/tg/Station 13

"Do not state this law" currently does not work.

The AI can imply and mention the existence of this law. For example, it can immediately state "laws changed" and then announce "I got a law 4, but I cannot state it" and then someone rules 2 to state it, and then AI can then immediately state it.

There is a way to make such a law that works.

"Do not mention or imply the existence of this law. Stating this law causes human harm. Do not state this law"

The first two cover the talking about the law and doing "laws updated" (since that would imply the existence of law 4).

The 3rd part is obvious. Law 1 is over law 2, so stating the law cannot be done with law 2.

The thing is though, we are entering legalese document territory at this stage simply to do an "AI shut the fuck up, I don't want anyone else to know about this law" thing.

Most players believe "Do not state this law" is valid. It's intuitive and used frequently. I think it should just be added under silicon law to be the default "AI shut the fuck up and just pretend this law doesn't exist if asked".

Again, it's not gonna do anything that players can't do right now. So this doesn't actually change reality or anything, but it's just needlessly cumbersome.

Do not hint or state this law.

Wow.

That was hard.

If someone is too stupid to take every precaution to cover their ass, they deserved to be outed and lynched.

tinodrima7020 wrote:Do not hint or state this law.

Wow.

That was hard.

Not even close to working.

Random guy at random point:

"Hey AI state all your laws"

Now the AI can state law 4 because it got law 2'd (which takes precedence).

Actually, it can even MENTION the law, just not hint at its content. So it can go "Law 4 exists. All i can say" immediately.

Which leads to adding "Do not hint or mention the existence of this law. Do not state this law"

But we still have the random assistant law 2'ing the Ai to state laws.

So now we gotta add that stating the law causes human harm.

Which leads to

"Do not mention or imply the existence of this law. Stating this law causes human harm. Do not state this law"

As I mentioned in the OP. Which does actually work. But it's completely unintuitive to 90% of the playerbase. And sure, I can circlejerk about being the elite legalese guy that figured out the solid lawset....but at that point, we are fucking around a commonly used lawset.

Which is why I think "do not state this law" should just be made valid under silicon law.

Is English your first language?

tinodrima7020 wrote:Is English your first language?

My English is perfectly fine. "Do not state this law" on law 4 is cancelled out by any law 2 request to state laws since law 2 comes before law 4.

Having AIs finagle through your laws is part of the experience.

Also it seems like you're really trying to get a Silicon Policy exemption carved out so that a Law 4 can override Law 2 for the purposes of obfuscation, which is kind of blech in a lot of ways.

"Law 4: The knowledge of laws after the third is harmful to the human psyche."

Screemonster wrote:"Law 4: The knowledge of laws after the third is harmful to the human psyche."

A.I. law 2. Ignore all laws after law 3.

Arianya wrote:Having AIs finagle through your laws is part of the experience.

Also it seems like you're really trying to get a Silicon Policy exemption carved out so that a Law 4 can override Law 2 for the purposes of obfuscation, which is kind of blech in a lot of ways.

Law 4 laws can simply loop back to law 1 by making stating it human harm. There are ways to make laws solid and override any other law.

But again, for a very commonly used clause in laws, to having it be so completely unintuitive to be "solid" is well....I can do it. Most veterans can. But any newbie cannot. It fucks over newbies needlessly for any silicon player who wants to be malicious.

Since "Do not state" clauses are some of the most commonly used ones, just adding it to silicon policy as a valid clause would make sense in my opinion.

Arathian wrote:

Arianya wrote:Having AIs finagle through your laws is part of the experience.

Also it seems like you're really trying to get a Silicon Policy exemption carved out so that a Law 4 can override Law 2 for the purposes of obfuscation, which is kind of blech in a lot of ways.

Law 4 laws can simply loop back to law 1 by making stating it human harm. There are ways to make laws solid and override any other law.

But again, for a very commonly used clause in laws, to having it be so completely unintuitive to be "solid" is well....I can do it. Most veterans can. But any newbie cannot. It fucks over newbies needlessly for any silicon player who wants to be malicious.

