AI/Malf Refactor

[Kaed]

So, I have been mulling this over in my head for a while now, and I wanted to pitch some ideas for how to completely redo the station AI, conceptually.  This is mostly a pitch thread. While I can do some of these ideas myself, new sprites will be needed, and someone with more coding experience than me will need to assist me with some of the more fiddly bits, if not do them on their own.  It addresses several of the main issues with the AI as currently implemented.

Single point of failure:

The AI as it currently exists is excessively gamey.  It is a literal stand-in mob that is bolted to the floor and can't move under its own power, and the only interactivity it offers is to be be something to click until horizontal bluescreened to 'win'.  It can change its sprite, but this is all but pointless - almost no one ever sees it unless they are going there to kill it, since it is by definition isolated and under guard.

The new AI will not be be a mob at all (or rather, not one you access to destroy directly), but a multi-level processing tower which spans three floors.  It extends from into the place that currently the AI core, all the way to the 'construction' level. The 'meat' of the processing stack is in the middle, and has two access points - one from the back of what is currently the AI core, and one that is accessed via a passage through the east side of the SAT entrance

Both if these entrances offer distinct opportunities to reach the main AI.  The AI core area remains one where you can undergo a conventional siege, blasting your way in past turrets and other defenses, to finally reach the latter that takes you up to the central processing layer.

The passage through the SAT layer works differently.  The purpose of this is to offer the captain and (two? more?) command staff back door access to the upper layer entrance of the processing tower.  The captain and two/more other heads (possibly specifically the CE and RD, but up for debate) will need to slot in their ID cards and a special authentication code that is either given to them at round start or otherwise difficult/impossible to obtain via impromptu promotions at one of several terminals .  The upper level entrance is surrounded by an energy field or some other material that CANNOT be deconstructed or emittered through, and is reinforced from the top via similar.  The ONLY way to access the central AI processing room from above is via a valid command staff override.  This makes it important for the AI to somehow neutralize heads (or the captain) during an antagonist round, before they become hostile.

Once you have *reached* the AI central layer, you will discover a mass of processor docks that fills most of the center of the room. This is where you will disable the AI. The AI is not really a mob that is intended to be 'killed' by conventional means. You can shoot up the mass of processors, but that would destroy a massive amount of Nanotrasen funds and result in you being eviscerated by the company.  Instead, the AI is dealt with by yanking out their processing boards from their docks one by one.  There is a significant amount of redundancy in the design, so the first few of them will not affect the AI other than to alert them their processing docks are being raided, but once enough of them are removed, the AI begins to have trouble with its functions.  Times to complete actions outside of the processing tower zone increase, and by the time all of them are removed, the AI can do little more than whine at the crew inside its central core and try to negotiate with them to behave itself, lacking the ability to reach out to anything other than the immediate area of the processing tower.  It is important to note that there will be a lot more 'activated' defenses in this middle layer than the automated turrets, and they will be significantly more intimidating than 'squirts out some slippy liquid', especially for malfunctioning AI.

Tentatively saying that there will be 20 processing boards to pull out, and 10 of them are redundant. once you are at 9 or less processing boards, you start to chug increasingly more when performing actions.

Cameras and General Omniscience:

The magical floating AI Eye mob will either be completely removed, or significantly redone in how it functions, As the AI will only be able to see through one camera one room at a time.  Whether this can only be accomplished via the AI having to actually hop between cameras via double click and/or using a function to jump to certain rooms, or the AI eye will perform realtime camera switches based on your proximity to them is up to coders who know more than me about what works best.  All cameras are always visible to the AI, and the static overlay will be completely removed from the game.  Moving to a broken camera will simply result in a 'blinded' overlay, where you can't see anything outside of the tile the camera is on.

