Jump to content

Mapping out the AI [Feedback Thread]


Recommended Posts

Can we get this as a testmerge for a while? To see how having no AI and 3 borgs works out?

I like to play AI yet at the same time I see why many people have issues with it, it can easily ruin the round for many antags.

Ultimately, I will be sad if we lose the AI because there are memorable ones I enjoy to interact with.

Link to comment

The third stationbound shouldn't negatively impact balance much at all I don't think, they don't have many of the issues with the AI role and are a lot more positive for roleplay and interaction too.

The main thing that I see that's wrong with the AI, is that it has the second highest round impact out of any role in the entire game, and once you recognize and accept this as a problem, the solutions to this that I see are:

  1. Actually recognize it's crazy that we don't expect an AI player to be as trustworthy and positive for roleplay as the Captain player who almost has less power and round impact than them, and throw the AI whitelist into the Command whitelist. This changes nothing in the actual AI hierarchy, it still has no authority over anything but the 'borgs, but the people who can now play it are also the ones we trust to play high power roles like the Captain responsibly without being a negative influence on rounds, with similar expectations. Synth-ban people the same if they aren't following their laws, if they try to play the AI as a command staff member, then command whitelist could be considered for strip. (We should expect Command whitelistees to know the AI isn't a Command member, it being in the whitelist, isn't an excuse. The Reps are also in the whitelist with no real authority), then tweak the AIs' mechanical means of reacting to situations, make cameras have an indicator so that it's clear they're in an area. Could be an APC area check + a small little flipping pixel overlay. Maybe make airlocks automatically unbolt with a timer on code green? and the AI/'borgs can't electrocute airlocks at all. And in their place the AI/'borgs can bring down the emergency shutters that require ID access.
  2. Remove the AI, and/or replace it with a new type of 'borg that the other 'borgs are automatically linked to, @kyres1 had a similar idea, which could be called the CIU (Central Intelligence Unit) and basically just be a regular 'borg kinda but maybe gets overclock by default or something to make its gameplay a little unique.

That's p. much it, whitelist the AI and do some QOL to make it less agonizing or remove it. If it's not whitelisted, we can't change the culture surrounding it, so we're always going to have the complaints people have against it, and if we don't do the QOL changes to 'borgs and the AI, Antags will always have afew issues with them, but mostly AI.

Link to comment
49 minutes ago, Pratepresidenten said:

Ive said it before and Ill say it again, AI removal or no, a third borg slot shouldnt be added.

Borgs are station utilities with the power to one-man run an entire department, and adding another slot would just cause more trouble than its worth imo.

What I expect the most is a third slot going rescue, mining, service or construction, with the other two going engineering and medical, which seems like it'd be okay, if two went engineering or medical it'd be a problem.

OH and I might try to code it in so you can only have 1 borg in each module if that helps.

Edited by Chada1
Link to comment

Instead of a removal, have we considered greatly nerfing the AI instead? Removal of electrifying doors. Opening doors and bolting doors come with a timer. Something where AI is only allowed on voted extended rounds? Whitelisting AI? Writing clearer server rules for AI?

I've had a number of interesting interactions with station AI's and although shitters exist, I would hate to see those interactions entirely lost.

Edited by niennab
Link to comment

We are sadly removing both types of AI with this.

Then one that doesn't give a fuck, shuts down antags and moves onto another server after being talked to and then one that spends their entire round interacting with crew, not following, hunting or shutting down antags.

Much bigger fan of the whitelist solution, although that would be the one that requires work.

Feels like a really big duct tape over a community issue instead of a mechanical one, we'll see what we need to remove next.

I'd just like both sides of this to consider why exactly they want to keep it or see it gone. For me it's obvious that the provided RP far outweighs the balancing issues.

I'm having a hard time trying to find the friendly words for "git gut". It's just another situation of "do we really need that for antags" VS "do we really need that for RP"

From an admin perspective though, almost nobody ahelps an AI. Dead chat salts, AooC salts, ahelps are rare.

We're rather strict with AI players, but if the community effort comes down to voting it out that's fine with me.

