Limit AI's ability to be extension of security

[Kaed]

In some instances the AI can be a lifesaver, reporting attacks on crewmembers or dangerous intruders being present, but much more often their ability to easily see and track any crew member in camera range is used to tattle on anyone performing even minor infractions by immediately alerting security that a crime is in progress, then opening every door on the path for security to arrest that person who hacked open a door or whatever.

We've done some to alleviate this by preventing AI from bolting doors to trap every antag they see, but as a protocol rather than mechanical nerf, I propose that AI, not being a crew member but rather a non-person piece of station equipment, can only be used as an advisory body to non-violent crimes, rather than being treated as a witness.  For example:

-The AI sees someone hacking into engineering, and reports this to security.  Under our current standards, this would probably result in at least one officer or overeager HoS rushing over to arrest the culprit sans warrant, because they have spotted them in the area they are not supposed to be that the AI reported them as having broken into.

-Under the new standard, officers may report to the scene following the AI's advice, or even check the area on cameras, but even if they find the crew member where they aren't supposed to be they must perform more work than simply taking the AI's word for it that a break-in happened and immediately performing an arrest.  Are the doors damaged? Can they confirm with the staff running that department that the person who broke in isn't supposed to be there/was not invited in?  If the department is willing to back up their surprise visitor either through being intimidated, compelled, or simply anti-authority (looking at you, engineering..), or perhaps the antag can come up with believable reason why the AI must have been mistaken about what happened, then a warrant with relevant evidence is required to perform the arrest. This also applies to searches if it is code green.

 

If the AI's testimony cannot be taken as official evidence for the sake of arrest, security officers have to do more work than simply respond to AI alerts with immediate non-warranted arrests.  This mostly applies to MINOR infractions, where no one has been hurt, and you are generally expected to have a warrant before arrest if you haven't physically seen a crime happening.

Edited June 12, 2020 by Kaed

[Pratepresidenten]

7 hours ago, Kaed said:

If the AI's testimony cannot be taken as official evidence for the sake of arrest

I feel this is a really hard thing to balance properly. Personally, I never ever take the AI's statement/testimony as fact. If there are no evidence or witnesses and everything hinges on the AI saying they saw X do Y, then there is literally no case. Dismissed.

But the main issue I see with wrangling AI "Authority" is:

The AI is meant to be an absolute, infallible overseer and you can without question trust what it is reporting and make any arrest based on its information.

Balanced with

Fair play, not partaking or encouraging validhunting and making the round overall a better experience for everyone involved.

 

Maybe all thats needed is just a line somewhere for security to know that the AIs testimony is a corroboration at best?

[Kaed]

2 hours ago, Pratepresidenten said:

But the main issue I see with wrangling AI "Authority" is:

The AI is meant to be an absolute, infallible overseer and you can without question trust what it is reporting and make any arrest based on its information.

 

Maybe all thats needed is just a line somewhere for security to know that the AIs testimony is a corroboration at best?

The main principle that I can see being applied to that statement is that while the AI might be an 'infallible overseer' (a statement I regard with extreme skepticism, because people on server tend to regard even the slightest strange behavior from the AI as evidence that it is malf/subverted), it is still not a person and can't provide legally admissible testimony to someone's guilt.  Its statements cannot be considered objective because everything is a result of following a set of behavioral guidelines that are not meant to be used for catching criminals.  

For instance, with some finagling in concept, one could judge that the law that says to 'serve/protect the crew with priority according to rank' could lead an AI to falsely accuse a lower ranking crew member of the crime made by their superior - according to the way the law chain goes, the higher ranking person has priority over their subordinate's safety.  This rarely happens IC because usually AI players want to catch the bad guys (which is the problem this suggestion is trying to curtail) but it is possible.  

And while you might disregard AI testimony, I've seen security teams fall back on AI testimony as evidence.

Edited June 12, 2020 by Kaed

[Pratepresidenten]

Yuh, there's definate room for change/improvement when it comes to this topic!

[Carver]

At what point do you simply remove the AI instead of adding a bunch of nonsensical rules? The thing is supposed to be one of the most trustworthy assets on-board, previously beside loyalty-implanted personnel - and now rather alone in that regard.