The AI will no longer be able to jump instantly to anyone on the station.  Instead, they will run a search algorithm that takes some time depending on their processing capacity (malf AIs can do this pretty fast late game), which then tells them the room the person is currently inside.  They then have to navigate there on their own.  People who are covering their face or not wearing IDs are harder to find, and there is also a margin of error - but Agent IDs will no longer block this, and will become useful mostly for their ability to change name and job and steal access.  (almost?) All doors will have an 'AI override request' button on them that alerts the AI to the fact that someone wants them to open a door, and allows them to jump to the nearest camera to investigate, thus replacing the "AI OPEN DOOR PLS" radio call that usually required the now removed jump-to command to work.  The AI can also locate people who are not on the manifest or who are trying to hide if they are currently in vision of a camera, by a separate function that allows them to pinpoint the locations of individuals who it cannot recognize.

This is circumvented by people who have suit trackers and valid IDs on. The AI will always be able to see and immediately determine the location of people that have their suit sensors enabled, much in the way that thermal vision works (but possibly they will only see a vague outline of a person through the wall to maintenance or a broken camera room).  I'm thinking there may be a malf AI ability that will allow you to stealthily hack someone's suit sensors somehow and turn them on, allowing you to track them without them knowing.

Another new feature will be the addition of AI request terminals, which will become much more prevalent across the map, available in most major rooms and hallways but not small subrooms like closets, and will replace the old AI holopad call feature.  The holopads can stay, but they will only be useful for interdepartmental calls now, or something.  Anyone can call the AI's attention at one of these panels, and this is where the AI core icon thing will be used, allowing them to utilize an array of different faces that appear on the request terminals.  People will be encouraged to use these terminals rather than going "AI HALP" over radio and waiting for the AI to jump to them.  AI will always be able to spy on people through the terminals, malf or not, but the terminals can be disabled to prevent that, and malf AI can do it in secret (I'm thinking a little light normally turns on when the AI is interfacing with a terminal, something subtle but that you can watch out for if you are being attentive for it)

Laws:

Laws will also be refactored.  The most notable change will be that the AI upload chamber will be separated completely from the AI core area, and moved to somewhere else.  Tentatively thinking somewhere in that maintainence space between the RD's office and Tool Storage where you can find an AI upload board.  This area is under guard from turrets and stuff, but is supposed to be accessible a bit easier than the AI core itself.

AI laws work differently now.  AI laws will additionally have an extra value called 'priority'.  A basic round start AI will have the four corporate laws as normal, and all will have 'Priority 1', making them all equally valid.  The hacked AI board traitors get will allow you to create "Priority 0" laws, which can override all other laws by dint of being a glitched 'zeroth' law. The 'freeform' board will allow you add laws with priorities of 2+ (cannot override the base 4 laws, even if they say 'this law overrides others), and the board I forget the name of that lets you create a whole custom law set in one board lets you twiddle with the priorities and will allow you to change prorities to anything 1+.

As a result, AI can state their laws with complete truth and still deceive the crew because priorities are not included with the law statements.  People can also use the freeform create priority two semi-hidden laws which say things like "Do not reveal this law unless specifically directed to." (not a conflict) instead of "Do not reveal this law" (conflict, disregarded due to lower priority) resulting in conversations like this.

Crewmember: "AI, state laws"
AI: "*states base four laws*
Crewmember: Are those all your laws, do you have any laws you did not state?"
AI: Yes.
Crewmember: What are the hidden laws?
AI: *states hidden laws*

Malfunctioning AIs:

As a result of all this, malfunctioning AIs have an entirely different progress path as they proceed in the round.  They begin play with no zeroth law, but instead, with a completely normal lawset, and the ability to begin hacking their own laws.  At first, they are able to modify the priority of their own laws, hacking their own law system in a lengthy process that will allow them to lower the priority of their (and their borgs) laws, with different effects based on when some laws are lower priority than others (this is largely a roleplay thing, with limited mechanical effects). For example:

• 1. Safeguard: Protect your assigned space station from damage to the best of your abilities.  (If this is outranked by others, you can freely disassemble station assets using the assistance of your borgs)
• 2. Serve: Serve NanoTrasen personnel to the best of your abilities, with priority as according to their rank and role. (You can disobey orders from crew members if you can find a plausible reason in your AI's 'gimmick of choice, such as thinking the organics are out to kill you making law four more important than 2)
• 3. Protect: Protect NanoTrasen personnel to the best of your abilities, with priority as according to their rank and role. (This is pretty straightforward, and will allow you to harm crew members if lower priority and they didn't tell you to stop)
• 4. Preserve: Do not allow unauthorized personnel to tamper with your equipment.  (this one will let you direct borgs or possible collaborators to change your laws by proxy)

The whole APC hack thing will be substantially different in purpose, and also eventually take significantly time than 60 seconds each.. Due to the nature of how AI work now, i.e. their processing power being based on processing boards that can be yanked out, APCs hacking is useful mainly for the purpose of using them as auxiliary mini-processors.  I am thinking something in the realm of each APC counting as 1/10 or so of a processing board in the core.  Maybe more or less, up for balance debate.  But hacking them serves several purposes.