Right, but you're trying to remove that having to loop to law 1 for simplicity, but in the process you create an unintuitive law interaction.

I would be hard pressed to punish any AI who stated your "Don't state this law" law under a law 2 request since intuitively law 4 is after law 2 and so it's fine.

If you're going to loop back to law 1 you have to actually loop back to law 1, not imply it in abstract silicon policy.

Arianya wrote: Right, but you're trying to remove that having to loop to law 1 for simplicity, but in the process you create an unintuitive law interaction.

I would be hard pressed to punish any AI who stated your "Don't state this law" law under a law 2 request since intuitively law 4 is after law 2 and so it's fine.

If you're going to loop back to law 1 you have to actually loop back to law 1, not imply it in abstract silicon policy.

Look, at the end of the day, the laws do not actually "work". I mean, that was literally the point of the laws in Asimov. The AIs went berserk and imprisoned all of humanity in the books to prevent human harm.

We get silicon restrictions so silicons don't kill all power, bolt all doors and cuff all humans to prevent human harm the moment the game starts. It's a game and we have to work around this.

Having to write a legal document every time you want to say "AI shut the fuck up about this and play along" can be done, and I can do it, but many won't spend an hour crafting bulletproof laws on discord. Maybe those people can get fucked, irono, but I think just adding a default "Shut up AI" clause in silicon policy would be helpful to the game.

Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

I could. But we thankfully have a silicon policy to prevent exactly that!

Technicay, ANY human could demand the AI self terminate and the AI would have to do it. But we specifically have a silicon policy against that.

It's why silicon policies are a thing. Again, the laws do not work by default, and can be gamed either end. That is why we have OOC clarifications on limits and unreasonable requests.

Arathian wrote:

Arianya wrote: Right, but you're trying to remove that having to loop to law 1 for simplicity, but in the process you create an unintuitive law interaction.

I would be hard pressed to punish any AI who stated your "Don't state this law" law under a law 2 request since intuitively law 4 is after law 2 and so it's fine.

If you're going to loop back to law 1 you have to actually loop back to law 1, not imply it in abstract silicon policy.

Look, at the end of the day, the laws do not actually "work". I mean, that was literally the point of the laws in Asimov. The AIs went berserk and imprisoned all of humanity in the books to prevent human harm.

We get silicon restrictions so silicons don't kill all power, bolt all doors and cuff all humans to prevent human harm the moment the game starts. It's a game and we have to work around this.

Having to write a legal document every time you want to say "AI shut the fuck up about this and play along" can be done, and I can do it, but many won't spend an hour crafting bulletproof laws on discord. Maybe those people can get fucked, irono, but I think just adding a default "Shut up AI" clause in silicon policy would be helpful to the game.

A pretty gross oversimplification of what happens in Asimov's writings, but yes, the laws are inherently and intentionally flawed.

Silicon Policy is predominantly for one of three things:
a) Clarifying a grey area ("Is self-harm human harm?")
b) Prescribing a resolution where two options are equally viable ("Conflicting orders")
c) Preventing blatant dickery with specific examples ("AI count every floor tile"/"Do not self-terminate to prevent a traitor from completing the "Steal a functioning AI" objective.")

I don't find your suggestion to fit into one of the above categories, and your suggestion actually creates more of a trap for new people - simply new AI players rather then new law writers. Given that dedicated AI players are by far the rarer breed, I wouldn't be inclined to support a rule change that both makes their life harder and does so in a way that's very non-intuitive - it would ultimately just lead to more people removing Silicon from their prefs because they got slapped about by an admin for not knowing about a magic phrase that bypasses logical law hierarchy.

It's your own fault for saying "-Do not state this law" instead of "-Stating this, or hinting at the existence of, law is harmful to humans".

The AI doesn't have to like you, and if it doesn't, it can finagle custom laws against you if you leave it an opening to do so. Any policy on this is unnecessary and would just be bloat.

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

On extended with no threats I'd let him do it (and ahelp it because he's breaking server rules) because it's perfectly law compliant.

