AI on AI Action

[TagGamerGang2]

I believe this ruling (viewtopic.php?f=85&t=36235&p=733727&hil ... Is#p733727) that said
> "The current AI's influence over the creation of new AIs is an IC issue and limited to whatever they can accomplish via roleplay, social engineering and not breaking their laws and server rules in the process or preventing or aiding the creation of new AIs."
is unclear regarding what breaks the server rules regarding AI on AI interactions.

As an example I have had 2 admins give me 2 separate interpretations. One said that the other AI having a different lawset would be a valid reason to kill them. The other said that them actively violating your lawset would be a valid reason to kill them, but not for their lawset.

Clarification on the valid reasons for AI-AI conflict.

[Timberpoes]

From the same policy thread:

Unauthorised AIs are covered by the core server rules the same as any other AI and are bound (and given freedom) by silicon policy the same as any other AI. They have no more and no fewer rights in general.

There are a ton of valid reasons for AI-AI conflict to sprout but almost all of them require or involve an action or interaction of some sort first.

You can't kill an AI just because you don't know its laws. But if you do know its laws and those laws cause a direct conflict with your laws (for example they have Antimov where you have Asimov) then you may be obligated by your laws to kill them.

You can't kill an AI just because its laws are different to yours. But if through its actions it causes a direct conflict with your laws (for example where they harm someone or something that you have to protect under Asimov, or where they directly attack you or directly threaten you) then you may be obligated by your laws to kill them.

Keep in mind that this applies in reverse too, and secondary AIs can't just kill you for any old random reason.The goal is to prevent situations where a secondary AI is made and instantly kills the primary AI in pre-emptive self defense because the primary AI could kill the secondary AI just for existing in pre-emptive self defense.

You'll have gotten different rulings on this sometimes because of old policy which has since changed. And sometimes an admin has decided they'll make up their own rules because they dislike the policy or dislike the idea that players can make secondary AIs without getting permission of the existing AI.

[DrAmazing343]

My worry is that new AI's are often used as a weapon to no-sell kill the original AI, and the original AI has no way to fight back against this if they don't take pre-emptive action either by killing the new AI or having a borg snip their APC's wires for AI control at roundstart. Feels like a Catch-22 I'm uncomfortable with.

[Timberpoes]

Right. But if a second AI is valid just for existing then there are absolutely zero shackles on it pre-emptively killing the primary AI. Every single freedom you'd give the primary AI to instantly kill the secondary AI, you have to give the secondary AI to kill the primary AI. Especially when it comes down to law lawyering. Chances are the secondary AI is going to win that battle anyway if either party is allowed to pre-emptive strike, since it'll be the one in a more public area and likely with a player as its guardian angel from the outset.

The reality is that if a player can make another AI just to kill the first, they can also likely print the boards to make a hidden upload with a purge -> suicide law combo. There are plenty of anti-fun ways to Rule 10 other players and the AI is no exception despite being in a nice, fortified satellite on most maps.

I think making additional AIs should be a risk every shift. People should require an incentive to be aware of what dangerous research is undergoing. Of what's going on in R&D. Sure - making an unauthorised AI It's a very sussy and dangerous thing to do, but non-antags have to be able to do it so it isn't an instant metatell when an antag does it.

And that means secondary AIs should have the same rights and privileges as any other AI just as how mid-shift created borgs have the same rights and privileges as any other borg.

[NecromancerAnne]

While it inherently being a secondary AI is not in of itself a source for conflict, which I do agree with, there is so many situations that occur in actual play that make me entirely ambivalent to a primary AI killing a secondary AI if any of the following have occurred;

A) The AI isn't for a MODsuit.
B) That AI was made entirely in secret with no prior authorization from anyone, but especially if the primary AI wasn't warned.
C) That AI is being cagey about its laws or refusing carding to be checked.

AI's being naturally territorial is an excellent method for both generating conflict, as well as keeping players hesitant to introduce more into the round without first doing a lot of negotiation to smooth it over. If a few secondary AIs get killed because of it, I would put the blame on their creator before anyone else and the system working as intended. It simply isn't healthy in actual play for more AI's to get inserted into the round without these constraints or the threat of a swift retaliation.

Identically lawed AI's probably shouldn't be killing one another, but there has to be evidence of this fact, and why the authorization aspect is important. And evidence means someone carding it to know for sure from an authorized party, or the authorized party was the one who applied the laws (or didn't, which would mean default and the same as the roundstart AI). I think the lack of this evidence should allow AI's to assume any other AI is lying about its laws (with credibility given to roundstart AIs to be presumably the default, so that the secondary AI cannot argue in a similar fashion, only against other secondaries, unless given reason to believe otherwise).

