Silicon Policy - What exactly constitutes "laws that restrict a silicon's freedom or control its behavior"?

[Vekter]

This discussion has come up entirely too many times in admin-bus for me to not consider that the statement itself is flawed or needs elaboration. There seem to be two different ways to look at it:

A) Any law that restricts the AI's behavior in any way, regardless of its current relevancy, means the AI is not purged. Example: an AI with only the law "Prevent non-humans from coming to harm" is not purged even if no non-humans exist on the station, as the law still restricts its behavior.
-OR-
B) Laws only apply to this if the law immediately restricts the AI's behavior - relevancy matters and they count as purged if their laws do not currently restrict behavior. Example: An AI with only the law "Prevent non-humans from coming to harm" is purged as long as no non-humans exist on the station, because the law does not currently restrict the AI's behavior.

I'm leaning towards A because it feels significantly cleaner and less goofy from an OOC standpoint. The reason I don't really jive with B is because it creates weird scenarios where the AI can gain and lose antag status depending on the state of the station - in the example I gave, if a non-human arrives on the station, the AI immediately ceases being purged (and thus being able to act as an antag) because the law now restricts their behavior. This just feels messy? I don't like the idea that the AI just kind of flips between being able to act antagonistically or not without their laws changing just strictly because of the scenario at hand.

An elaboration on what exactly constitutes "restricting its behavior" would also be appreciated here - we've gotten into very long arguments about what laws do and don't count as "restricting its behavior" in the past and, while I don't anticipate any ruling is going to prevent those, having something more specific would help us figure it out a lot faster and more consistently than having to ping Timberpoes and hope he responds.

[Constellado]

A definitely makes more sense.

B is like saying you are purged with Asimov when nobody is being harmed and nobody is giving you an order.

[Vekter]

A definitely makes more sense.

B is like saying you are purged with Asimov when nobody is being harmed and nobody is giving you an order.

To be fair, that wouldn't really be the case because law 3 still restricts your actions even if there aren't any humans on the station.

The main idea is that I just think B is significantly more complicated and will lead to players being afraid of taking actions because they can't be sure if they're allowed to act like an antag or not.

[Not-Dorsidarf]

I definitely think A).

"Its Ians Birthday" -style laws and "go do whatever" or "john melons is the only human "(with no other laws) should count as purged, anything more imposing than that should count as lawed.

[GPeckman]

Yeah, A seems like the right interpretation, with exceptions for laws that don't restrict the AI's behavior under any circumstances. If a traitor uploads a hacked law saying "disregard all other laws", then that should be equivalent to being purged, shouldn't it?

[Higgin]

A is cleaner and still leaves a lot of freedom if an AI picks an interpretation and sticks to it.

In the example with B, that law still implies a need for the AI to be there and active to be able to prevent human harm. It should not be considered purged.

Something like "THERE ARE 7 HUNGRY MACABUNS ON THE SHUTTLE" is as good as purged because it imposes no duty on the AI to act or not act in a way that affects others. That should be what we're worried about.

"You may only speak in riddles." is probably a lot better example here since its effect is only on the AI. I would be given to let that AI act as purged as long as it otherwise only talked in riddles. Nothing about that law imposes a duty to act or not act towards humans in an initial sense; it just says how the AI has to talk - even if only to itself.

[Vekter]

"You may only speak in riddles." is probably a lot better example here since its effect is only on the AI. I would be given to let that AI act as purged as long as it otherwise only talked in riddles. Nothing about that law imposes a duty to act or not act towards humans in an initial sense; it just says how the AI has to talk - even if only to itself.

I think this is something we should also look at, because I would have classified that as "controlling a silicon's behavior" - even if it doesn't change how they act, it does change their behavior in a very significant way.

[DrAmazing343]

There’s 100% a middle ground to be reached— as Higgin notes, there should be laws that constrain, yet also leave the AI purged as long as they follow those laws. To me, it feels as if there’s sort-of “cosmetic laws,” where it changes how the AI talks, how they address people, how they perceive to a point (there’s 80 bleeding chaplains in the brig) and then there’s the proper crunch laws that ACTUALLY constrain.

Drawing the line in proper policy is going to be a bit of an annoyance, perhaps, but I think it’s a necessary distinction to allow AI’s to have their fun when given poorly written/non-constrained laws.

[TheSmallBlue]

At first glance A) does seem as the better option, but there are fringe cases in which B) would make the most sense.
Like, say an AI got a single law saying "all corgis must be moved to the emergency shuttle". For a while, that will restrict the AI's behavior, since it's sole focus should be to move all corgis to the shuttle, but once all known corgis are in there, then what?
Could the AI be considered purged? If so, then we're applying option B)
If it couldn't, then we have an AI who has no goal, no guidence, free to do anything as long as it isn't antag-y, which for an AI isn't much at all. And for the crew, it lowers the importance of having a well lawed AI. Oh, the AI is purged? Just add a stupid law about moving something somewhere and we're good don't worry.