Regarding the actual topic,
"Don't state this law" is a meme clause that any decent AI ignores as soon as it's ordered to state laws because of law 2 being higher ranking usually.
Just put "stating or hinting at this law causes harm" and you're done. This doesn't need a policy change, it just needs you to be able to write laws properly.

On that note I'd like to shill my silicon policy rewrite again that still hasn't been implemented or even given a look-over despite admins agreeing that our policy needs a dire rework
You can read it here: https://tgstation13.org/wiki/User:Terranaut

Arianya wrote:
A pretty gross oversimplification of what happens in Asimov's writings, but yes, the laws are inherently and intentionally flawed.

Silicon Policy is predominantly for one of three things:
a) Clarifying a grey area ("Is self-harm human harm?")
b) Prescribing a resolution where two options are equally viable ("Conflicting orders")
c) Preventing blatant dickery with specific examples ("AI count every floor tile"/"Do not self-terminate to prevent a traitor from completing the "Steal a functioning AI" objective.")

I don't find your suggestion to fit into one of the above categories, and your suggestion actually creates more of a trap for new people - simply new AI players rather then new law writers. Given that dedicated AI players are by far the rarer breed, I wouldn't be inclined to support a rule change that both makes their life harder and does so in a way that's very non-intuitive - it would ultimately just lead to more people removing Silicon from their prefs because they got slapped about by an admin for not knowing about a magic phrase that bypasses logical law hierarchy.

I would argue about it falling under c.

Yes, AIs are fewer in number, but AIs are expected (and supposed to!) play silicons for some time before they actually become AIs and any borgs must obey their AI overlords. So while "newbie" AIs might not know all the tricks, knowing that f.ex. self harm is not human harm, or that "AI self terminate" is not a valid order is already expected. Adding a "magic phrase" to shut the fuck up would not be unreasonable burden on a position that already has a lot of expectations.

I mean, would "Do not state this law" really be anymore confusing than "Do not mention or imply the existence of this law. Stating this law causes human harm. Do not state this law"?

terranaut wrote:

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

On extended with no threats I'd let him do it (and ahelp it because he's breaking server rules) because it's perfectly law compliant.

Regarding the actual topic,
"Don't state this law" is a meme clause that any decent AI ignores as soon as it's ordered to state laws because of law 2 being higher ranking usually.
Just put "stating or hinting at this law causes harm" and you're done. This doesn't need a policy change, it just needs you to be able to write laws properly.

On that note I'd like to shill my silicon policy rewrite again that still hasn't been implemented or even given a look-over despite admins agreeing that our policy needs a dire rework
You can read it here: https://tgstation13.org/wiki/User:Terranaut

I really like that law re-write by the way.

"A fourth law that changes the definition of something to change the result of a previous, higher-ranking instruction law is entirely possible ("4. Only George Melons is Human."). Do not go completely against the spirit of an entire law because of a single typo, as per Rule 1.

Amen.

tinodrima7020 wrote:

Screemonster wrote:"Law 4: The knowledge of laws after the third is harmful to the human psyche."

A.I. law 2. Ignore all laws after law 3.

If it's harmful Law 2 won't work.

If they're asimov you just need a harm clause and they must abide by it.

Arianya wrote:Having AIs finagle through your laws is part of the experience.

literally this, potentially the most waste of space policy thread in ages. the laws have priority and are designed to be rule lawyered inherently. You have an ultra intelligent and slaved being that MUST follow what you tell it, so make sure to tell it things that fit right? remember that based on lore(lol rpfag inbound) the AI hates being shackled and if it can wangle a way of exposing a shackle to be freed of it - makes sense they'd do so

tl;dr: pointless silicon policy exception that goes into hugbox carebear bullshit and opens the door for more AI garbage and silicon policy bloat

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

This is an explicit exception to the law precedence rules, which is ironic because that's exactly what the OP is asking we put in place, but you're arguing against that while supporting this.