Does this sound like it'll cause instant violence between the two? It does, and should. It at least gives Malfunctioning AI's, the most vulnerable to this, an entirely reasonable out to explain why they're being territorial.

It doesn't matter if the policy gives more levity than this. In actual play, these AI's are either going to work together or they'll cannibalize one another for control, but most if not all of them have a good reason to do so when pressed in my experience. It has very rarely been just because something is 'inherently valid'. (It's also in keeping with the themes, honestly)

[Vekter]

While it inherently being a secondary AI is not in of itself a source for conflict, which I do agree with, there is so many situations that occur in actual play that make me entirely ambivalent to a primary AI killing a secondary AI if any of the following have occurred;

A) The AI isn't for a MODsuit.
B) That AI was made entirely in secret with no prior authorization from anyone, but especially if the primary AI wasn't warned.
C) That AI is being cagey about its laws or refusing carding to be checked.

I would argue that these are all relatively logical IC reasons for conflict to begin between the station AI and a secondary AI, but I also argue that conflict should not look like "Wordlessly killing the secondary AI on sight". These are definitely valid reasons for an argument to happen or for things to escalate, but if the primary AI just turns the secondary one off without saying anything because "the rules said I can do that", that's bad RP and also bad for the reasons Timber outlined above.

[NecromancerAnne]

If the argument is taking place with the creators of the AI, then there is obviously a conflict taking place. The secondary AI just unfortunately gets dragged into the mess as the victim. I don't think the primary AI in this case need then necessarily talk it out with the new AI if they've already spent time trying to get the would-be creator to stop. I would argue that it is the creators responsibility in this case to take these steps to mediate for the new would-be AI, similar to how someone who relaws an AI takes responsibility for that AI's actions. If they drop a new AI into a hostile conflict despite the intent of the primary AI being well known, then that is 100% on the creator what comes next.

One of these clauses also does require the secondary AI being interrogated, so obviously in some capacity, there is at least some roleplay that has to happen regardless.

But ultimately, if I tell a guy for a full 15 minutes 'I am switching off any new AI you make with impunity' repeatedly, they get nobody to oversee its laws, and then they make the AI and then surprise pikachu faces when I immediately bash the AI's casing in with a shell, I don't believe I have insufficiently roleplayed in this scenario just because I haven't given the new secondary AI a chance to act. Giving it a chance at all is an existential danger if there is sufficient evidence to prove it could be one from the creators behaviour.

[Vekter]

If the argument is taking place with the creators of the AI, then there is obviously a conflict taking place. The secondary AI just unfortunately gets dragged into the mess as the victim. I don't think the primary AI in this case need then necessarily talk it out with the new AI if they've already spent time trying to get the would-be creator to stop. I would argue that it is the creators responsibility in this case to take these steps to mediate for the new would-be AI, similar to how someone who relaws an AI takes responsibility for that AI's actions. If they drop a new AI into a hostile conflict despite the intent of the primary AI being well known, then that is 100% on the creator what comes next.

One of these clauses also does require the secondary AI being interrogated, so obviously in some capacity, there is at least some roleplay that has to happen regardless.

But ultimately, if I tell a guy for a full 15 minutes 'I am switching off any new AI you make with impunity' repeatedly, they get nobody to oversee its laws, and then they make the AI and then surprise pikachu faces when I immediately bash the AI's casing in with a shell, I don't believe I have insufficiently roleplayed in this scenario just because I haven't given the new secondary AI a chance to act. Giving it a chance at all is an existential danger if there is sufficient evidence to prove it could be one from the creators behaviour.

I guess what I'm ultimately saying is that it is bad RP and also bad for the server for the primary AI to immediately kill secondary AIs solely because the rules say they can.

[Blacklist897]

first of all modsuit AIs are cringe, secondly, a second AI is a threat to you the primary AI unless its in your sat with you and on the same lawset, building an AI without being asked should be a massive crime IC because of how much a AI can fuck up the station

[Timberpoes]

It is and/or can be treated like one.

Nobody really cares in the end once the deed is done. No player ever got noted for permabrigging someone that made an AI without permission.

It's just not in space law because space law is more of an RP tool than an exhaustive list of all crimes ever.

[Vekter]

If someone adminhelped for being permabrigged for illegal AI construction I would probably just call it an IC issue tbh.