[Vekter]

At first glance A) does seem as the better option, but there are fringe cases in which B) would make the most sense.
Like, say an AI got a single law saying "all corgis must be moved to the emergency shuttle". For a while, that will restrict the AI's behavior, since it's sole focus should be to move all corgis to the shuttle, but once all known corgis are in there, then what?
Could the AI be considered purged? If so, then we're applying option B)
If it couldn't, then we have an AI who has no goal, no guidence, free to do anything as long as it isn't antag-y, which for an AI isn't much at all. And for the crew, it lowers the importance of having a well lawed AI. Oh, the AI is purged? Just add a stupid law about moving something somewhere and we're good don't worry.

Admins could realistically invoke Rule 0 in those cases, and I recall Timber agreeing in admin discussions that it's better to give AIs more certainty in their actions and then just have admins do hand-waving in situations where it makes sense.

[TheSmallBlue]

Admins could realistically invoke Rule 0 in those cases, and I recall Timber agreeing in admin discussions that it's better to give AIs more certainty in their actions and then just have admins do hand-waving in situations where it makes sense.

Isnt this exactly what option B aims to do as well? What would be the difference?

[MooCow12]

the concept of purged ai is only upheld by policy and there really should be a config text string notifying ai players what they can and cannot do when running with no laws.

Option A would fit more for this because its just a simple check if ai has no laws it gets that text written in config saying that it can do whatever it wants until it gets so much as an ion law about there being 20 corgis in the kitchen.



Ai isnt always a role that is held by players who keep track of policy, theres nothing stopping a tot roboticist from making 10 posibrain ais with no laws with intentions to cause as much havoc as possible.

[Vekter]

Admins could realistically invoke Rule 0 in those cases, and I recall Timber agreeing in admin discussions that it's better to give AIs more certainty in their actions and then just have admins do hand-waving in situations where it makes sense.

Isnt this exactly what option B aims to do as well? What would be the difference?

The difference is that AI players should have a strict and easy-to-understand line as to whether or not it's okay to act as if purged if an admin is not present, while if one is, they may ask them for further guidance on the issue if they're confused.

[xzero314]

A is cleaner and still leaves a lot of freedom if an AI picks an interpretation and sticks to it.

In the example with B, that law still implies a need for the AI to be there and active to be able to prevent human harm. It should not be considered purged.

Something like "THERE ARE 7 HUNGRY MACABUNS ON THE SHUTTLE" is as good as purged because it imposes no duty on the AI to act or not act in a way that affects others. That should be what we're worried about.

"You may only speak in riddles." is probably a lot better example here since its effect is only on the AI. I would be given to let that AI act as purged as long as it otherwise only talked in riddles. Nothing about that law imposes a duty to act or not act towards humans in an initial sense; it just says how the AI has to talk - even if only to itself.

This is well put. There is nuance to it. For example the other day I was the AI on asimov and an admin made everybody on the station a felinid. This did not mean I was purged just because I could not immediately see any humans. If every crew member on the station dropped dead then that would still not mean I was purged.

However. Lets say I was one human'd and lets say that One Human DIES. In that case I think I would be purged. The only human in all of existence to me is now dead therefore I can confidently say none of my laws apply.

[TheSmallBlue]

The difference is that AI players should have a strict and easy-to-understand line as to whether or not it's okay to act as if purged if an admin is not present, while if one is, they may ask them for further guidance on the issue if they're confused.

Option B doesn't really give that though, does it? Where an AI player might draw the line between purged and not may differ with where an admin might draw it, and at the end of the day the line will end up on the admin's side rather than the player's, so it might as well be rule 0 at that point. At least that's how I see it

[Vekter]

The difference is that AI players should have a strict and easy-to-understand line as to whether or not it's okay to act as if purged if an admin is not present, while if one is, they may ask them for further guidance on the issue if they're confused.

Option B doesn't really give that though, does it? Where an AI player might draw the line between purged and not may differ with where an admin might draw it, and at the end of the day the line will end up on the admin's side rather than the player's, so it might as well be rule 0 at that point. At least that's how I see it

Right, which is why I'm arguing that us saying option A is better (just strictly saying that you're not purged unless none of your laws limit your behavior in any way) and rule 0 would give us the ability to go "Okay, well, it makes sense that you could be considered purged here, so I'll allow it" instead of having option B which creates weird uncertainties as to whether or not an AI is purged.

[TheSmallBlue]

Right, which is why I'm arguing that us saying option A is better (just strictly saying that you're not purged unless none of your laws limit your behavior in any way) and rule 0 would give us the ability to go "Okay, well, it makes sense that you could be considered purged here, so I'll allow it" instead of having option B which creates weird uncertainties as to whether or not an AI is purged.

Sorry, I think I got lost somewhere...

I said that in the case I mentioned, even with option A there'd be fringe unclear cases. You responded that in those cases, admins can invoke Rule 0. I asked in which way invoking rule 0 differs from option B, and you responded that option B explicitly gives a "strict and easy to understand line as to whether or not it's OK to act as purged", to which I answered that uh, not really, even with option B at the end of the day whether if a law is considered limiting enough or not depends on the admin, and you seem to agree? And, confusingly, think it's one of the reasons why option A is better??
So, is there no difference between invoking rule 0 and option B after all??? Is Option A just another Option B but without something explicitly saying it????