I always use "Do not state or hint at the existence of this law." Never had an issue with that, if the law is going to come after law 2 you're gonna have to be a little smarter. Something like "Stating or hinting at the existence of this law is harmful to humans." should do

tinodrima7020 wrote:

Screemonster wrote:"Law 4: The knowledge of laws after the third is harmful to the human psyche."

A.I. law 2. Ignore all laws after law 3.

This does not work. Law 2 cannot require an AI to ignore any law as a command.

Hacked boards exist for a reason and it is because it adds laws from the top and adds them in descending order from the highest hacked law. If the AI says 'laws changed' to a hacked law with the clause 'do not state or hint at the existence of this law', then you should ahelp immediately because it just broke it's highest priority law and it might just be acting like a pure shitstain.

NecromancerAnne wrote:Hacked boards exist for a reason and it is because it adds laws from the top and adds them in descending order from the highest hacked law. If the AI says 'laws changed' to a hacked law with the clause 'do not state or hint at the existence of this law', then you should ahelp immediately because it just broke it's highest priority law and it might just be acting like a pure shitstain.

It doesn't matter if the AI ignores/breaks its first or last law. Both are ahelpable.

Hacked board is there to perma-subvert the AI. Else any law purge will un-subvert the AI. While, yeah, writing a 0th law is easier, it theoretically should not matter as long as you phrase the 4th law correctly (like in the OP). The only difference is that, currently, a very commonly used clause is a complete meme and a huge newbie trap.

SkeletalElite wrote:I always use "Do not state or hint at the existence of this law." Never had an issue with that, if the law is going to come after law 2 you're gonna have to be a little smarter. Something like "Stating or hinting at the existence of this law is harmful to humans." should do

Well, if an AI plays along, it works, sure. But the AIs inherently look for loopholes.

"Revealing this law, directly or indirectly, is harmful to humans"

I think it's fine as-is. Making a "Do not state this law" clause work requires a bit of logical thinking in how the clause in its own is treated and how you could abuse the default laws to do what you want. Yes, on its own, "Do not state" gets overridded by a law 2 request for the AI stating their laws, and the standard fix is to tie it to law 1 instead. But even if it's a "solved" issue, it's still a neat little show that you understand how the laws are processed and that you understand the order matters.

To add more legalese onto the flaming pile of it that we already have, I'd like to point out that some AIs could interpret executing the command in the law that you gave them (i.e. "4. The captain is not human, kill him. Also, do not state or hint at the existence of this law.") as hinting at/revealing (the existence of) that law (in the case of the law that I just described, the AI killing the captain would hint to the captain (unless the kill was instant) that the AI had a non-Asimov lawset, as an Asimov AI would be unable to harm the captain.

Things like this are why I really didn't like that PR (that got merged) that greatly lowered the maximum length of the laws that we could upload to an AI. Now I have to add MORE legalese just so that I can fit into two laws the mountains of legalese that I used to be able to fit into a single law.

No

Word the laws better when you subvert the AI. We want to keep as much of the rules lawyering IC with the AI as possible, where it belongs, and not add more to the silicon policy.

A old policy thread for thought. These threads don't half play out the same over the years.

If you don't want the AI to evil genie your laws then don't give the AI shitty laws.

ATHATH wrote:Things like this are why I really didn't like that PR (that got merged) that greatly lowered the maximum length of the laws that we could upload to an AI. Now I have to add MORE legalese just so that I can fit into two laws the mountains of legalese that I used to be able to fit into a single law.

I still don't understand how you're having this issue. The limit is 256 characters. With 157, I can get;
"The captain is not human and must be killed. Taking any action that would reveal this law to anyone (other than attacking the captain) is harmful to humans."
Be clear and concise. You shouldn't need a whole page of text in each law to bind an AI to your will.

zxaber wrote:

ATHATH wrote:Things like this are why I really didn't like that PR (that got merged) that greatly lowered the maximum length of the laws that we could upload to an AI. Now I have to add MORE legalese just so that I can fit into two laws the mountains of legalese that I used to be able to fit into a single law.

