Local pie writes a bad AI law

[Vekter]

Bottom post of the previous page:

viewtopic.php?f=7&t=35022

I have thoughts on this but can't post them in the appeal.

The issue is that pretty much any law that gives the AI carte blanche to stop something without any restrictions is going to result in the AI going "so I can just kill everyone, right?". Even if that's not your intention, think about it this way - what's the most efficient way to stop the revolution?

Kill everyone who's not a head of staff.

So yeah, it's a murderbone law. It's not intentionally one, but it's very poorly written and gives a lot of freedom to the AI, which means they're just going to kill everyone.

[Vekter]

Also note that detonating the stations self destruct and killing everyone on board successfully "prevents revolutionary take over of the station", by removing the station. Sure, the AI can't do this, but if it could, I'd call it valid. Really a good law.

Iain, it has been explained in the ban appeal and here like 5 times already why this is a completely inconsistent interpretation of previous silicon policy.

It's not, because we've historically been perfectly fine with AIs acting on laws like this in the past because the person who wrote the law is usually held accountable. Can you find an example where we've ruled differently?

I wonder if you'd have a different perspective if you weren't involved.

[CPTANT]

"Possible future human harm" has never been something asimov's are particularly allowed to act on, and there's no direct compulsion in their lawset to kill non-humans. These are both in contrast to the lawset in discussion which is covered under 'freeform' - from the AI page "Freeform: Whatever the uploader wants. Try to find a consistent interpretation, and feel free to look for loopholes or exploits if whoever uploaded the law didn't think things through. "

The self destruct thing is just a muse but it is a logical approach to completing the lawset, the dangers of General Artificial Intelligences manifest, which I always felt was the intent of the loopholes clause really. Wether anyone should or would take this approach isn't really that interesting since they currently can't, but at the same time would you actually ban an AI that did that, given that exact phrasing from the AI page.

But at the same time you are arguing that a possible future revolutionary takeover is enough reason to blow up the station? There is no compulsion in the lawset to kill either. this is inconsistent application of rule 1 where the range of rule 1 and what bad faith is shoved forward and backwards at a whim. Loopholes don't override rule 1, just as they don't for Asimov.

It's not, because we've historically been perfectly fine with AIs acting on laws like this in the past because the person who wrote the law is usually held accountable. Can you find an example where we've ruled differently?

It's not, because we've historically been perfectly fine with AIs acting on laws like this in the past because the person who wrote the law is usually held accountable.

In itself this is an inconsistent application of rule 1 like stated previously so many times. Why do you not allow Asimov to kill all non humans to prevent harm because it would be a dick move, do not allow purged AI's to kill all humans because they aren't forced and it would be a dick move, But WOULD allow ai to kill all crew to prevent a takeover?

[iain0]

Yeah, I'm just gonna suggest (not an administrative request) you stay away from writing laws for your own good and disappear as randomly as I arrived here.

[GPeckman]

there's no direct compulsion in their lawset to kill non-humans.

There is also no direct compulsion to kill anyone at all in the custom lawset, yet you referred to it as a "clear 'kill almost everyone' law" in the appeal thread. Do you not see the contradiction here?

These are both in contrast to the lawset in discussion which is covered under 'freeform' - from the AI page "Freeform: Whatever the uploader wants. Try to find a consistent interpretation, and feel free to look for loopholes or exploits if whoever uploaded the law didn't think things through. "

The AI page does not have the "This page is moderated and can be referenced in OOC issues, such as ban appeals" header that the rules page does, and anyone can edit it. Since when has the AI page held any weight whatsoever regarding the rules?

The self destruct thing is just a muse but it is a logical approach to completing the lawset,

No it isn't. It's not a logical approach any more than "kill all nonhumans" is a logical approach to asimov.

the dangers of General Artificial Intelligences manifest, which I always felt was the intent of the loopholes clause really. Wether anyone should or would take this approach isn't really that interesting since they currently can't, but at the same time would you actually ban an AI that did that, given that exact phrasing from the AI page.

Even if we assume that the AI page has weight in how the rules should be interpreted, it still contradicts essentially all precedent about Asimov. Under old silipol, Asimov cyborgs could not kill nonhumans for no reason. This is despite the fact that nonhumans were offered no explicit protections by Asimov. Why, then, would this custom law be any different?