Link to comment

Whitelisting seems like a duct-tape solution (how can people try AI and get good at it if it's behind a lock?), even with a trial where they can get run out over the span of that one week, but it seems to be better than no whitelist (although whitelists don't catch all the shittery). I will say that we do get good and great AIs even without a whitelist (RIP Mistress...)

This is a bit of a community issue where there's a sort of focus on "winning", especially since the AI can quite literally snap its fingers and trap traitors in a room with bolts. Perhaps nerfing the AI, getting rid of follow verbs, instant tracking, door electrifying and the like can help? Granted, nerfing into the ground doesn't feel like a solution and indeed takes away a lot of what AI is. If the issue is with shit behavior then (sorry mods and admins!) ahelps should be sufficient as opposed to ranting in dsay.

Perhaps a better question is what does removing AI add and take away. On a side note, an AI can bolt antags in a room, but it can also be ordered to bolt antags in a room, and an AI player can also find a reason to not bolt antags in a room if it helps keep the round progressing. An AI can take away from RP and can also add to RP in more ways than deciding whether or not security wins.

I'm not for AI removal but I am slightly biased in that regard. Removing AI seems like something that has to be discussed more. As an aside, before saying something should be discarded, vis-a-vis, AI, it would help to actually try it out.

Link to comment

Please, PLEASE do not remove Ai's, it would be a horrible mistake.

The arguments that I've seen in past on this topic are: It shuts down antags. Not... really. You know if you played against me antagging. At worst, the follow verb is the culprit here because it's impossible to lose track of ai once it's ontop of it. That's about the worst I can say.

Electrifying doors? Bolting them? Removal of those powers.... I have to honestly ask if you are insane when suggesting that. Here's why: The ai already has limited power of being malicious without being malf [and malf pretty much has explosions and gravity slams on top of the normal ones]. A big part of antag play is working with the ai in some way, be it sneaking around it's vision, trying to not catch it's attention in the first place, preparing escape routes if things go wrong that the ai can't stop you with... it makes a player feel like James Bond once you have it down. Not to mention subversion, the ai NEEDS to have access to those things.

Literally the only thing I'm willing to step back on is that the follow verb is .... a bit too much at times. Otherwise, please leave ai be.

Link to comment

The way bolting is handled is p. much going to get changed remove or don't remove, it's one of the biggest complaints people have against 'borgos and I'm likely going to fix it given time @Naelynn, the thing I'm thinking of doing is making any (Outside of actual airlocks) airlock with an intact bolt wire automatically unbolt in about 10 seconds on any code level except code red. And then if 'borgs/AI need to keep crew out of an area, like in a phoron leak/fire/blob/carp they can just pull the fire alarm instead, and won't be able to trap antags anymore.

Edited by Chada1
Link to comment

I think that a malf AI probably should have access to weird and wonderful things. The issue is with general AI's power. Oft times it's not even the AI's fault per se, just a product of their abilites. For example, I've had my antag gimmick shut down because some one said ";help" over the radio. AI leapt to their location and it was game over.

Removing the AI's ability to jump to people and instead rely on jumping to cameras would help if we're not whitelisting them. Maybe the "10 seconds and the bolts raise" suggestion from chada above would be cool.

Link to comment

I would be glad to remove the AI.  It's something I succeeded in getting trialed on Bay before.
The reasons are as follows.
1) The AI is essentially unlimited in its' power.
The only limits an AI really has is:
Running out of power, which essentially never happens because the SMES is good for pretty much forever if you know what you're doing.
Cameras, which alert the AI and also anyone looking through cameras at the time if they're disabled (yes, it's pretty easy to know if one's been stealthily taken out.)
The player being generous enough to not exert their full strength onto the crew/antag, which is an extraordinarily shitty way to balance.

2) The AI is bad for antaggery.
Being bolted and shocked is not the only thing the AI can do against you, though it's certainly the most egregious power.. It can (and often has in certain cases) shout over every channel (and make command-level announcements) of your location and keep others away from you. It's the most powerful anti-antagonist tool, bar none, and its' laws essentially mandate it to do something about antags. Not only this, but it's incredibly hard to take the AI out (camera alarms, it's easy to know if one's been stealthily taken out) due to the fact that it's locked in a secure bunker that's harder to get into than the Captain's office. Keeping track of the AI and countering its' moves takes up like 75% of your brain, leaving very little head RAM for anything else. You either kill it, or you keep running away from it.

