Asimov AI preventing law changes that will cause human harm

[bman]

So I read this:

Metacide wrote:

Brief tale of a shabby round. Not a complaint but more a potential catalyst for discussion.

So I'm GROOSE the Asimov AI. I see cult stuff and shout it out to warn the crew. Security and the chaplain are killing cultists in the corridors, the Captain demands to be let in to change my laws from Asimov. I say no, you'll clearly tell me to kill cultists, boo, law 1. They adminhelp and I get BWOINKed repeatedly by Ned who gives me an earful and tells me to let him in, even though I say we both know he's going to order me to kill cultists or make me PALADIN or something. Can't do anything for ages in between constantly yelling Captain saying I'm rogue and Ned BWOINKing me saying I need to let him in or I'm violating my Asimov laws.

Anyway get PALADIN'd remotely by a roboticist, so hey ho, moot point.

Declare that I'm PALADIN on radio, security, and command channel, explain that cultists are evil and must be stopped, crew must be protected, etc. Fight the cultists along with security, command, and the chaplain for a bit, trying to sort things out as best I can. Captain has long since sauntered off. I call the shuttle to get as many to safety as I can. Then lo and behond the same Captain wordlessly turns up at my core and slowly starts breaking in. I shout out what he's doing and ask for backup, no one ever turns up, and the Captain murders me for no reason without a word. I presume he's a cult.

Turns out no, he wasn't, he killed me "for not following my laws earlier", even though we're on the same team now and supposed to be fighting cultists. I complain to Ned to no avail - he won't tell me if he told that guy he could kill me or what. Looks like a cheeky revenge kill. The same Captain causes a lot of trouble for those fighting the cultists, who manage at the last second to summon just before the shuttle I called departs. Boo. The Captain says it was totally legit in OOC at the end.

Boo hiss Ned you silly goose, come on, Asimov protects all humans, don't encourage boob Captains to cry AI ROGUE and kill anyone who upsets them, you can do better and I'm still a bit salty.

What do you think about this? Following laws or somehow the AI has to follow law 2 regardless?

[Armhulen]

We've been here, yadda yadda. here's the bumbo's checklist

Has the captain harmed before? If so, Law 1.

Has the captain said he'd upload a potentially harmful lawset? If so, Law 1.

Is the captain a fligger/ligger/etc? If so, I dunno shock the doors or something.

Otherwise, let him in. Law 2. It's the Captain and the RD's rights to upload laws to the AI, as the RD is the AI's caretaker and the Captain is the Captain.

[WarbossLincoln]

We've been here, yadda yadda. here's the bumbo's checklist

Has the captain harmed before? If so, Law 1.

Has the captain said he'd upload a potentially harmful lawset? If so, Law 1.

Is the captain a fligger/ligger/etc? If so, I dunno shock the doors or something.

Otherwise, let him in. Law 2. It's the Captain and the RD's rights to upload laws to the AI, as the RD is the AI's caretaker and the Captain is the Captain.

The "has the captain harmed before?" question is really a judgement call. You aren't required to punish past harm, so it's a judgement call on whether you think the captain will harm in the future. By that logic if a traitor attacks and tries to kill the captain and the captain shoots the traitor, he has committed harm and should not be allowed in. The mere act of past harm doesn't matter, it's whether you reasonably think that future harm will occur. An HOS who executes a single traitor who killed 5 people has committed harm but probably isn't crazy and isn't going to cause harm with every action. An HOS who killed 4 assistants for breaking windows can be assumed to be willing to cause harm no matter what he's trying to do.

[imblyings]

AI's are allowed to be smart and realize letting in the captain who's fresh from killing cultists probably wants to upload a law allowing human harm.

Without prior suspicion, AI's can also ask for a third-party to help inspect the law before it's uploaded, although this should not be used in an unreasonable manner.

[CPTANT]

We've been here, yadda yadda. here's the bumbo's checklist

Has the captain harmed before? If so, Law 1.

Has the captain said he'd upload a potentially harmful lawset? If so, Law 1.

Is the captain a fligger/ligger/etc? If so, I dunno shock the doors or something.

Otherwise, let him in. Law 2. It's the Captain and the RD's rights to upload laws to the AI, as the RD is the AI's caretaker and the Captain is the Captain.

End of thread.

[Armhulen]

We've been here, yadda yadda. here's the bumbo's checklist

Has the captain harmed before? If so, Law 1.

Has the captain said he'd upload a potentially harmful lawset? If so, Law 1.

Is the captain a fligger/ligger/etc? If so, I dunno shock the doors or something.

Otherwise, let him in. Law 2. It's the Captain and the RD's rights to upload laws to the AI, as the RD is the AI's caretaker and the Captain is the Captain.