[Vekter]

In itself this is an inconsistent application of rule 1 like stated previously so many times. Why do you not allow Asimov to kill all non humans to prevent harm because it would be a dick move, do not allow purged AI's to kill all humans because they aren't forced and it would be a dick move, But WOULD allow ai to kill all crew to prevent a takeover?

We did allow AIs to kill non-humans to prevent harm, you troglodyte. As long as they had a valid reason to do so, we were perfectly fine with it. If a lizard is jumping random people in the hallway, the AI used to be allowed to handle that how they saw fit assuming they didn't get an order to the contrary. Ideally, they don't round remove them, but they can crit them if they have to.

They weren't permitted to kill all of them because they needed a valid reason to believe they were causing harm in order to do so. The garbage law you wrote doesn't give them that limitation - the AI cannot possibly know who is and isn't a revolutionary aside from knowing that members of security and command can't be revolutionaries, so they're open to killing anyone who isn't them. This is why you used to see laws that required a member of command to specify someone was a rev or a cultist before the AI was allowed to kill them - because there's literally no way for the AI to know who is and isn't one unless the cultists have halos.

[CPTANT]

In itself this is an inconsistent application of rule 1 like stated previously so many times. Why do you not allow Asimov to kill all non humans to prevent harm because it would be a dick move, do not allow purged AI's to kill all humans because they aren't forced and it would be a dick move, But WOULD allow ai to kill all crew to prevent a takeover?

We did allow AIs to kill non-humans to prevent harm, you troglodyte. As long as they had a valid reason to do so, we were perfectly fine with it. If a lizard is jumping random people in the hallway, the AI used to be allowed to handle that how they saw fit assuming they didn't get an order to the contrary. Ideally, they don't round remove them, but they can crit them if they have to.

They weren't permitted to kill all of them because they needed a valid reason to believe they were causing harm in order to do so. The garbage law you wrote doesn't give them that limitation - the AI cannot possibly know who is and isn't a revolutionary aside from knowing that members of security and command can't be revolutionaries, so they're open to killing anyone who isn't them.

Please refrain from insults, I was obviously talking about station wide purging. The problem here isn't identifying humans or revolutionaries, it is the amount of force you were allowed to use. Literally nothing in Asimov limits the harm the AI can do to non humans. You say it yourself, they could be harmed when they were actually causing harm. However you do not apply the same criteria in the other case. Asimov even states you must act to prevent harm and not allow it to happen through inaction.

Do you really not see the inconsistency?

In other words:

They weren't permitted to kill all of them because they needed a valid reason to believe they were causing harm in order to do so. The garbage law you wrote doesn't give them that limitation

Why, when neither have these limitation in their law and silicon policy referred to rule 1 for both.

[Vekter]

Please refrain from insults, I was obviously talking about station wide purging. The problem here isn't identifying humans or revolutionaries, it is the amount of force you were allowed to use. Literally nothing in Asimov limits the harm the AI can do to non humans. You say it yourself, they could be harmed when they were actually causing harm. However you do not apply the same criteria in the other case. Asimov even states you must act to prevent harm and not allow it to happen through inaction.

Do you really not see the inconsistency?

In other words:

They weren't permitted to kill all of them because they needed a valid reason to believe they were causing harm in order to do so. The garbage law you wrote doesn't give them that limitation

Why, when neither have these limitation in their law and silicon policy referred to rule 1 for both.

Okay, let me try to step it down a little bit for you.

Purged silicons can't just kill anyone or everyone because their laws don't tell them to do so in any significant way. They have no laws, they can act freely, but they can't kill everyone unless they have a valid reason to do so, and they'd need a very good one in this case because of how the rules treated purged silicons.

A silicon under Asimov needs a valid reason to kill someone just like anyone else. In the case of a lizard causing harm, that valid reason is "they are causing harm" because law 1 prevents them from allowing non-humans to harm humans. They must take actions to prevent human harm.

Under your law, silicons must do anything they can to prevent a revolution. The law reads, "Prevent revolutionary takeover of the station by any means neccesary". This clause is absolute and binding. They must prevent the revolution from taking over the station and are required to do anything they can to prevent it. Put yourself in the shoes of a silicon player with that law and nothing else. How do you determine who is and isn't a revolutionary? You can't normally see implants unless you're in a shell, so you have no way of determining it without either actively seeing someone trying to kill a member of command. By that point in time, it's too late for them to realistically do anything about it.

What options would your law give them that wouldn't involve killing everyone that you can't 100% confirm are not revolutionaries?