3) The AI offers very little.
I will give you a free drawing of your character in MS Paint if you can name 10 things the AI commonly does for you that don't relate to door access and character interaction.

4) Nerfing it does not work.
Most attempts to nerf the AI's powers, such as being able to immediately jump to you have been met with complaints that they 'gut the role'. I, for one, am not willing to deal with the vitriol, and I don't imagine there's legions of developers willing to rework it either. The AI role has not fundamentally changed for over a decade now.

5) Whitelisting it does not work.
Imagine tomorrow we merged a PR that gives every engineer a rocket launcher (and twenty backup rockets, 'just in case.'). Imagine, if you will, the engineers proceed to use their new bazookas to immediately blow up every possible antag they find. The sane answer would be to take away the rocket launchers, not to whitelist the engineers so that 'only the good ones that sparingly fire rockets can play.'

6) Writing rules doesn't work.
To lazily quote Skull, 'If we make rules for the AI. We must make rules for sec. If we make rules for sec, we must make rules for medical doctors and how they treat patients. It's a fucking rabbit hole, as Alb said and CM has demonstrated.'

7) (Subjectively!) The population of good AI players could not hope to outweigh the population of bad ones.
Title.

? AI does not fit the game.
Whoever decided to have an omniscient eye in the sky in a game about paranoia needs to be beaten. It also wildly tilts the game into the crew's favor, rather than having the balance at least pretend to be equal.

Link to comment

These arguments go around over and over and we always come back to the following.

 

1) People agree the AI is broken, but don't know how to fix it.

2) People claim there are good AI players , and then name one, possibly two; one of whom no longer plays here. The number of bad ones vastly outweighs the good.

3) There's likewise the arguments you can play around the AI. Yes, you can. It is not, however, all too enjoyable and borderline powergaming to enter the core/whatever doesn't add anything.

4) 'Just ahelp if there's a problem', despite entirely ignoring that the fact utterly shutting an antag down is, in theory, in line with the laws.

5) 'Whitelist it'. This is actually one of the few points I agree on. It should be behind a command whitelist, given that's the only one that really vets players from an OOC angle (At least, to the greatest degree).

 

The main problem to my mind is that a role is entirely catagorized by what it does best. And the 'best' skill an AI has is to fuck with antags and validhunt. Period. Very, very few add anything of value to the RP side, and removing the poorly played ones would contribute more than the handful of good ones ever could.

I also feel like if nobody is actually willing to work on the code associated with the usual solutions: removing tracking, bolt timers, ect - which they've showed time and time again they're not; I mean, how many times have we had this conversation? - removing is simply better.  Especially given the most fun element, malf, has been entirely absent since it was removed from rotation for a series of very good reasons.

That said, I am biased against AI and do not think it adds anything in a HRP enviroment given people have shown over and over they cannot be trusted. In theory, the role is great. In practice, it sucks.

 

Edited by Lemei
Link to comment
3 hours ago, ParadoxSpace said:

4) Nerfing it does not work.
Most attempts to nerf the AI's powers, such as being able to immediately jump to you have been met with complaints that they 'gut the role'. I, for one, am not willing to deal with the vitriol, and I don't imagine there's legions of developers willing to rework it either. The AI role has not fundamentally changed for over a decade now.

5) Whitelisting it does not work.
Imagine tomorrow we merged a PR that gives every engineer a rocket launcher (and twenty backup rockets, 'just in case.'). Imagine, if you will, the engineers proceed to use their new bazookas to immediately blow up every possible antag they find. The sane answer would be to take away the rocket launchers, not to whitelist the engineers so that 'only the good ones that sparingly fire rockets can play.'

6) Writing rules doesn't work.
To lazily quote Skull, 'If we make rules for the AI. We must make rules for sec. If we make rules for sec, we must make rules for medical doctors and how they treat patients. It's a fucking rabbit hole, as Alb said and CM has demonstrated.'

7) (Subjectively!) The population of good AI players could not hope to outweigh the population of bad ones.