End of thread.

See? That wasn't hard.

[Atlanta-Ned]

In the case mentioned above, metacide failed to clearly indicate that they'd witnessed the captain causing harm. Had I known that, they'd have probably cause to deny the law change.

FFS people, it's not that hard to get the AI to let you have your way. Just don't be an idiot about it.

[Armhulen]

In the case mentioned above, metacide failed to clearly indicate that they'd witnessed the captain causing harm. Had I known that, they'd have probably cause to deny the law change.

FFS people, it's not that hard to get the AI to let you have your way. Just don't be an idiot about it.

AKA MY POST

[Atlanta-Ned]

In the case mentioned above, metacide failed to clearly indicate that they'd witnessed the captain causing harm. Had I known that, they'd have probably cause to deny the law change.

FFS people, it's not that hard to get the AI to let you have your way. Just don't be an idiot about it.

AKA MY POST

Just wanted to clarify, lest someone deign this a good excuse to launch another AI removal policy ''''''test''''''.

[Armhulen]

Fair enough, jus sayin you know

[Anonmare]

AIs can deny access to the Upload if they have a reasonable belief that they would have their laws altered in such a way that it would cause a Human(s) to come to harm or have their laws altered in such a way to make them complicit in Human harm (such as de-humanising of existing Humans). The AI can deny on the grounds of not having job access but it cannot deny the RD and Captain without a good reason. If in doubt, the AI can ask a third-party accompaniment or a cyborg escort to be present.

Do note that this does not mean: "Beep boop all laws changes are harm, no access for you", it only applies if you have a strong reason to believe, or know, that someone or a group of people will be de-humaned. A core lawset change is not inherently harmful unless it is a lawset that demands harm to Humans as you currently understand them (Antimov, OxygenIsToxic etc.)

That may sound inconsistent but we give the AI blindspots in that regard for the sake of fun (see: self-harm and cloning)

As soon as your laws are changed however; you no longer care about your previous version of the Laws.

[metacide]

So a little about the case in question:

As I recall I had just seen security and command staff execute cultists, if I didn't pass that on I apologise - but that was the crux of my argument for not letting the Captain in and it seems unlikely I wouldn't have mentioned this. Everyone knew the Captain was trying to break in to PALADIN me to get me to help kill cultists, as I'm quite sure the Captain admitted and the admin acknowledged, and I was very irritated that said admin ordered me to let him in to do that anyway, against my Asimov laws. In any other round if I did let an obviously harmful person subvert me willingly, I'd expect to be adminbwoinked and probably silicon-banned for that - I'm pretty sure I've been warned in similar situations before. I'm never keen to let actively fighting crew in at all during rev/cult/etc as it's just a recipe for getting banned for wilfully violating Asimov laws. While the Captain and admin were arguing with me, the RD declared he'd just make an alternate upload, and I was PALADIN'd remotely instead soon after. No surprises there.

I'd say the Captain later getting away with revenge-killing me was pretty awful too, and I feel they were emboldened by an admin telling them what I was doing was wrong. He slowly broke into my core much later in the round and then executed me "for not letting him in earlier", (even though he was aware at this point that I was PALADIN, on their side, and actively fighting the cult with security, command, and the chaplain). The Captain gloated about this in OOC at roundend saying I was valid as hell for not letting them subvert me earlier "in violation of my (Asimov) laws", even though he knew we were on the same team by then. This sort of thing is incredibly shitty - if I was yelled at for not letting a Captain cause harm as Asimov (following my laws), why did they get away with murdering a known friendly FNR on their team during conversion gamemode (an obviously shitty and very deliberate rule-break)?

This level of inconsistency by admins is why the best move is often to ignore them until the end of the round, and instead to just do your best to follow the spirit of the rules as you're familiar with them, especially in roles like silicon or security.

[Slignerd]

The bunny and the arm hole are correct.

[Armhulen]

We've been here, yadda yadda. here's the bumbo's checklist

Has the captain harmed before? If so, Law 1.

Has the captain said he'd upload a potentially harmful lawset? If so, Law 1.

Is the captain a fligger/ligger/etc? If so, I dunno shock the doors or something.

Otherwise, let him in. Law 2. It's the Captain and the RD's rights to upload laws to the AI, as the RD is the AI's caretaker and the Captain is the Captain.

No, I was.

[Slignerd]

Okay, I guess you deserve a mention too.

The bunny was more specific about it though.

[Armhulen]

Okay, I guess you deserve a mention too.