If the AI can see what you're doing, either be faster with it or disguise yourself properly. If a CCTV records you doing a crime with your face and similar perfectly visible, then you'd be just as free to be apprehended - if you're conveniently on-scene when law enforcement or security arrives, you'd also be apprehended. If I were to need to go through a million extra loops because of a different ruling, you'd be damned sure I'd be keeping you held in brig whilst that investigation went underway if you were seen or caught on-location by the AI or anyone.

Edited June 13, 2020 by Carver
It should be remembered that you're not in a public location, the entire station is private property so under no circumstance are you free to just walk off during an investigation either.

[Kaed]

12 hours ago, Carver said:

At what point do you simply remove the AI instead of adding a bunch of nonsensical rules? 

 Never, and this is a needlessly aggressive somewhat off topic statement. Thus, putting aside the extremist "this is too frustrating to deal with make it go away" stance, the rebuttal here would be that we don't need an infallible source of information on server.  The more time we spend communicating with other characters in this medium, i.e. role playing, rather than immediately escalating to submit or die scenarios because an overseer entity that you have no way to track the vision of happened to spot you doing a naughty, the more rich and dynamic our experience will become.

"Git gud, scrub" arguments also fail to address any sort of issue by dismissing the fact that there is an issue other than the individual player's skill level, they are both reductive and elitist. Why must the onus of effort fall entirely on the antagonist, while security team members can just "computer box say tator man bad" and bring 1-3 other friends with them to curb stomp the unfortunate door hacker with no additional recourse or effort?

Edited June 14, 2020 by Kaed

[Chada1]

We don't need to do this, we already removed bolting @Kaed, the AIs ability to be an extension to security is completely limited to keeping an eye on suspects, they can't lock people down anymore near at all.

The final thing we could do is remove the AIs ability to jump to and automatically follow persons, but that's it really. That coupled with that most Antags have the ability to use an AI cloaking device now too (If they have access to an uplink) would do it.

Edited June 14, 2020 by Chada1

[Carver]

On 14/06/2020 at 04:41, Kaed said:

the rebuttal here would be that we don't need an infallible source of information on server.

Then you don't need an AI.

On 14/06/2020 at 04:41, Kaed said:

The more time we spend communicating with other characters in this medium, i.e. role playing, rather than immediately escalating to submit or die scenarios because an overseer entity that you have no way to track the vision of happened to spot you doing a naughty, the more rich and dynamic our experience will become.

Two answers to this: Either port the device that tracks when an AI is watching you, or don't perform actions yourself that escalate to 'submit or die' in plain view of a camera network. The basic measures of stealth apply.

On 14/06/2020 at 04:41, Kaed said:

"Git gud, scrub" arguments also fail to address any sort of issue by dismissing the fact that there is an issue other than the individual player's skill level, they are both reductive and elitist. Why must the onus of effort fall entirely on the antagonist, while security team members can just "computer box say tator man bad" and bring 1-3 other friends with them to curb stomp the unfortunate door hacker with no additional recourse or effort?

Because this is a video game and assuming you're willing to learn how to play the game, then you should find relative success for your efforts in regard to skill. It is not intended to be some form of easy street theatre play where everything goes smoothly without fail. If you do something that necessitates that response, then you must be willing to accept the consequences of your actions and plan around them - something I learned well long ago both playing Antagonist roles and Security (both of which have a fair amount of mechanical crossover that helps you learn to play the other).

[Zyrus]

On 13/06/2020 at 16:03, Carver said:

At what point do you simply remove the AI instead of adding a bunch of nonsensical rules?

This is a HRP, not a LRP powergaming server. There are players (like me) that enjoy playing humanlike AIs, and would be very angry if it was removed just because powergamey whiners don't like that their gimmick was found out early.

[Cnaym]

On 15/06/2020 at 19:11, Carver said:

It is not intended to be some form of easy street theatre play where everything goes smoothly without fail.

Nailed it Carver.

Changing the AI will not improve our antags, we had far better antags even with bolted doors. All it did was sort out bad AI players, which is a good thing.