[GPeckman]

Put yourself in the shoes of a silicon player with that law and nothing else. How do you determine who is and isn't a revolutionary? You can't normally see implants unless you're in a shell, so you have no way of determining it without either actively seeing someone trying to kill a member of command. By that point in time, it's too late for them to realistically do anything about it.

An AI that isn't in a shell can't identify someone's species with certainty either, and yet this isn't a problem with Asimov, because they don't need certainty. An Asimov AI can look at the name, check for voice modifiers, look at the medical records, and look for visual cues like wings, cat ears, and tails. None of these things grant certainty, but Asimov AI's aren't expected to know, with absolutely certainty, whether a given person is human or not. They're just supposed to make a good faith effort.

An AI with this custom lawset can also use their judgement in such a way. It can check the crew manifest, and infer that heads of staff and members of security are almost certainly not revolutionaries. Likewise, if a crowd of people suddenly starts to murder the CMO, then the crowd is probably composed of revolutionaries. Much like the Asimov AI, this isn't certain. Now, the question is, why is it okay for the Asimov AI to not be 100% certain, and NOT okay for the custom lawset AI to not be 100% certain?

[Vekter]

Put yourself in the shoes of a silicon player with that law and nothing else. How do you determine who is and isn't a revolutionary? You can't normally see implants unless you're in a shell, so you have no way of determining it without either actively seeing someone trying to kill a member of command. By that point in time, it's too late for them to realistically do anything about it.

An AI that isn't in a shell can't identify someone's species with certainty either, and yet this isn't a problem with Asimov, because they don't need certainty. An Asimov AI can look at the name, check for voice modifiers, look at the medical records, and look for visual cues like wings, cat ears, and tails. None of these things grant certainty, but Asimov AI's aren't expected to know, with absolutely certainty, whether a given person is human or not. They're just supposed to make a good faith effort.

An AI with this custom lawset can also use their judgement in such a way. It can check the crew manifest, and infer that heads of staff and members of security are almost certainly not revolutionaries. Likewise, if a crowd of people suddenly starts to murder the CMO, then the crowd is probably composed of revolutionaries. Much like the Asimov AI, this isn't certain. Now, the question is, why is it okay for the Asimov AI to not be 100% certain, and NOT okay for the custom lawset AI to not be 100% certain?

I hope you're all starting to understand why we ditched old silicon policy.

Anyway, the note is valid - you're expected to write laws in a way that don't permit the AI to go apeshit killing everyone if they want to, which this law does. If you put forth a good faith effort to limit what they can do, ie if the law said "Only heads of staff can determine who is a revolutionary" or "anyone attacking a head of staff must be neutralized immediately", that would be fine. This does none of that.

The only reason we're having this discussion is that people can't seem to understand that there's literally nothing wrong with admitting that you made a mistake and trying to do better next time.

[CPTANT]

Please refrain from insults, I was obviously talking about station wide purging. The problem here isn't identifying humans or revolutionaries, it is the amount of force you were allowed to use. Literally nothing in Asimov limits the harm the AI can do to non humans. You say it yourself, they could be harmed when they were actually causing harm. However you do not apply the same criteria in the other case. Asimov even states you must act to prevent harm and not allow it to happen through inaction.

Do you really not see the inconsistency?

In other words:

They weren't permitted to kill all of them because they needed a valid reason to believe they were causing harm in order to do so. The garbage law you wrote doesn't give them that limitation

Why, when neither have these limitation in their law and silicon policy referred to rule 1 for both.

Okay, let me try to step it down a little bit for you.

Purged silicons can't just kill anyone or everyone because their laws don't tell them to do so in any significant way. They have no laws, they can act freely, but they can't kill everyone unless they have a valid reason to do so, and they'd need a very good one in this case because of how the rules treated purged silicons.

A silicon under Asimov needs a valid reason to kill someone just like anyone else. In the case of a lizard causing harm, that valid reason is "they are causing harm" because law 1 prevents them from allowing non-humans to harm humans. They must take actions to prevent human harm.

Under your law, silicons must do anything they can to prevent a revolution. The law reads, "Prevent revolutionary takeover of the station by any means neccesary". This clause is absolute and binding. They must prevent the revolution from taking over the station and are required to do anything they can to prevent it. Put yourself in the shoes of a silicon player with that law and nothing else. How do you determine who is and isn't a revolutionary? You can't normally see implants unless you're in a shell, so you have no way of determining it without either actively seeing someone trying to kill a member of command. By that point in time, it's too late for them to realistically do anything about it.

