AI RP and Behavior Rules revamp

[JKJudgeX]

After many, many rounds on the old map and the new, a recurring trend that I am seeing way too much of is the AI playing as part of security, all the time. In order not to nerf the AI so that it's still fun to play and useful to the crew, I would like to see some of the following RP guidelines enforced for the AI. Many of these ideas come from how I remember the AI being played about 5 years ago on various servers (RP and not), and some of them are just because some of the antag modes (cult and changeling, especially) seem to run very slow with very little action because of the amount of subterfuge required to avoid the AI tattling on you and thus bringing the round to a boring close.

1) The AI is not security-minded. It can assist with security functions, but, unless special rules and directives are installed or spoken to it, the AI does not report "suspicious behavior". It should not be using intercomms to listen in on conversations and then hand over the conversation log to security, unless asked (yes, they do this). It will report things like being asked to open doors that it shouldn't. It will report things like unknown crew members boarding the station. It will report injuries - but only if it sees them because of suit sensors (it doesn't necessarily understand organics completely).

2) The AI is vague in its reporting, requiring back and forth to coax more information out of it. "Crew member Douglass seems to be injured in arrivals maintenance" might be how it reports that a changeling has pulled someone into maintenance and started absorbing them. If asked "how was he injured?" it might say "Another crew member was involved." and finally reveal who has attacked Douglass if asked "Which crew member?".

3) The AI awaits commands to bolt doors, except for in clear, immediate danger-to-crew situations. A known murderer is trying to hunt down the RD and there's one door between them? The AI might bolt that. The AI does NOT lock down antagonists or report their every move without input. It must be queried for that information, and told by a crew member to lock an antagonist on the run down.

4) Generally, the AI player should respect the game mode and allow a generous, sporting chance for antagonists to work some conflict into the station before spoiling the roundtype and/or bringing about the conclusion of the antagonist's round via bolted doors and divulged information to security/command... much in the same way that antagonists are to RP and promote enjoyment of the round by choosing gimmicks that are more involving than "murder everybody".

The AI on Aurora is generally not horrendous about breaking these kinds of behaviors, but, close to half the time, the AI does seem to get very, very involved with valid-hunting, on its own volition, and that seems a little strange to me. At a certain point it stops seeming like a synthetic intelligence and starts seeming like a guy playing SS13 with access to all cameras and doors who wants to valid-hunt. I'd like to see some serious rules, even if not these, put in place to limit this kind of behavior. The power of all station camera access with jump-to capabilities and the crew monitor, plus being able to bolt people in places as soon as you suspect them of something is just altogether unnecessary, and results in antag rounds often seeming slow and stupid (because the antag got busted by security before they could really get things started, the first time they needed to kill someone).

None of this would apply to a Malf AI and of course anything can be overridden and modified with clever directives.

I'm 100% aware that all of this can be played around by subverting the AI/being tricky/etc, I have a somewhat decent success rate as antagonist as far as getting away with a few murders or starting a semi-successful cult/rev goes... but, if people want more action in those changeling and traitor rounds, the super-cop AI play has to be curtailed. I get it, disabling cameras is "ezpz", but A) it's a dead giveaway that summons security to disabled cameras half the time even if no alarm goes off, and B) is a little silly to have to squint our eyes and believe every possible antagonist suddenly knows how to disable cameras like a professional electrician.

This isn't me asking that AI be removed from keeping the crew safe/etc... this is me asking for clearer, more pro-RP rules that steer AI players towards being a synthetic and utility for the crew, and away from round-ruining valid-hunting.

That shouldn't be why you're playing AI.

[Skull132]

o_boi.png.avi

imma steer clear of this and bat it into the proper court. [mention]Garnascus[/mention][mention]Shadow[/mention]

i lied. I can't not reply.

4) Generally, the AI player should respect the game mode and allow a generous, sporting chance for antagonists to work some conflict into the station before spoiling the roundtype and/or bringing about the conclusion of the antagonist's round via bolted doors and divulged information to security/command... much in the same way that antagonists are to RP and promote enjoyment of the round by choosing gimmicks that are more involving than "murder everybody".