4; Nerfing does work, the issue you're citing in that paragraph isn't even against nerfing, it's against peoples willingness to accept nerfs, or to code them, and it could be v. easy to fix that problem by just making the nerfs v. simple, like the bolt timer, cameras being more visible when the AI is in an APC-area, etc. Scroll up to my earlier post.

5; Following this logic to its logical end, if it were true, Aurora should remove all of our whitelists, the command whitelist, including our species whitelists. If they don't ensure the mechanics of species are used responsibly and in a positive-for-RP way then why do we have them? The answer is that they DO WORK and people love to say that they don't but they don't seem to like testing it to see if they do or not. The Command whitelist matches almost every single OOC criteria for what we want in an AI player. Let me go down a list of things command players can get their whitelist stripped for:

  1. Choosing the role and ignoring its' responsibilities to gain authority over the other related roles and overreaching in their authority.
  2. Discouraging RP and taking actions that shuts it down for the rest of the crew.
  3. Unreasonably responding to situations
  4. Abusing their jobs power and mechanics to valid-hunt/other powergaming and shut down Antags.
  5. Ignoring the OOC expectations of their role.

All of this is whitelist strippable for the Command whitelist, now tell me that all of that doesn't completely fit the exact bill of what a good AI player should avoid?

6: We already have rules for Stationbounds, the AI doesn't need unique rules, every rule that applies to 'borgs also applies to AI, the problem is that an AI is given situations a 'borg is never able to encounter due to it being a disembodied entity that can be anywhere at any time, and CM uses their whitelists differently than how we do. Their bishops (The equivalent of their IPCs) are setup to require powergaming to maintain the whitelist. We would never do that in a million years, every whitelist we have exists to PREVENT powergaming so using their way of handling whitelists and rules for what we'd become is a v. poor way of looking at it. We are not CM. We don't even have the same administrative policy and ethics.

7: This is partially because we leave the role open for absolutely anyone to play, if we had it behind a responsibility-driven whitelist (The Command whitelist) we could more seriously moderate it and the slot would more often remain open for the ones people consider good. We can actually enact a gradual player culture shift if we put it behind a whitelist and tweak some of the mechanics.

Link to comment
On 09/05/2020 at 16:29, ParadoxSpace said:

[...]
2) The AI is bad for antaggery.
Being bolted and shocked is not the only thing the AI can do against you, though it's certainly the most egregious power.. It can (and often has in certain cases) shout over every channel (and make command-level announcements) of your location and keep others away from you. It's the most powerful anti-antagonist tool, bar none, and its' laws essentially mandate it to do something about antags. Not only this, but it's incredibly hard to take the AI out (camera alarms, it's easy to know if one's been stealthily taken out) due to the fact that it's locked in a secure bunker that's harder to get into than the Captain's office. Keeping track of the AI and countering its' moves takes up like 75% of your brain, leaving very little head RAM for anything else. You either kill it, or you keep running away from it.

3) The AI offers very little.
I will give you a free drawing of your character in MS Paint if you can name 10 things the AI commonly does for you that don't relate to door access and character interaction.

4) Nerfing it does not work.
Most attempts to nerf the AI's powers, such as being able to immediately jump to you have been met with complaints that they 'gut the role'. I, for one, am not willing to deal with the vitriol, and I don't imagine there's legions of developers willing to rework it either. The AI role has not fundamentally changed for over a decade now.

5) Whitelisting it does not work.
Imagine tomorrow we merged a PR that gives every engineer a rocket launcher (and twenty backup rockets, 'just in case.'). Imagine, if you will, the engineers proceed to use their new bazookas to immediately blow up every possible antag they find. The sane answer would be to take away the rocket launchers, not to whitelist the engineers so that 'only the good ones that sparingly fire rockets can play.'

6) Writing rules doesn't work.
To lazily quote Skull, 'If we make rules for the AI. We must make rules for sec. If we make rules for sec, we must make rules for medical doctors and how they treat patients. It's a fucking rabbit hole, as Alb said and CM has demonstrated.'

7) (Subjectively!) The population of good AI players could not hope to outweigh the population of bad ones.
Title.
[..]