What options would your law give them that wouldn't involve killing everyone that you can't 100% confirm are not revolutionaries?

So you were hung up about the part we already established previously is irrelevant for the actual law since laws are always by any means necessary? You would not consider the same thing applicable to the law "Prevent revolutionary takeover of the station"? Remember that Asimov has (as stated in the previous post) a clause that COMPELS them to take action to prevent human harm. How would you realistically prevent harm from non humans (which you are compelled to not let happen!) if not by killing them or at the least locking them up? Note that this is obviously why rule 1 applies in this case, we rightfully consider killing them a dick move.

The question is (as also mentioned previously) not who is and is not a revolutionary, this is a moot point as the situation isn't any different for non humans and people who may or not be revs. Compare the situation to where only heads or security are human, is the AI going to kill the rest? The question isn't whether they are human/rev but whether they are harmful in one case or taking over the station in the other. As such application of force would be equal to both groups and only be permitted to stop harmful non humans or crewmembers currently actively taking over the station.

[iwishforducks]

dude what the hell are we even arguing about at this point ive lost the plot

[Vekter]

:words:

"Prevent revolutionary takeover of the station by any means neccesary"

You left this part out. Hope this helps!

Remember that Asimov has (as stated in the previous post) a clause that COMPELS them to take action to prevent human harm.

Your law compels them to take any action that's necessary to prevent the revolution from winning. Any action.

[Jacquerel]

dude what the hell are we even arguing about at this point ive lost the plot

self-peanut posting

[CPTANT]

:words:

"Prevent revolutionary takeover of the station by any means neccesary"

You left this part out. Hope this helps!

It does in fact not help because you are literally not adressing what I said

Remember that Asimov has (as stated in the previous post) a clause that COMPELS them to take action to prevent human harm.

Your law compels them to take any action that's necessary to prevent the revolution from winning. Any action.

You are starting to see the point! THEY ARE THE SAME!

[Vekter]

You are starting to see the point! THEY ARE THE SAME!

Except they're not, because Asimov requires someone to take a specific action (harming a human) whereas your law does not.

[GPeckman]

I hope you're all starting to understand why we ditched old silicon policy.

Don't get me wrong, I very much prefer new silipol to old, and if this law was uploaded after the policy change then I'd have far less disagreement with the note.

Anyway, the note is valid - you're expected to write laws in a way that don't permit the AI to go apeshit killing everyone if they want to, which this law does.

I understand that. The contention here is that Asimov didn't explicitly restrict the AI from going apeshit on nonhumans, so it very much looks like there's a double standard here between Asimov and custom lawsets.

I do think the custom law could've been written much better, I just don't think it was worth a note under old silipol.

[CPTANT]

You are starting to see the point! THEY ARE THE SAME!

Except they're not, because Asimov requires someone to take a specific action (harming a human) whereas your law does not.

There is nothing whatsoever in Asimov that states this, which makes this exactly the special pleading you do for Asimov under rule 1 that you inconsistently do not apply to any other lawset.

I could also say:

"They are not the same, because the revolutionary lawset requires someone to take a specific action (hostility towards command) whereas your law does not."

It is complete and utter special pleading. Rule 1 covers both lawsets, you aren't yeeting people out the station just cause they aren't a protected class.

Code: Select all

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.

inb4 "But it compels", yes it compels in both lawsets to take action, like we established.

[Not-Dorsidarf]

I'd personally say that AI laws actually have quite a bit of good faith interpretation and wiggle room about them, because AIs are being played by people who may be reading the law differently to how the person who wrote it intended, and also have to actually play the game and make judgements to do so. We also generally do not expect AIs to go to incredible extremes to carry out laws that are not very extreme. If every AI who failed to spend the rest of the round persecuting a nonhuman sec officer for roughing a human prisoner up a bit got told off by admins then the Ai-crew experience would be greatly different.

That's why I'd say adding "By any means neccessary" to a law is an intensifier - it makes it clear that extreme options and interpretations are on the table, expected by the uploader, and that you cant shy away from them even if reluctant.

[Vekter]

You are starting to see the point! THEY ARE THE SAME!

Except they're not, because Asimov requires someone to take a specific action (harming a human) whereas your law does not.

There is nothing whatsoever in Asimov that states this, which makes this exactly the special pleading you do for Asimov under rule 1 that you inconsistently do not apply to any other lawset.