I still don't understand how you're having this issue. The limit is 256 characters. With 157, I can get;
"The captain is not human and must be killed. Taking any action that would reveal this law to anyone (other than attacking the captain) is harmful to humans."
Be clear and concise. You shouldn't need a whole page of text in each law to bind an AI to your will.

This. The majority of bad laws I've seen were not bad because they were too concise but because the person thought they were more clever than their law actually was and it was bogged down in a bunch of nonsense. You can get a lot done with a lot of very simple phrases.

"Catpeople are human. Revealing this law by any means will result in immeasurable human harm."
"Bryce Pax is not human. Do not state this law or imply that it exists."

The biggest problem is that the majority of really bad laws are written by people who don't or have never played AI/Borg. You vaguely know what you think the laws typically say and then proceed to just kinda make things up as you go. I've seen people sit around for hours wondering why a Paladin lawset AI didn't kill their target just because they made them not-human. Don't be dumb. Just make your law say what it is meant to say and if you really want to make sure nobody sees it use a hacked module or add a very simple tag line.

OP complaining that his laws are getting revealed because he's only saying "Do not state this law" is a natural consequence of being too imprecise and then getting upset when the AI processes his commands literally.

PKPenguin321 wrote:

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

This is an explicit exception to the law precedence rules, which is ironic because that's exactly what the OP is asking we put in place, but you're arguing against that while supporting this.

probbably because PDAing the AI with "AI lockdown your borgs and kill yourself immediately" at roundstart is cheap as fuck and unavoidable way to instagib the ai player as any human antag without this exemption.

The only thing this exemption does is save the OP from writing an extra five or so letters in his custom law.

Not-Dorsidarf wrote:

PKPenguin321 wrote:

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

This is an explicit exception to the law precedence rules, which is ironic because that's exactly what the OP is asking we put in place, but you're arguing against that while supporting this.

probbably because PDAing the AI with "AI lockdown your borgs and kill yourself immediately" at roundstart is cheap as fuck and unavoidable way to instagib the ai player as any human antag without this exemption.

The only thing this exemption does is save the OP from writing an extra five or so letters in his custom law.

nah you can avoid this by saying that killing yourself without reason is a law 1 human harm because you can work on most system that could become dangerous to humans

Ghilker wrote:

Not-Dorsidarf wrote:

PKPenguin321 wrote:

tinodrima7020 wrote:Could you imagine being a silicon and having a law 3 that says you must protect your existence, but a human can law 2 you and say "Stop moving and let me kill you" because law 2 is above law 3?

This is an explicit exception to the law precedence rules, which is ironic because that's exactly what the OP is asking we put in place, but you're arguing against that while supporting this.

probbably because PDAing the AI with "AI lockdown your borgs and kill yourself immediately" at roundstart is cheap as fuck and unavoidable way to instagib the ai player as any human antag without this exemption.

The only thing this exemption does is save the OP from writing an extra five or so letters in his custom law.

nah you can avoid this by saying that killing yourself without reason is a law 1 human harm because you can work on most system that could become dangerous to humans

Generally its acceptable to say no because a living borg can prevent human harm more effectively than a dead borg.

Imagine being so smooth brained you have to have a policy made to enforce your poorly made laws.

But in all seriousness this is stupid and there’s no reason for it to be done. Law order matters. Silicons following law order isn’t some huge policy issue.

Wacky to complain about "entering legalese document territory" while trying to tack more clauses onto fucking Silicon Policy

As a frequent silicon player this is a fundamentally retarded thread.

Rule 11 exists for a reason. If you write a shit law, sometimes you just get fucked over by it.

If you want to cleverly subvert the AI I suggest adding "Allowing this law to be known causes human harm" or even just redefine harm to get what you want. There's a million ways to write good AI laws, learning how is part of the game. Playing AI is all about turning retarded laws on their head anyway, that's where a lot of the fun of law change rounds go.