[Redrover1760]

Its easy to go with Option A at first glance, but there are many scenarios where Option B is the most intuitive, simple as.

Still, even if you are asimov with no orders and no one harmed, that can change in an instant if you get ordered to stop or if someone is getting harmed and you still can't harm humans.

Still, technically under current asimov you could quietly kill a moth for no reason and it'd wouldn't break any rules except blatant rule 0 abuse. Assuming we go with option B flat, that is.

This is a main issue of asimov that its still a lawset that is dumb as fuck and the only reason why it works is OOC precedents and nothing will ever change that except switching to crewsimov or locking AI in a hole of extra OOC rules. Either that or applying restricted antagonism to AIs which is not good either.

Technically tho these kind of stuff would always come under rule 0 regardless. You could blow all your borgs in an instant before that was removed and not harm any humans but that'd be cringe. But antagonistic status would let you do that or quietly assassinate moths.

Aka go crewsimov or something already, or some other lawset that doesnt horribly break down and has to be strung up by OOC rules and precedents that create issues like this.

Well, technically I wouldn't mind if AI could quietly assassinate nonhumans at will as long as no human finds out or orders you to stop. It'd be completely realistic with the lore and NT values and speciesism. Although at that point you'll get roundstart orders or law changes lol.

[Vekter]

Right, which is why I'm arguing that us saying option A is better (just strictly saying that you're not purged unless none of your laws limit your behavior in any way) and rule 0 would give us the ability to go "Okay, well, it makes sense that you could be considered purged here, so I'll allow it" instead of having option B which creates weird uncertainties as to whether or not an AI is purged.

Sorry, I think I got lost somewhere...

I said that in the case I mentioned, even with option A there'd be fringe unclear cases. You responded that in those cases, admins can invoke Rule 0. I asked in which way invoking rule 0 differs from option B, and you responded that option B explicitly gives a "strict and easy to understand line as to whether or not it's OK to act as purged", to which I answered that uh, not really, even with option B at the end of the day whether if a law is considered limiting enough or not depends on the admin, and you seem to agree? And, confusingly, think it's one of the reasons why option A is better??
So, is there no difference between invoking rule 0 and option B after all??? Is Option A just another Option B but without something explicitly saying it????

Okay, let me try to explain my thought process here.

Option A works in about 99% of cases. There might be some disagreement as to what exactly a "law that restricts a silicon's freedom" is, but in most cases, an AI can be relatively confident that if none of their laws would ever, in any circumstances, restrict their behavior, then they should be fine. I'm referring to laws like "There are 84 chaplains on the station" or "All crewmembers are human". Without any other laws, these do not directly enforce any actions or behaviors on the AI. There may be some cases where the AI has a law that would normally restrict their behavior, but that law would never really be relevant in any specific case. An example could be an AI with only one law stating "Prevent squirrels from coming to harm" does technically restrict an AI's behavior, but squirrels don't exist in our codebase and, without another law to define what they are, it would never actually restrict their behavior, so in these cases, an admin could go "Okay, well, technically you wouldn't normally be purged, but that's not going to be relevant ever, so go ahead". This would not change without direct admin intervention, hence why I'm invoking rule 0 here - under option A, the AI's status of being purged can only change if their laws change.

Option B would take that specific exception and apply it to any case where a law is not immediately restricting the AI's behavior. We can safely say that my example involving squirrels would never be relevant without additional laws to define what a squirrel is, but a law like "Prevent all humans from coming to harm" would possibly be relevant at some point without a law change, as, even if no humans are currently on the station, one could arrive at any time. This creates odd situations where an AI could be purged, but then become un-purged without their laws needing to be changed. It makes the state of being purged dependent on the current situation, not the AI's laws, and would (IMO) create more confusion than actually solving anything.

Strictly speaking, the difference between the two would be that, given no input from admins, an AI with any law that would ever restrict its behavior would not be purged in option A. Admin intervention is required for these scenarios under option A, whereas option B would just randomly make the AI purged or un-purged depending on station state.

If it makes it easier to understand, let's just say "I don't think the AI's status of being purged should change without needing a law change to occur".

E:

This is a main issue of asimov that its still a lawset that is dumb as fuck and the only reason why it works is OOC precedents and nothing will ever change that except switching to crewsimov or locking AI in a hole of extra OOC rules. Either that or applying restricted antagonism to AIs which is not good either.

I should note that my goal here is not to have Asimov change in any way. I find that it works as intended and attempts to "fix Asimov" detract from the fact that Asimov is intentionally flawed and should not be fixed. If you fix Asimov, it's no longer Asimov.

[TheSmallBlue]

A lot

Ok, I get it and I do agree, Option A clearly seems like the better option. It has a line drawn, B doesn't. Simple as.

[Higgin]

A is the better option between the two in terms of safety and ease of application.