Stealth is will never be a good as a / or for a gimmik, nor will a powerfantasy be good when 80% of crew just return to extended because they cannot do anything else.

This is not a singleplayer hitman game, but more of a Payday session with drunk friends. The fun comes from the mistake both sides make. I never even see getting a headshot in spessmen but the officer running in to slip on oil will always crack me up.

Instead of being invisible, or worse yet winning, an antag should focus on being fun to interact with. That will give you more breathing room than any nerf or buff ever can. If people enjoy the antag, they will keep them around and interact more with them. I had longer conversations inside the brig than some of our current antags manage to get outside of it.

This is the issue, you will not fix that by changing other jobs, but only by playing a better antag than this, in hopes that one or two will pick up on it and improve.

Because giving them more stuff has not worked.

Because telling them at 1:50 in AooC to finally do something has not worked.

Because giving them a buddy for certain roles has not worked.

Because drafting from everyone who voted secret has not worked.

Even removing the AI cannot fix the current mindset of both, security and antag players.

You can run the experiment yourself, play an absolute useless AI, never helping security find anything or anyone, it makes no difference for the antags at all.

[Zelmana]

Agreed with everything above Cnaym and Carver posted.

 

We keep trying to solve player responsibility and proper roleplay expectations with rulesets or code changes. There are some acceptable use of this but we need to be cognizant that we do not go too far.

[Lordnesh]

5 hours ago, Cnaym said:

You can run the experiment yourself, play an absolute useless AI, never helping security find anything or anyone, it makes no difference for the antags at all.

I don't play a useless AI, but I definitely play more hands off. And this is true. I played a round where security was searching for a reported weapon, and I was fairly certain I knew where it was, based on my observations. However, I treated that knowledge as meta knowledge, since at the time I hadn't been requested to observe the target at the time. Despite essentially helping the antag get away with it, it didn't really change anything.

I don't know what the answer to the problem of security and antag is, but I don't think restricting AI players more is the right one.

[Zelmana]

5 minutes ago, Lordnesh said:

I don't play a useless AI

 

5 minutes ago, Lordnesh said:

 I was fairly certain I knew where it was, based on my observations. However, I treated that knowledge as meta knowledge

I think these conflict each other. An intelligence with many petaflops of processing power is able to make reasonable abstract thoughts regarding weapons + security searching for weapons. If you knew security was searching for a weapon, and had your previous observations as notably suspicious, you are intentionally going against your lawset to assist the crew. 

There could be an argument for "i was perhaps being balanced with crew and antag play" but I don't think in this obvious of a situation.

Edited September 14, 2020 by Zelmana
added words

[Zelmana]

To me, from the way you play AI as a complete non-assistance beyond anything that is direct orders, going as far as to give a blind eye to things that security is actively trying to discover, you should just stick to the limited capabilities of borg and IPCs.......

[Lordnesh]

I was already watching them, out of boredom, way before security even thought to have me watch them.

Essentially, I knew where things were going, so I started watching the antag. Noticed some strange behavior when they found it security was looking for them, and thought to myself "that's suspicious". However, at that time I didn't really have an IC reason to be watching them.

Depending on alert level, as long as an antag isn't doing anything blatant, I'll just pretend I wasn't even looking.

[Lordnesh]

1 minute ago, Zelmana said:

To me, from the way you play AI as a complete non-assistance beyond anything that is direct orders, going as far as to give a blind eye to things that security is actively trying to discover, you should just stick to the limited capabilities of borg and IPCs.......

I am very helpful, when people ask for my help. 

 

From the wiki for AI.

What the AI should do

As AI you have the power to strongly influence the round and you should always be aware of that and consider your actions, and the appropriate responses, before you ruin someone else’s fun. Remember that the game is not about winning but about the RP and the experience of the round.

For example it can be a real killer if the AI immediately calls out anyone looking suspicious, giving an accurate description of, for instance, a wizard. It would also be considered metagaming and against the rules.