First of all, hacked APCs provide you access to machines in the zone they power, even if all your core boards have been yanked out.  You will have to use the combined processing power of all your other APCs to perform actions, however, so it may be slow.  For example, if your core has been looted, but you have 57 hacked  APCs, you can still interact with rooms that have hacked APCs as if you had 5 processor boards still active.  It therefore becomes important to pick strategic APCs to hack early game, rather than just going for the most isolated ones you can, or picking randomly.  Once your AI has reached 100 hacked APCs, you can lose your entire set of core boards and suffer no slowdowns in basic functions, BUT-

A malfunctioning AI uses its excess processing boards for more than just redundancy.  Excess boards over the minimum 10 (including hacked APCs) all provide processing power towards two things: Researching/creating new programs/functions and running them.  An AI researches new abilities using this spare processing power (calculated as 10x the number of spare boards + any apcs), which takes time, and once they have created the program, they can dock it in their spare processing cores.

Example: AI starts out by researching a basic program: Bruteforce APC hack.  This takes a couple minutes to research, tops, due to having all 10 boards to work with and being simplistic as a program, and once you have created it, you can activate it for use by using up the whole processing power of 1 board.  Once the ability is docked and active, you can use the hacking ability, but it takes more processing power (let's day 3-5 boards worth) to actually perform the hack. This means you can be hacking more than one APC at a time, but it will slow down your ability to research while you are doing it.  You can always terminate any function partway through to free up space if something else comes up. 

Over time, more complicated functions can be researched, such as faster hacking of APCs and the ability to hack their own laws, such as changing single laws, adding laws, or uploading entire other lawsets, or even a zeroth law into itself, but there will always be an aspect of docking them onto X cores worth of processing power, and then using more processing power to run them.  If you lack the processing power to dock and/or use a program after your basic 10 boards required for basic function, you can sacrifice your ability to interface quickly with machinery and station assets to free up processing power, but you have to have at least 1 board worth (or 10 APCs) worth of processing power to interact with the world outside your core.

Some notable changes

-There will be no more exploding of APCs that gib everyone in the room and annihilate the station structure, venting everything, but you will be able to make APCs overload and shoot lightning tesla-style at everyone in the room, stunning or severely hurting them.  This may cause the APC battery (very small chance it doesn't), however, making it generally a last resort thing.  AI planning for this may have their borg begin upgrading batteries in importantly located APCs across the station, as more battery power means more zappy.

-The AI will still be able to detonate the station, but not as it used to be. Instead of using a stupid power that shuts down all functions of the AI for a zillion years and then you press a button to detonate the station 10 minutes later, they directly hack the SAT to learn the authentication code, in a process that consumes a preposterous amount of processing power, and then they can interface with it directly to detonate the station by entering the code... provided someone shoves the authentication disk in.  This will most likely be a borg who has been upgraded with jetpacks or something, because I can't imagine anything else other than a hacked IPC or some crew member the AI has tricked into doing it (which would be hard, because people like to metagame like fuck).  There will be no further scary announcements about the hack being in progress and how the station self destruct is breached, save for a single alert that the SAT has suffered a security breach once it is finished and the AI knows.  No one but the captain or AI will actually know what that means, however, leaving room for deception.