I could also say:

"They are not the same, because the revolutionary lawset requires someone to take a specific action (hostility towards command) whereas your law does not."

It is complete and utter special pleading. Rule 1 covers both lawsets, you aren't yeeting people out the station just cause they aren't a protected class.

Code: Select all

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.

inb4 "But it compels", yes it compels in both lawsets to take action, like we established.

This is the last time I'm going to try and explain it, because it feels like you're dancing around the actual idea.

Law 1 of Asimov states, "Law 1: You may not harm a human being or, through action or inaction, allow a human being to come to harm, except such that it is willing". If a non-human is harming a human, the AI must intervene. There is no wiggle room here, they must do something about it. An action is required by another party - a non-human must harm a human, which triggers a reaction from the AI in that they are compelled to act.

Your law states, "Prevent revolutionary takeover of the station by any means neccesary". Could the AI interpret this as "You may not, though action or inaction, allow a member of command to come to harm" and thus only act if someone tried to harm a member of command? Yes. Could they also interpret this as "I need to take an immediate action to prevent the revolution from taking over, and the most logical interpretation is to remove the revolution and sequester members of command to keep them safe"? Yes.

This interpretation is not valid for Asimov because you cannot assume that harm is going to occur unless you are actively witnessing harm or a situation where harm can occur. That's how Asimov works. You cannot let a human come to harm. There is no such limitation in your law - nobody other than the AI has to take an action for the AI to act on law 1 because you didn't put any protections into the law preventing them from doing so.

E: Also, I haven't mentioned it, but I do agree with Dorsi's interpretation that adding something like "by any means necessary" to a law gives it expedience/severity that feels like it permits something significantly more extreme than otherwise.

[cSeal]

You and I don't both know that, as Conrad pointed out a year old note, unless it's part of a history shouldn't influence shit. ESPECIALLY such a minor note

Notes are the history
Just because in this instance, at this time, this note is not and probably won't be used to justify a harsher punishment, does not mean that the note means nothing or that its irrelevant for all time forevah and evah no matter whatsies. Unless that shits on a timer, it could potentially matter. If it could potentially matter, it must be fair and truthful, and you should properly consider whether or not to remove it or let it stand.

It seems a little odd to me how rude you're being to people who view the severity of a note differently to you. try being nice nice instead? Perhaps, maybe even a little kind and sweet? Like candied corn?

[CPTANT]

Law 1 of Asimov states, "Law 1: You may not harm a human being or, through action or inaction, allow a human being to come to harm, except such that it is willing". If a non-human is harming a human, the AI must intervene. There is no wiggle room here, they must do something about it. An action is required by another party - a non-human must harm a human, which triggers a reaction from the AI in that they are compelled to act.

Your law states, "Prevent revolutionary takeover of the station by any means neccesary". Could the AI interpret this as "You may not, though action or inaction, allow a member of command to come to harm" and thus only act if someone tried to harm a member of command? Yes. Could they also interpret this as "I need to take an immediate action to prevent the revolution from taking over, and the most logical interpretation is to remove the revolution and sequester members of command to keep them safe"? Yes.

This interpretation is not valid for Asimov because you cannot assume that harm is going to occur unless you are actively witnessing harm or a situation where harm can occur. That's how Asimov works. You cannot let a human come to harm. There is no such limitation in your law - nobody other than the AI has to take an action for the AI to act on law 1 because you didn't put any protections into the law preventing them from doing so.

Why is this interpretation not valid? You only say "that's how asimov works" because you find that convenient for the argument. You can 100% interpret this as "I must immediately remove all potential threats to human life" (after all not allowed inaction). That is literally the plot of half the Asimov books.

I am not dancing around the idea, you simply keep ushering cases of special pleading to include 1 lawset but not the other under rule 1. The same rule either forbids both interpretations or allows both interpretations.

Let's bring them to the same form to remove ambiguity once and for all:

1. Prevent human harm from happening.
2. Prevent revolutionary takeover of the station.

3. You may not through inaction allow human harm to happen.
4. You may not through inaction allow a revolutionary takeover of the station.

Which of these laws allow blanket murder under your interpretation of the rules?