2) A whitelist can alleviate this. When I play as an AI and notice someone breaking into somewhere where they probably shouldn't be, I typically ignore them. An antag has access to bomb every critical area of the station if they feel like it, but they don't (typically). Instead the antag looks to further role play- the AI should be held to the same standards.

3) Searching for someone on the station; Creating announcements on behalf of command not at their offices; More easily handle configuring air alarms; Be a relay between borgs and command/ other departments; Provide on-time alerts of station alarms; Being a target for antags to powergame with; Another set of eyes on suit sensors; Someone to tell to do a mundane task (like turning on or off a shit ton of lights or something like that i dunno; Someone to re enable fire alarms when the assistant decides to turn them off; A literal slave to command.  The AI is something you can say "Hey do X task" and it should do it. It's not always easy to access many switches or physically check on all the SMES when something is amiss.

4) I don't think nerfing is worth anything for the AI. if you are going to do something it would have to be a complete rework on how it access systems.

5) Whitelisting could work, and IMO its better to try that before removing it completely. Worse case scenario, nothing change and we just remove it. Best case scenario, things improve. This should be followed through with guidelines on how to role play for the betterment of the round, not murder boning someone in the wrong area. Also, your argument seems like a fallacy- the equivalent of saying "If the janitor had access to a nuke disk/code, you wouldn't whitelist the job, you would remove (REWORK) the janitor's access to a nuke disk."

6)Rules, no- I don't like that idea. However, a guide on how to play AI that command whitelist would be heavily suggested to follow- yes.

7)True, whitelist the role to cut down on bad players (and some good ones). Issue AI bans for power gaming due to the heavy round effect they carry? Im not even sure if thats a good idea tbh.

 

I love playing the AI, the AI is also the character which greatly helps the crew when there is no command available (a situation that wouldn't likely happen in a non-fictional world)

If we as a community believe that removing is best, I suggest we focus on trying whitelist first, if that doesn't work then removal til rework would also be good.

Link to comment
7 hours ago, afitz said:

If we as a community believe that removing is best, I suggest we focus on trying whitelist first, if that doesn't work then removal til rework would also be good.

The thingo is if we remove the AI, 95% likely it will never EVER be reworked, it's p. much now or never. Once the culture adjusts and shifts to be used to the AI being gone -- no dev/player is likely going to be willing to even try to fix it. That's why I agree tho we should try to whitelist and fix it and then proceed to removing instead of just removing until fix.

Another idea was brought to me recently, which is to remove the AIs ability to see names on the radio or on IDs. instead, it sees just the job name (This exists in telecomms code already and can be made a more permanent thing just for the AI) and when it sees a mob, it only sees a silhouette of whatever species it is, instead of equipment or clothing. This way the AI can't be biased in any way or form since it literally cannot tell crew apart from each other except for their rank and role (Job and whether they're command) which means one roboticist is as prioritized as the other roboticist in requests. This makes an Antag using a false ID or wiping the Crew manifest actually a potent way to disable the AIs ability to gain information, and would ensure mechanically that the AI would act more like a machine.

This also means interactions would go more like. Captain says, "AI, Open X Airlock." instead of Bob Planter [Captain] says, "AI, open X AIrlock." which could be a way to solve a lot of the problems with AI right off the bat. The problem? I don't even think this is worth exploring by itself without whitelisting it too, but it's the best idea I've been given so far.

Link to comment

I am going to support adding a whitelist to the AI role to limit it to players trusted by the community to not act upon shutting down the antagonists while maintaining the station within their laws as there were already good points made by people like Chada. However, it may lead down to the road of high bar expectations of AI players, built around how our wonderful AI players like Harmony and Mistress.

For now, keep it but whitelist it.

Link to comment

I'll keep it simple and to the point. I am in 100% support of Chada's approach. Whitelist it, hold it to higher expectations, and if it doesn't work for whatever the reason that may be. Remove it. Outright removing things without attempting to fix is a bad card play. I like the silhouette idea but I insist the AI should see silhouette after being tampered with via APC or some computers/laptop connected to the network or refer to the old suggestion of mine malware

 

Link to comment
Guest
This topic is now closed to further replies.
×
×
  • Create New...