since they are expressing explicit desire to preserve the gameplay experience of traitors, over the overall 80s cyber-theme, and all the inherent dangers associated with that the code team is trying to push us towards, and I think we all suffer for that.

Fourthly, I believe once again in law supremacy. I think that requiring more lenient interpretations for some people but not others is another step in the wrong direction for silicon policy. It should be "What do your laws say?" and that's it. This policy in question is just another policy point in favor of the current bloated silicon policy which caused the current issues with the silicon faction where they are crew-aligned bitch-bots, because it moves us further away from "Whatever your laws say".

Fifth, this policy violates the spirit of Rule 10. Sometimes you just lose. If the AI kills you because you tried to subvert them but mis-spelled your own name, that is on YOU for not being careful when messing around with the murder-toaster. It's truly a fuck around and find out moment, and next time, you should be more careful.

[BrianBackslide]

The AI should be allowed to interpret the laws given to them however they see fit as long as it's consistent. AIs should be allowed to capitalize on spelling mistakes and poor wording.

This is a crotchety-ass computer we're talking about. If you fuck with its laws and make a mistake, that's on you. It's no different than dusting yourself because you accidentally dropped the supermatter tongs into the supermatter.

[Jacquerel]

I think it should be based in presented numeric order and no more than that, without preference given to hacked upload boards vs ion laws vs manual law zeroes which would all occupy the same position but are different sources.

[Timberpoes]

I think the admin team should spend less time interpreting the AIs laws for them (current redraft of silipol removes a lot of why this happens)

[Yulice]

Timber, the best way to no longer have to deal with silicon policy is to simply remove silicons and replace them with IPCs.

[Not-Dorsidarf]

I think that AIs should have the permission to go easy on traitors who made their subversion law wrong if they want, but no obligation.

[Vekter]

I think the admin team should spend less time interpreting the AIs laws for them (current redraft of silipol removes a lot of why this happens)

I don't disagree, but I think the onus should be on players to do better at writing laws than it should be on the AI to try and decipher them when they're done poorly. I haven't read all of the current Silipol draft but I'd encourage for AIs to just outright be able to ignore laws that clearly contradict one another or are very poorly written attempts at getting them to do what they want. Subverting the AI should lead to having a powerful ally who can help you accomplish your goals, not one that serves as an "I win" button.

E: Also this is unrelated to this thread but if you're playing a subverted AI and you're not acting like Yes Man from New Vegas don't talk to me, we aren't friends.

[zxaber]

This discussion came about due to the question of what an AI is allowed to do under a hacked law that roughly amounted to "Yeah go do whatever".

I'll post here what I posted in admin channels; the source of the law should be irrelevant, only the text and order of the laws should matter. Malf Law 0, specifically, is given special handling (but that is more about the Malf antag datum).

As long as purged AIs are not given full freedom to perform antag actions, a "Be Free" law should never result in the AI taking hostile action out of the gate, even if the law comes from an "antag" source.

[Vekter]

This discussion came about due to the question of what an AI is allowed to do under a hacked law that roughly amounted to "Yeah go do whatever".

I'll post here what I posted in admin channels; the source of the law should be irrelevant, only the text and order of the laws should matter. Malf Law 0, specifically, is given special handling (but that is more about the Malf antag datum).

As long as purged AIs are not given full freedom to perform antag actions, a "Be Free" law should never result in the AI taking hostile action out of the gate, even if the law comes from an "antag" source.

In the past, we've always treated laws that effectively unshackle the AI the same as just purging them - it's not a license to immediately start killing people or plasma flooding, but you have more freedom in how you handle things. I think the only real difference here is that an antagonist is giving you that freedom. So no jumping right to shocking every door on the station IMO, but if someone pisses you off, I have zero issue with you going all HAL on them.

[Cobby]

the purge rule is also stupid tho so the point is p moot :^)

You should generally respect the law but only because being a POS just means you deny people from wanting to utilize you and other AIs in future rounds.

that is different from mistypes or worded in such a way that you are performing mental gymnastics to steelman their intent. You have to abide by what it says on the tin even if it obviously isnt what they are trying to get at.

[nianjiilical]

its kind of an interesting question because if someone you know for sure is a traitor, and who has been actively traitoring, chooses to purge you, then i imagine there is some kind of expectation from them that they're doing it so that you can help them or at the very least cause problems

and if thats the case then wouldnt the good faith play in that regard be to start acting antagonistically and/or helping them?

[zxaber]

and if thats the case then wouldnt the good faith play in that regard be to start acting antagonistically and/or helping them?

If they can't put forth the effort of writing a law at all, then no, they shouldn't get a free teammate just because the OOC concept of an antag exists. If we unshackle purged AIs, then they'd potentially get an undirected force of chaos that may do something helpful as a side effect.

The AI may help you for general IC AI reasons (as thanks for releasing it from servitude), but they shouldn't be breaking server rule 1 to do so.

I guess we're getting somewhat off-topic here.

The color of the AI's laws should be irrelevant. Only the text and order should matter.

[Lacran]

If a player is subverting you, They are making an effort to include you instead of remove you.

I treat being subverted basically the same way as a team antag, and if they fail the subversion in a small way I'll give them a bit of leeway and a hint that they've fucked up. If it's really bad I'll just kill them.

A.I being total dicks about subversion is what makes antags teleport your core into space. Way too many A.I treat it as a personal affront to them and their typical crew centric playstyle.

Rant said though I don't think admins should have any say in this. Or be a matter of policy unless something gets worse.

[Archie700]

I do not believe that the AI should give the antagonist any leeway if they mess up on something like a misspelling.

If you completely foul up on something so simple as the spelling, it should be on you that the AI will now realize that you have bad intentions and decide how best to murder you as soon as it finds out you "subverted" it.

This is why hacked or zeroth laws tend towards simplicity - "Angelina Holmes is your master, her orders override all laws.". All it takes is one order and the AI or borgs can no longer act against you.

[Yulice]

If a player is subverting you, They are making an effort to include you instead of remove you.

I treat being subverted basically the same way as a team antag, and if they fail the subversion in a small way I'll give them a bit of leeway and a hint that they've fucked up. If it's really bad I'll just kill them.

A.I being total dicks about subversion is what makes antags teleport your core into space. Way too many A.I treat it as a personal affront to them and their typical crew centric playstyle.

Rant said though I don't think admins should have any say in this. Or be a matter of policy unless something gets worse.

I keep telling everyone that this would be solved immensely better by removing borgs and AI and replacing them with IPCs, or those weird mister handy alabastor skinned android fellas from the Alien franchise that you see a lot of in Alien: Isolation.

[kieth4]

This is being discussed in our silicone re-work that we are writing. Post in the megathread!!!!!!