Don't just act like any normal crewmember. In fact, if possible try not to even act human. You don't even have to like humans, as long as you do your best to keep them safe. The AI has Law programming but it also has personality quirks. As a rule of thumb you should first check if any laws are threatened or not. If yes, then you have to act. If not, then you should consider the situation:

[Zelmana]

That's fine.

However, you made it seem as though you watched someone put a gun or something in a locker and then security was searching for said gun, and then failed to assist because they did not directly ask you, "AI, can you search for this gun?"

All good, and I agree AI should not auto-rat on antags and use some tact for the sake of RP. But suggesting that the AI is not an all seeing eye goes completely against the design of the role, as well as the entire robot lore.

[Lordnesh]

As the AI I don't feel it is my job to tell people how to play their role, or hold their hand. If they want my help, then they can certainly ask for it. I will never refuse a request for help. I'm almost always more than happy to be helpful. But what I will not do, is utilize my out of character knowledge about the game to call out the antags. That's not fun for the antag, and it's not fun for security.

As the AI you have a lot of influence over the round, simply by having such an unobstructed view of almost the entire station. It is important to use that information wisely. So everyone can enjoy the round. Not just security, and not just the antag.

The way I look at is this, as the AI you have a lot of information at your disposal. However, you can't pay attention to everything at once. And how do we deal with that problem now? With filters. Filters limit the amount of data we see to something more manageable.

Station on green? Filters are set to the most restrictive, such that only blatant acts are "noticed".

Edited September 14, 2020 by Lordnesh

[Cnaym]

9 hours ago, Lordnesh said:

As the AI I don't feel it is my job to tell people how to play their role, or hold their hand.

How to find the worst AIs quickly, do something different, watch the player OOCly completly loose their shit for not understanding it, face an AI that requires you to explain everything you did only to get told how to do it "the right way".

It's a voice control for the airlocks until it gets tasked with something different. Announcements about current situations make no sense, even if the AI understands the situation, it cannot know what command wants to share with the crew. Always ask command and crew how to help,  heck even basic stuff as setting up smelters for miners to help out goes a long way, but ordering around or "correcting" crew as AI is one of the biggest holes you can dig yourself into ICly and OOCly.

Back to the topic though, I strongly agree with the fact that the AI can make or break antag stuff, or just safe people from carp and such. I do however recommend not relying on the AI, or if playing as AI, not making the crew depend on you. There is nothing more boring than suit sensors in the entire game, ignore those, work with alarms and signals from crew that require active interaction, like fire alarms or holopods -> The later means someone wants to play with you, the prior means 9/10 times you are ruining someones round.

On 12/06/2020 at 09:09, Kaed said:

The AI sees someone hacking into engineering, and reports this to security.  Under our current standards, this would probably result in at least one officer or overeager HoS rushing over to arrest the culprit sans warrant, because they have spotted them in the area they are not supposed to be that the AI reported them as having broken into.

There is a ton going wrong there.

It should be made clear for the AI to not report such minor issues, instead let forensics riddle a litte, report a door issue to engineering after the antag is gone and watch two departments interact and RP.

Security arresting over messing with airlocks -> Invite the person to the lobby, over the public radio, let them explain what they did and why and then fine them tops. If you are an antag and tell me the panel was open so you decided to close it, I will ignore anything an AI has to say about it.

Antag getting spotted -> Use that agent ID, give yourself another name and use suit sensors while breaking into the captains office, always makes for a good meme to send the valid hunters another way. Being stealthy only helps you in getting a couple items to start out, apart from that it only leads to ganking and boring hostage situations. Be fast instead of sneaky, a lot more fun for everyone involved.

Last but not least, ahelp that. Ahelp the AI and the officer, heck make a staff complaint against the entire team if nobody took your ticket. You get an okay boomer for ahelping an AI rating you out, but more often than not the AI gets one of those fancy notes along the lines of "told to give antags some more breathing room". People do not get to play AI or officer for long if they collect those.

[Karhast]

16 hours ago, Cnaym said:

Last but not least, ahelp that. Ahelp the AI and the officer, heck make a staff complaint against the entire team if nobody took your ticket. You get an okay boomer for ahelping an AI rating you out, but more often than not the AI gets one of those fancy notes along the lines of "told to give antags some more breathing room". People do not get to play AI or officer for long if they collect those.