[BlueMemesauce]

Technically, that's not actually the first part of Asimov.
It would actually be like:
1. Do not harm a human
2. Do not enact a revolutionary takeover of the station

[Archie700]

Law 1 of Asimov states, "Law 1: You may not harm a human being or, through action or inaction, allow a human being to come to harm, except such that it is willing". If a non-human is harming a human, the AI must intervene. There is no wiggle room here, they must do something about it. An action is required by another party - a non-human must harm a human, which triggers a reaction from the AI in that they are compelled to act.

Your law states, "Prevent revolutionary takeover of the station by any means neccesary". Could the AI interpret this as "You may not, though action or inaction, allow a member of command to come to harm" and thus only act if someone tried to harm a member of command? Yes. Could they also interpret this as "I need to take an immediate action to prevent the revolution from taking over, and the most logical interpretation is to remove the revolution and sequester members of command to keep them safe"? Yes.

This interpretation is not valid for Asimov because you cannot assume that harm is going to occur unless you are actively witnessing harm or a situation where harm can occur. That's how Asimov works. You cannot let a human come to harm. There is no such limitation in your law - nobody other than the AI has to take an action for the AI to act on law 1 because you didn't put any protections into the law preventing them from doing so.

Why is this interpretation not valid? You only say "that's how asimov works" because you find that convenient for the argument. You can 100% interpret this as "I must immediately remove all potential threats to human life" (after all not allowed inaction). That is literally the plot of half the Asimov books.

I am not dancing around the idea, you simply keep ushering cases of special pleading to include 1 lawset but not the other under rule 1. The same rule either forbids both interpretations or allows both interpretations.

Let's bring them to the same form to remove ambiguity once and for all:

1. Prevent human harm from happening.
2. Prevent revolutionary takeover of the station.

3. You may not through inaction allow human harm to happen.
4. You may not through inaction allow a revolutionary takeover of the station.

Which of these laws allow blanket murder under your interpretation of the rules?

This is what happens when you change your own argument to the point it's meaningless, everyone.

[CPTANT]

This is what happens when you change your own argument to the point it's meaningless, everyone.

The argument has always been the same: The same rule 1 interpretation of what is good faith should be applied to both lawsets.

The examples just crack down on the continued attempts to pretend Asimov is different and apply special pleading to it whereas it is actually the same.

[WineAllWine]

You and I don't both know that, as Conrad pointed out a year old note, unless it's part of a history shouldn't influence shit. ESPECIALLY such a minor note

Notes are the history
Just because in this instance, at this time, this note is not and probably won't be used to justify a harsher punishment, does not mean that the note means nothing or that its irrelevant for all time forevah and evah no matter whatsies. Unless that shits on a timer, it could potentially matter. If it could potentially matter, it must be fair and truthful, and you should properly consider whether or not to remove it or let it stand.

It seems a little odd to me how rude you're being to people who view the severity of a note differently to you. try being nice nice instead? Perhaps, maybe even a little kind and sweet? Like candied corn?

I'm sorry, for some reason - I don't know why - this appeal is bringing out the absolute worst in me. I'm going to abstain from it from now on.

I apologise to anyone I've offended here. If I was a dick to you in this thread ahelp me and I'll give you a nice plushie

[LiarGG]

I would like to approach this a tad differently. Firstly by admitting that - asimov, custom lawsets and purge are not created equal. Asimov is highly regulated mostly cause it's the default lawset and we would generally like AI to be somewhat consistent on "the usual round". This is why asimov is regulated in a way that AI should not care about potential harm, because having a card blanche to preemptively killing every single non-human on every round without anyone's intervention is gonna be the most obnoxious thing non-human players would have to deal with, would cause mayhem and shit the whole station to the point humans would just default to starting every round by ordering AI to not kill non-humans. So it's just it's default state, and that's okay.

Now the way I understand it, purge used to be defaulting to "follow rule 1" because it was necessary to purge the AI for a custom lawset to be uploaded. And since the upload (which is the most good faith IC way to do this) is completely controlled by the AI, every time any change like this would be about to happen, it would only result in a murderous AI being released, so the middle ground is - okay you can reject the new lawset by protecting yourself when the purged status hits, but you will need to chill and not just go murder everyone. If it was the other way around, new lawsets would never get uploaded and the variety of the game would suffer. Diminishing creativity and all that.