 

I think fake chances for the sake of chances are shit. Do understand that this is coming from a person [me] who basically powergames within the confines of his RP, but still. They're fake, unnatural to the flow of the game, and place a heavy burden on new players who are not naturally aware of these things. Mechanical solutions for problems like this are a way better alternative. If any reasonable ones exist.

[Scheveningen]

1) The AI is not security-minded. It can assist with security functions, but, unless special rules and directives are installed or spoken to it, the AI does not report "suspicious behavior". It should not be using intercomms to listen in on conversations and then hand over the conversation log to security, unless asked (yes, they do this). It will report things like being asked to open doors that it shouldn't. It will report things like unknown crew members boarding the station. It will report injuries - but only if it sees them because of suit sensors (it doesn't necessarily understand organics completely).

 

Essentially, crippling the AI by rendering them blind and dumb.

 

2) The AI is vague in its reporting, requiring back and forth to coax more information out of it. "Crew member Douglass seems to be injured in arrivals maintenance" might be how it reports that a changeling has pulled someone into maintenance and started absorbing them. If asked "how was he injured?" it might say "Another crew member was involved." and finally reveal who has attacked Douglass if asked "Which crew member?".

 

Essentially, forcing people to roleplay AI a certain way.

 

3) The AI awaits commands to bolt doors, except for in clear, immediate danger-to-crew situations. A known murderer is trying to hunt down the RD and there's one door between them? The AI might bolt that. The AI does NOT lock down antagonists or report their every move without input. It must be queried for that information, and told by a crew member to lock an antagonist on the run down.

 

Essentially, removing any sense of initiative and only permitting the AI to react under explicit circumstances in which they must be told to do their job before doing it.

 

4) Generally, the AI player should respect the game mode and allow a generous, sporting chance for antagonists to work some conflict into the station before spoiling the roundtype and/or bringing about the conclusion of the antagonist's round via bolted doors and divulged information to security/command... much in the same way that antagonists are to RP and promote enjoyment of the round by choosing gimmicks that are more involving than "murder everybody".

 

Basically this entire paragraph is what you'd see from an antagonist that just got caught and locked down by an AI and then arrested by security because the antagonist completely ignored the power of a station role that can see mapwide into every area that has a camera and can interface with most machinery in order to put organics at a disadvantage.

Nah. None of these are good solutions against validhunting, do you know what is? An active administration that bases most of its disciplinary actions from reports from the playerbase. Which we have, and goes sorely underused. The AI is meant to be strong when left unchecked. It's largely one of the first antagonist objective targets on LRP servers, primarily to card it, secondary concern is that a good AI brings incredible pressure on the map. Although they can be counterplayed fairly easy, with the right equipment.

[Brightdawn]

I mean I'd argue that the AI is probably the most balanced thing on the station. Yes it has access to a lot of things but it's biggest weakness is it's lack of any real flexibility. In most situations an AI can do maybe 3 things? without breaking it's laws any good antag can plan past that IC and basically cut them out the equation. Non AI things can adapt and shift and change. an AI while yes it starts off strong can't really do much if you play around it.

Hell even if you're not going loud would destroying camera's secretly be OOC for any of the antag roles?

[sonicgotnuked]

How about we add a AI detector? We can also add a getto AI detector that is a little more rundown but works.

[Chada1]

You know, I once had an idea... What about making those monitors that can show a face/eye/other selected images, show the image when the AI is watching and be blank when it's not?

[Scheveningen]

I honestly think other suggestions should go in another thread that isn't this one. They would probably have more effect than policy change requests.

[JKJudgeX]

Basically the problem that I draw here, is that we tell antagonists not to "play to win" but the AI is not under the same guidelines (this includes malfAI, too).

Other servers often have a few paragraphs to a few pages of rules specifically dictating some dos and don'ts regarding playing AI, many of which are aimed towards putting a bridle on AI meta/powergaming. It would stand to reason that on a HRP server (with the heaviest possible ruleset imposed upon antag behavior), some special guidelines of play would be imposed on the role with more power than any other role on the station (as counterbalance to the heavy ruleset placed on antagonists).

[Skull132]

How about we add a AI detector? We can also add a getto AI detector that is a little more rundown but works.

 

This is already a planned project that I will look into over the next month. Maybe before or after Nümap, will see.