Your experience is extremely different from mine, to say the very least. I have never, ever, not once, at all, seen security catch trouble from the administration for taking the AI at its absolute word. Half of the time they don't even bother following up with forensics, they simply believe the AI and then the valids flow on ever more. I'd like for you to be right, but as best I can tell, nobody seems to care if an officer or HoS blindly follows the AI's word.

[Cnaym]

3 minutes ago, Karhast said:

I'd like for you to be right, but as best I can tell, nobody seems to care if an officer or HoS blindly follows the AI's word.

Ahelp all the things. If you think staff does not care you are posting into the void. If you do think staff cares -> Ahelp

If you want to help staff handling it, go apply. Honestly take a trial, take a look at what they handle and what not, it's an overall fun experience.

The issue is the "best as you can tell" part, only way to be certain is ignoring it and not ahelping (in which case nothing happens) or going for some moderator trial and taking a look at it from the other side. 

In conclusion: Ahelp all the things.

[Karhast]

Yes, I did ahelp all the things, and the ruling was that security did things just fine. I wouldn't be posting here if I didn't think that this is just policy by now.

[Amunak]

I'd actually love if AI still had an option to bolt doors; there are absolutely valid reasons to do so even outside ruining fun for people. Door leads to unpressurized area or something? Bolt it so that crew doesn't accidentally wander in. Doing stuff like that is quite fun as an AI player, IMO.

To help the issue with AIs fighting antags too much we could (and IMO absolutely should) add OOC notes to their role for start-up where it tells them what to do and what to avoid. Even just saying something like "Please remember that as an AI player you are extremely effective against non-crew and thus also potentially effective at ruining everyone's fun. Don't forget to give all players some leeway; you don't have to report every petty crime, call out every suspect or provide unfallible evidence without asking. Your role is supposed to be supportive and assistive, not assertive." could probably help, especially for newer players. Maybe then we could even bring back bolting...

And if not, perhaps some other mechanics could be added to facilitate AI blocking access (for at least willing people), while not blocking access for antagonists. Perhaps an opposite to "IDscan" - a toggle where yellow lights light up on an airlock, and some special permission (which, say, engineers, EMTs and command would have) or intent (maybe a confirmation "Are you sure to enter?") is required to override it?

One huge problem with AIs being able to bolt only when antag is IMO the fact that OOCly when people see more than handful of bolted doors that otherwise wouldn't be they can guess it's malf AI.

To prevent AIs ruinging antags we could also base the camera follow feature on suit sensors; if they were turned off completely the tracking could be made impossible. But I'm not sure what other repercussions that'd have.

Edited October 17, 2020 by Amunak

[KingOfThePing]

7 hours ago, Amunak said:

I'd actually love if AI still had an option to bolt doors; there are absolutely valid reasons to do so even outside ruining fun for people. Door leads to unpressurized area or something? Bolt it so that crew doesn't accidentally wander in. Doing stuff like that is quite fun as an AI player, IMO.

Dont open emergency shutters if you dont know what is on the other side, they are everywhere. They also almost always go down, so there are not really many reasons to bolt doors except deny access. At least I cant think of much from the top of my head

7 hours ago, Amunak said:

To help the issue with AIs fighting antags too much we could (and IMO absolutely should) add [snip snip]

People don't care, they already know that, it's not working

7 hours ago, Amunak said:

One huge problem with AIs being able to bolt only when antag is IMO the fact that OOCly when people see more than handful of bolted doors that otherwise wouldn't be they can guess it's malf AI.

Not a problem anymore since malf is a voted gamemode and not in the secret rotation

7 hours ago, Amunak said:

To prevent AIs ruinging antags we could also base the camera follow feature on suit sensors; if they were turned off completely the tracking could be made impossible. But I'm not sure what other repercussions that'd have.

This is an idea that was brought up in the past, and I still think that's pretty good. Camera tracking is way worse than bolting doors imo

[Lordnesh]

I would support linking camera following to suit sensors. Although, agent IDs have a function that make it so the AI can't track you.