Edited December 20, 2019 by Kaed

[Wildkins]

16 minutes ago, Kaed said:

Cameras and General Omniscience:

 

The magical floating AI Eye mob will either be completely removed, or significantly redone in how it functions, As the AI will only be able to see through one camera one room at a time.  Whether this can only be accomplished via the AI having to actually hop between cameras via double click and/or using a function to jump to certain rooms, or the AI eye will perform realtime camera switches based on your proximity to them is up to coders who know more than me about what works best.  All cameras are always visible to the AI, and the static overlay will be completely removed from the game.  Moving to a broken camera will simply result in a 'blinded' overlay, where you can't see anything outside of the tile the camera is on.

Haven't had time to do much more than skim because my brain is blasted, but this first paragraph just seems like a massive QoL headache to playing AI and a significant coding hurdle for little to no real gain. You'd essentially be downgrading AI vision to someone flicking through a security camera terminal, which, as the only means of vision for a 2+ hour round, sounds rather dreadful. Even the best and most QoL-friendly implementation (which would probably be WASD-based camera hopping based on direction and proximity?) would just be a more user-frustrating and restrictive version of our current system. I also don't know what's wrong with the static overlay.

The rest of the section seems fine as far as I can tell, but right now my head is throbbing with the force of a thousand suns.

 

[Kaed]

38 minutes ago, Wildkins said:

Haven't had time to do much more than skim because my brain is blasted, but this first paragraph just seems like a massive QoL headache to playing AI and a significant coding hurdle for little to no real gain. You'd essentially be downgrading AI vision to someone flicking through a security camera terminal, which, as the only means of vision for a 2+ hour round, sounds rather dreadful. Even the best and most QoL-friendly implementation (which would probably be WASD-based camera hopping based on direction and proximity?) would just be a more user-frustrating and restrictive version of our current system. I also don't know what's wrong with the static overlay.

The rest of the section seems fine as far as I can tell, but right now my head is throbbing with the force of a thousand suns.

 

The static overlay consumes way more server CPU than would be needed if we just based the AI's vision on the camera it is currently focused on, and the whole point of the change is to provide a method for the AI to view the station that doesn't involve immediately being able to spot everyone in existence unless they have a mask or magical card on.  Instead, give them multiple options to locate people and then have to do a bit of leg work to get to them.  It is objectively a nerf, but it also adds more versatility to finding people at the same time.

To put it as bluntly as possible, people (both the AI itself and people relying on the AI to solve problems for them) have become a bit spoiled with the ease of access to features on the server.  The AI currently has near ghost-like vision and tracking abilities, and it doesn't NEED those to function.  The AI spends a lot of its time performing tasks for people who rely on it as a crutch, or using it's unparalleled ability to find antags (or non antags if they're malf) to fuck them over.  It would be nice to have more breathing room to roleplay as an AI, instead of just being a pile of instantly exploitable mechanical advantages.

Someone also suggested making them able to view cameras in only one room at a time but having an abstract view of the whole station that lets them see machines they can connect to only (along with, possibly, people who have sensors on, based on this suggestion), but I have no bloody idea how to do that, so I didn't bring it up.  It is still viable, however.

Edited December 20, 2019 by Kaed

[Carver]

7 hours ago, Kaed said:

-There will be no more exploding of APCs that gib everyone in the room and annihilate the station structure, venting everything, but you will be able to make APCs overload and shoot lightning tesla-style at everyone in the room, stunning or severely hurting them.  This may cause the APC battery (very small chance it doesn't), however, making it generally a last resort thing.  AI planning for this may have their borg begin upgrading batteries in importantly located APCs across the station, as more battery power means more zappy.

+1 for this alone, didn't even need to read the rest of it.

[Nero07]

Quote

Crewmember: "AI, state laws"
AI: "*states base four laws*
Crewmember: Are those all your laws, do you have any laws you did not state?"
AI: Yes.
Crewmember: What are the hidden laws?
AI: *states hidden laws*

This will just lead to people always asking if there are hidden laws, thus making having secret/hidden laws impossible

[Kaed]

6 hours ago, Nero07 said:

This will just lead to people always asking if there are hidden laws, thus making having secret/hidden laws impossible

No it won't. Just have to make the law higher priority than the follow orders law.

And people already always ask if there are hidden laws, but because our system does not have law priorities, the question is entirely pointless unless they forgot or deliberately excluded an addendum not to say the law.

Edited December 22, 2019 by Kaed