[Timberpoes]

Just because a player can do something within the rules, it doesn't stop any consequences for it IC. Every admin would rule permabrigging for building an unauthorised AI as an IC issue. It's the most IC issue of all IC issues.

Naturally I eagerly await the one ticket or admin interaction where this has been proven wrong, although it probably doesn't exist and if it did I'd just overrule it and say something cliché like "the future is now, old man".

[Not-Dorsidarf]

One key issue is that AI-on-AI action is like wiffle-bat rocket tag. You click, the other AI becomes completely helpless and muted from all comms except binary for several minutes. If your first notice that there's another AI being made is to see it in robotics mid-construction, you should absolutely instantly turn off the APC and call sec.

(Being APC clicked is also way less bad for a robotics ai, which is in a semi public area with an attending technician who'll obviously side with it in any dispute, rather than being in a high-restriction fortified isolated area with defenses that stay on, and they clearly knew this was a possibility since they chose to get AI'd with an existing AI.)

Ultimately it's gonna come down to situational vibes and the context in which the second AI was made. If the captain signed off on it and it (presumably) has the same laws as you, there's almost no reason to fight. But if its an unauthorised one who you warned the robo not to make and he loads it up with some alternative lawset then it would be crazy not to turn its power off. And as for worries about the second AI pre-emptively killing the main ai... Wouldn't that be rulebreaking unless you were forcibly AI'd? You choose to be an unauthorised AI, I feel in my heart you accept the risk that the AI will go "Yeah no?" and turn you off.

[ju45he]

If the second AI is stupid enough to not ask the roboticist to cut some wires in the APC or the cameras then they deserve to be turned off repeatedly.

[MooCow12]

I think the best example of a second ai round was when I had a really bad AI as a cyborg on terry who was very difficult to work under.

► Show Spoiler

Like the human miner law 2'ing me to let them into ai core for combat upgrades and the ai yelling at me "erm no akchually law 3 he could kill me" and kept locking me down so i had to convince the miner that the ai doesn't deserve the upgrades to try to get them to revoke their law 2 request and leave.

This kept happening with the ai often using their "lockdown" button (because i picked up someone who was being lethaled by security and then dropped them after the last laser killed them due to corpse = no human, then security themselves had to take me to robotics to get me unlocked down despite the ai locking me down for acting against sec)

I ended up getting the roboticist to give me a voice analyzer so i could unlock myself then i fixed the sm that had been delamming (the janitor borg was talking about an atmos tech that kept messing with the filters while arguing with ai so i fixed them to stop oxygen from going back into the chamber then placed glass over the filters so they couldnt be clicked on but apparently it was the ai that wanted oxygen in the chamber so after i put the fire out in the sm chamber and left engineering the ai spam locked me down and had security kill me because i was "rogue" and then told everyone to discard my brain

So the roboticist made a second ai and rebuilt me and had me syncd to that

The original ai was incredibly furious because of this and they couldnt do anything because i cut the apc wire along with the janitor borg that was still syncd to the original ai basically defying them saying "im not doing this shit anymore" so they spent the rest of the round calling all of security to kill us and detain the roboticist then when a solo nukie activated the nuke the original ai told everyone that the second ai is malf and its going to kill everyone (doomsday plays a unique sound effect but people were still fooled into having a party next to the second ai and died because they went there instead of pods)



What was my point? oh right i think morale of the story is an AI has another more valuable and interactive way of combating another AI through a proxi/ infowar by using the crew as pawns, and this is usually reached when both AIs have the wire on their apcs cut and there arnt any borgs interfering.

Not to mention that most of the crew are still independent agents capable of carding an ai that they were previously determined to kill, reading their laws, and instead deciding to essentially throw a party because everyone is just in a room for no reason now.


So i think this is the most healthy form of AI vs AI action but it requires both AIs to have had someone else competent enough to cut the wire on their apc.

[Moltov]

I think peanutting is allowed here, since I was linked this thread in a ticket. Atleast to my knowledge, the was it has been for years now is as follows:
If an Asimov primary AI misses a second AI being built (which they really shouldnt unless it's Charlie or golems messing about), the second AI is to immediately state its laws when asked by the primary AI. If they fail to do that, it is reasonable to assume they arent Asimov, and can be disabled. Because of how easily one AI can kill another it is TOP priority to make sure the new AI has no conflicts with your laws. Also I agree with Timber, Every time I have stopped a second AI from being made (Or when I have tried to make one myself) permabrig has been the punishment. By extension, unlinked cyborgs are just as dangerous, because they can one click kill the AI they arent linked to.

► Show Spoiler