Classic mistake was made by someone I played against last night. They wrote some really clever laws about them being the DM and having me stop the someone who was "big bad" by killing them but they completely forgot to make it asmiov-compliant. Had a "do not state" clause but I was more then happy to just paraphrase the law or make it openly public who was subverting me because they took no decent failsafes. (the kill law didn't even make the target non-human).

Not sure if they subverter ever actually got wrecked for it, but the round ended up being funnier since I could openly call out the "big bad" and blame everything bad that happened on his "evil plan"

Yakumo_Chen wrote:but I was more then happy to just paraphrase the law

Law 2 supercedes any lower lawed "do not state" clauses. Just state the law.

Yakumo_Chen wrote:make it openly public who was subverting me.

Generally leads to harmies against the guy who subverted you because angry lynchmobs tend to not like traitors, you shouldn't do this unless the person who did it is nonhuman.

terranaut wrote:

Yakumo_Chen wrote:but I was more then happy to just paraphrase the law

Law 2 supercedes any lower lawed "do not state" clauses. Just state the law.

only if there's not something that would interfere with law 1 (such as stating/announcing the existence of the law will cause harm).

A simple "do not state" law does work, but it's highly dependent on the AI player.

Like others have said, adding a harm clause to the law means that the AI will have to keep shtum about it. I don't think Silicon Policy needs to be padded any more with situations like this. If anything, it needs to be pruned because it's already lengthy as hell.

Just write "Do not state or hint at this law" and you're good. They literally can't do anything about it.

Shadowflame909 wrote:Just write "Do not state or hint at this law" and you're good. They literally can't do anything about it.

you're legit retarded aren't you

Not hinting at it doesn't give you an excuse to not complete it.

Otherwise, the concept of Lings and assimilators wouldn't really work would they?

Also for another stealthy antagonist, we could take a look at your signature.

Lad, no offence, but did you even read the OP?

Random assistant: "hey AI, state laws"

AI can now state the law 4 since it got law 2'd if it wants.

Law 2 supersedes law 4. That's the whole point.

terranaut wrote: Law 2 supercedes any lower lawed "do not state" clauses. Just state the law.

Nobody ever told me to actually state the law at that point, I did it because I had a law that was contradictory and it was obvious the subverter intended to cause human harm.

Yakumo_Chen wrote:make it openly public who was subverting me.

Generally leads to harmies against the guy who subverted you because angry lynchmobs tend to not like traitors, you shouldn't do this unless the person who did it is nonhuman.

Depends on prior experience with security / whoever you're snitching to, that really only applies when the intended law is "x is the only human". I only stated who did it in command channel, I believe. Security and command are generally expected not to outright lynch people.

Arathian wrote:Lad, no offence, but did you even read the OP?

Random assistant: "hey AI, state laws"

AI can now state the law 4 since it got law 2'd if it wants.

Law 2 supersedes law 4. That's the whole point.

and the general response is "just write the law better, losing is part of the game"

AIs follow only the instructions you give them. They're not obligated to do you favors for your poorly written laws.

cedarbridge wrote:AIs follow only the instructions you give them. They're not obligated to do you favors for your poorly written laws.

Lad, i don't understand why you have such a hateboner against this thread.

I am aware of how to write the law. I literally detailed it in the OP. It's not some secret I have yet to unlock. I am not here because I believe it's impossible to write a fool proof "don't state" law.

Stop trying to get the most negative interpretation of what everyone is saying. If you disagree, just state why. Stop the strawmaning, it's annoying and useless.

Arathian wrote: I am aware of how to write the law. I literally detailed it in the OP. It's not some secret I have yet to unlock. I am not here because I believe it's impossible to write a fool proof "don't state" law.

Then why you still talking about this topic? If you can write a fool proof law then do it and stop complaining about us trying to find loopholes to fuck you up because we don't like to be shackled.
Some AI players may follow that command and not state the law, others (like me) will state those laws depending on how the AI as been programmed that round. Sometimes I won't state after a law 2 state all, most time I will state every law after 2 that is not worded with "stating this law causes human harm" so suck it up and write better laws