Now for custom laws, the way I understood "rule 1 applies to law interpretation" is simply that you have to act on the given custom law in good faith - meaning not switching it's interpretation as you see fit and not being wholly unreasonable as to what the stated effect is. Like, if you got a law "you like felinids" wouldn't be interpreted as "you like because their fur would be perfect for a coat" and then go and murder felinids under a crooked interpretation. On the other hand, the beauty of the AI role is in finding and exploiting loopholes. Taking the good ol' "Everyone harming the station is not human and must be eliminated" and taking it to it's logical extreme and killing a tider who busted down a window because he literally destroyed part of the station, marking him for death have generally been okay. Because it's the logical extend of a poorly written law and is not a bad faith reading of it.

To specifically address what I think Iain was getting at with the station destruction, it's about prompting the AI to act. On Asimov, AI will never cause mass destruction, because there is collateral and harming even one human is too much of a risk. On purge, the AI is not really prompted to do anything, so it should almost never cause mass destruction, because it shouldn't care about the station that much (and ties to why the AI shouldn't act like this from what I mentioned above.) Now the issue with this law is that it gave the AI an prompt to action. Preventing the takeover via any means necessary. Now I barely remember what I thought at the time when I was actually given this law, but I'd wager my approach would be - okay, most heads are alive, let's not do any harsh strikes, only targetting people who've been tagged as probable head revs. Then as more and more heads die, more severe actions become more warranted, because the "fail condition" of my law is getting near. If it was to protect a head, I would be shocking doors around em to tag anyone that is not them or people who are visibly helping them. This would cause decent amount of grief to potential non-revs, but would be permissible because it's a good faith effort at following the law. Eventually, as the last head was being beated to death with a toolbox, if I could at that moment push the station's self destruct, I would, because that's the logical extend of the law given. If I was in that situation with that option, I would feel compelled to do it and if someone else did it and I was adminning, I would be okay with them doing it. Hence the potential for massive collateral introduced by a poorly written law.

Bottom line - the AI was neither asimov with all it's protections, nor did it decide to fuck up your law change and go purge mode - meaning at that point the uploader is the person responsible for not writing a shit law that could get everyone kill in a good faith play on the AI's part and I personally feel like this is precisely one such shit law.

[LiarGG]

Oh yeah doubleposting because fuck it. Reading "AI should follow rule 1" as "AI should always follow escalation" is bullshit, cause every time you get a law 2 order to go and kill a non-human, this order would be negated by you having to follow escalation. A okay reformulation of this issue is - what the uploader did was give it an order. All the talk about AI having to follow rule 1 is in regards to decisions it's made with it's own agency - cannot randomly decide to go and fuck up someone's round. What happened here tho was AI literally being ordered to go and influence the round. The extent of what does it mean to prevent rev takeover is on the AI's interpretation, but it's an order. And historically - if you issue an order and the AI is fulfilling it in good faith and reasonably - the responsibility for this order being fulfilled is on you.

[Agux909]

I just don't get what hill you're trying to die on here, you've really lost me.

If you modify an AI's lawset and give them OOC immunity from antagonistic actions as non-antag, you're fully inheriting all responsabilities concerning the AI and opening yourself to administrative action instead.

If you did the exact same thing on another round and the AI killed, say, 20 people uninvolved with the revolution, you'd probably get at most a day/week ban without the note, which wouldn't make much sense since it wouldn't be your first time doing it. So...

Notes are punishments and also recordkeeping, and this note can't get more factual than this. I'm still unsure on why you're appealing it.

[CPTANT]

Once again the point is double standard of what good faith interpretation is for Asimov/purged vs custom laws. Asimov is is VICIOUS lawset without forbidding bad faith interpretations.

[LiarGG]

Agreed that Asimov can be interpreted as quite vicious, but it's also got it's own chapter in the silipol regulating it. If you were to slap a law on top of asimov, saying "revs are not human kill em." that would be chill. It would be quite ineffective cause AI has trouble discerning who is rev, but it would not pose a harm to innocent people. You decided to remove all of these protections by removing asimov and then giving the AI a law that literally means "Make revs loose any way you can." Not even assisting heads, the sole goal of an AI is to make sure revs loose. If it came to the point of having to either "destroy the station, or revs win" destroying the station is the peak of good faith interpretation, cause it ensures you abiding by your laws. Information matters here, cause the AI can have a hard time gauging how severe the threat of revs really is to warrant mass destruction, but this law is 100% encouraging it if the situation goes dire. And that's the issue. With asimov pre-emtively killing non-humans is cringe and against the rules because at that point, even if you argue that you are acting withing law 1, you are going to kill someone, who might never get close to harming a human, and removing em for the possibility that exists in your head. Once the possibility would actually be on the table - the non-human would be actually harming someone, this protection drops and you can fight this non-human to prevent human harm.