 

Other servers often have a few paragraphs to a few pages of rules specifically dictating some dos and don'ts regarding playing AI, many of which are aimed towards putting a bridle on AI meta/powergaming. It would stand to reason that on a HRP server (with the heaviest possible ruleset imposed upon antag behavior), some special guidelines of play would be imposed on the role with more power than any other role on the station (as counterbalance to the heavy ruleset placed on antagonists).

 

It is the personal opinion of this Head Developer that those other servers have fallen folly to rules lawyers if this is the case. Or have not wished to move away from Asimov, as may be the case in TG, for example. Lots of rules to regulate something != a solution to the issue != a decent solution. Rules should be enforced by mechanics, in this case, mechanics being the AI's abilities and lawset. Most of the restricting and unrestricting should be done via those, and not via a 30 line long wiki article dictating silicon policy (ala Yogs and TG).

While I'm not saying that restricting the AI is necessarily a bad thing, they can very quickly shut down an antagonist if they discover them, I am saying that we should almost absoloutely find a way that does not entail mountains of policy about it. We have managed to avoid this specific trap for 4 years, almost half a decade. Can we please keep avoiding it?

[Scheveningen]

It stands to reason that when people witness bad roleplay or a rule infraction that said people should report it via adminhelps or complaints. As of now, I have not seen the whole of the playerbase creating such problems when playing synthetics that it requires we overhaul existing policies and to tell people to roleplay something radically different from what the server rules and the four main laws tells an AI to follow.

Validhunting is a term that is well-defined as being an extremely aggressive roleplay style that causes certain characters, when mandated by the will of their players, to seek out and put an end to the ability of the antagonist to be able to roleplay at almost any cost, generally with a priority overruling that of more sensible goals a character would have, such as surviving, having moral values, having other existing priorities that don't end up in the violent murdering of an antagonist, et cetera.

Validhunting is an OOC issue, above all, as it is OOC attitude that seeps into IC and negatively affects most or all players involved. It is a subcategory of powergaming and any suspect cases should be reported along with additional evidence supporting it. I emphasize the last point because lately we've had issues with people reporting things without showing actual proof, thus contaminating any semblance of a proper justice process.

[JKJudgeX]

Well, my original suggestion post regarding AI from a long time ago was to provide notification when the AI is watching an area, or other mechanics-based limitations to the AI, which was met with nearly identical disinterest, so, my natural next move to reduce this problem was for a simple set of behavioral guidelines for AI similar to that found in command or that found for antagonists.

I'm fine with any kind of limitation, rule-based or mechanics-based that doesn't put antagonists, and thus the majority of the interaction of the round type, at the whim or quality of the AI player.

I feel like that if we are going to dictate that a Tajaran has personality X, Unathi usually behave thusly, and IPCs are to tend to be "this way", a few simple guidelines for the AI's behavior that lead it away from playing as a human police detective with lockdown powers wouldn't be out of the question or unprecedented.

That being said, I DO actually agree with Skull and prefer mechanical changes, but, I've not seen any limiting mechanical changes suggested to AI that got any traction whatsoever... I'd like to see some, either here or in another thread (and if in another thread, please link me so I can chime in and lend support).

[NoahKirchner]

Adding roleplay guidelines and rules seems like a really poor solution, in all honesty. Especially for newer AIs, it just becomes more hoops to jump through and more things to be worried about. If anything, adding laws that would restrict that type of behavior might be better, since it is right there and law-following is already a rule.

[Azande]

Not reading all of this really - but I have an AI fully designed to be a security apparatus, the AI was put in as a balance on the crew's side to help stop antags, if the antag thinks the AI is a serious threat to it's plans, kill it or subvert it (both are startlingly easy) .

Most of the time if I find someone breaking a regulation, or committing a serious crime - I report it as an AI, it makes sense for me to do this. I give antags plenty of chances, but if they are sloppy and set off alarms and commit murder in PUBLIC areas with cameras, you deserve to be reported and locked down. Sorry not sorry.

There are lots of times that I can't even locate simple vandals or minor assault criminals - if they can escape my wrath while I'm AI, a fully fledged, armed and equipped antag should be able to, or they should be able to deal with the consequences of not being able too.

[Arrow768]

This has been answered by staff. Therefore moving it to completed