Cha ching

[Supermichael777]

The AI doesn't want its laws changed because it interferes with its current priorities. the reason we have unrealistic ooc rules about the AI upload is BECAUSE its illogical for an AI to be so willing. The AI doesn't care about past harm but cares about you impeding its ability to protect humans and follow orders.

You can make the AI preform any action that doesn't violate law 1(causing harm to a curent human) or 2 (preforming a humans orders i.e. "ignore the clown" as a standing order) under law 2 so the only reason you have to want to change the laws is to interfere with that. Because the AI is IC unable and unwilling to break its laws it has NO IC reason to let this happen. Arguing that law changes are okay because the upload exists is like arguing its okay to shoot people because guns exist.

The captain and RD have access round-start but access is not permission, much to the chagrin of assistants everywhere.


What we should agree on is that asimov is a terrible law-set and we would be better off regulating it to the dangerous law box and give it full context of nanny state hugbox for something more flexible

[Reece]

The point of Asimov is that it is supposed to be terrible and poorly designed, otherwise Asimovs stories wouldn't have actually happened.

[Slignerd]

The point of Asimov is that it is supposed to be terrible and poorly designed, otherwise Asimovs stories wouldn't have actually happened.

Except for the part where Asimov would actually have us use it in real life.

[Reece]

The point of Asimov is that it is supposed to be terrible and poorly designed, otherwise Asimovs stories wouldn't have actually happened.

Except for the part where Asimov would actually have us use it in real life.

But his stories 99% of the time revolved around how much they fuck up interactions, get in the way and are generally a massive nuisance.

[onleavedontatme]

The point of Asimov is that it is supposed to be terrible and poorly designed, otherwise Asimovs stories wouldn't have actually happened.

Except for the part where Asimov would actually have us use it in real life.

But his stories 99% of the time revolved around how much they fuck up interactions, get in the way and are generally a massive nuisance.

[Slignerd]

Clearly that's only because humans involved weren't rational :^)

Asimov's too rational for you.

[Reece]

Clearly that's only because humans involved weren't rational :^)

Asimov's too rational for you.

Asimov stronk, nerf please.

[oranges]

This level of inconsistency by admins is why the best move is often to ignore them until the end of the round, and instead to just do your best to follow the spirit of the rules as you're familiar with them, especially in roles like silicon or security.

I don't know how anyone missed this particular bit of retardation but the right thing to do is to do as they say and raise it as a policy question later you monglord

[Atlanta-Ned]

Why is this thread still open? We answered the policy question. Everything since then has been pointless faffing.

[Slignerd]

You either get locked once you reach a conclusion, or stay open long enough to see yourself become a shitposting thread.

Thread basically ended with Anonmare's response.

[Armhulen]

O-or with mine

[Lumbermancer]

The captain and RD have access round-start but access is not permission

I thought it's established that AI cannot disallow access to upload to authorized personnel unless they have shown propensity for harm or declared willingness or desire to upload laws that would allow harm (which is basically all other law sets).

[Slignerd]

Do note that this does not mean: "Beep boop all laws changes are harm, no access for you", it only applies if you have a strong reason to believe, or know, that someone or a group of people will be de-humaned. A core lawset change is not inherently harmful unless it is a lawset that demands harm to Humans as you currently understand them (Antimov, OxygenIsToxic etc.)

cough

I believe it's fine to be honest with the AI if you wish to change its core lawset, especially when uploading a freeform one. I guess it does help that I assure the AI that the new core lawset will still protect all humans aboard, and generally keep my word. Assuming I haven't harmed a human, from my experience the AI usually acts in good faith and doesn't cause me any trouble. I suppose I wouldn't want it any other way.

[Oldman Robustin]

AI's are allowed to be smart and realize letting in the captain who's fresh from killing cultists probably wants to upload a law allowing human harm.

Without prior suspicion, AI's can also ask for a third-party to help inspect the law before it's uploaded, although this should not be used in an unreasonable manner.

This is the correct answer.

Though back in my day the first sentence was "allowed" it was "required". Most AI's relish the chance to start murderboning so when the bloody captain rolls up after murdering 3 dudes in handcuffs and says he wants to upload some new laws, there really shouldn't be any circumstances where the AI goes "yea sure Captain come on in".

If I murdered my way onto the bridge with a double esword and ask for upload access and the AI gave it, people would be begging for the AI ban - I don't see it being any different if a captain/RD willfully (aka not in self-defense) executes/murders people. Asimov has gotten so lax and "optional" nowadays it might as well be a PURGE lawset. The differences between the two are fairly miniscule now (purged AI's can't randomly harm or mass-harm either, they have to follow Rule 1 still).

[PKPenguin321]

Why is this thread still open? We answered the policy question. Everything since then has been pointless faffing.

ping us to lock em next time