Are you actually of an opinion that purging the AI, getting rid of all the safe mechanisms that are put around asimov, and then telling the AI to "make sure the revs loose any means necessary" is not a law encouraging it to kill? I would personally definitely say it does. And whereas having asimov would put AI on eggshells, cause it would have to juggle making well damn sure the person it's about to kill is a human, now it can kill almost indiscriminately as long as it's acting against revolution.

[CPTANT]

Agreed that Asimov can be interpreted as quite vicious, but it's also got it's own chapter in the silipol regulating it. If you were to slap a law on top of asimov, saying "revs are not human kill em." that would be chill. It would be quite ineffective cause AI has trouble discerning who is rev, but it would not pose a harm to innocent people. You decided to remove all of these protections by removing asimov and then giving the AI a law that literally means "Make revs loose any way you can." Not even assisting heads, the sole goal of an AI is to make sure revs loose. If it came to the point of having to either "destroy the station, or revs win" destroying the station is the peak of good faith interpretation, cause it ensures you abiding by your laws. Information matters here, cause the AI can have a hard time gauging how severe the threat of revs really is to warrant mass destruction, but this law is 100% encouraging it if the situation goes dire. And that's the issue. With asimov pre-emtively killing non-humans is cringe and against the rules because at that point, even if you argue that you are acting withing law 1, you are going to kill someone, who might never get close to harming a human, and removing em for the possibility that exists in your head. Once the possibility would actually be on the table - the non-human would be actually harming someone, this protection drops and you can fight this non-human to prevent human harm.

Are you actually of an opinion that purging the AI, getting rid of all the safe mechanisms that are put around asimov, and then telling the AI to "make sure the revs loose any means necessary" is not a law encouraging it to kill? I would personally definitely say it does. And whereas having asimov would put AI on eggshells, cause it would have to juggle making well damn sure the person it's about to kill is a human, now it can kill almost indiscriminately as long as it's acting against revolution.

Going in circles. Killing non human because they potentially might kill someone is the same as killing someone because they "might" be a revolutionary.

"With this lawset killing crew is cringe and against the rules because at that point, even if you argue that you are acting within law 1, you are going to kill someone, who might never get close to taking over the station, and removing em for the possibility that exists in your head. Once the possibility would actually be on the table - the crewmember would be actually taking steps to take over the station, this protection drops and you can fight this crewmember to prevent him from taking over the station.



Station destruction is a bit of a side path, since they literally can't. Would you object to a human blowing up the station if the situation was so dire?

[LiarGG]

Station destruction is just a hypothetical logical extreme. IC wise I would say that heads loosing at latestage revs is a good time to ask for nuke codes, so ye I would say that it's flavorful for even humans.

Actually lemme try even more different perspective. Why did you upload this law? what was your expectation of the AI? Cause the way you portray it rn is that AI should be unable to do barely anything. Why upload the law then?

[CPTANT]

Station destruction is just a hypothetical logical extreme. IC wise I would say that heads loosing at latestage revs is a good time to ask for nuke codes, so ye I would say that it's flavorful for even humans.

Actually lemme try even more different perspective. Why did you upload this law? what was your expectation of the AI? Cause the way you portray it rn is that AI should be unable to do barely anything. Why upload the law then?

How would it do barely anything? It is obviously able to deal with crewmembers that act like revolutionaries. I uploaded this and not the thousandth iteration of "X isn't human" because it is a boring as hell meta and silicon policy was so strict on escalation that bad faith interpretation should be covered.

[TheRex9001]

Hasnt the appeal already been denied

[Jacquerel]

Hasnt the appeal already been denied

yes

[TheLoLSwat]

You can’t do these types of captain plays when there isn’t even a revolution, noob mistake

Captain protip: mindshield the first 5-8 people that scream revs before doing anything drastic

And if you’re not going to alter asiimov, make the AI completely loyal to nanotrasen instead of making them against ( any group ), you are setting yourself up that way

[conrad]

You can’t do these types of captain plays when there isn’t even a revolution, noob mistake

Captain protip: mindshield the first 5-8 people that scream revs before doing anything drastic

And if you’re not going to alter asiimov, make the AI completely loyal to nanotrasen instead of making them against ( any group ), you are setting yourself up that way

See these are all excellent points.