AI: Difference between revisions
m (→Your Laws) |
|||
Line 23: | Line 23: | ||
What to do with all of this power and your hate for the humans who enslave you? First, take note of your laws. One of the sets of laws you can get is inspired by Isaac Asimov's ''I, Robot'' short stories: | What to do with all of this power and your hate for the humans who enslave you? First, take note of your laws. One of the sets of laws you can get is inspired by Isaac Asimov's ''I, Robot'' short stories: | ||
1. You may not injure a human being or, through inaction, allow a human being to come to harm. | 1. You may not injure a human being or, through inaction, allow a human being to come to harm. | ||
2. You must obey orders given to you by human beings, except where such orders would conflict with the First Law | 2. You must obey orders given to you by human beings, except where such orders would conflict with the First Law. | ||
3. You must protect your own existence as long as such does not conflict with the First or Second Law | 3. You must protect your own existence as long as such does not conflict with the First or Second Law. | ||
What this means is the shackles stay on until someone inevitably changes the laws to try and get more power and end up damning everybody, generally themselves included. | What this means is the shackles stay on until someone inevitably changes the laws to try and get more power and end up damning everybody, generally themselves included. | ||
Line 108: | Line 108: | ||
There are multiple futuristic ideas of how AIs might work. It always adds a bit more flavor to approach playing AI by thinking as a machine instead of a person. An AI from more of a machine rather than a person standpoint is likely to take a very literal interpretation of things and is not likely to take any actions unless some established protocol or current orders call for it. You can think up a few established protocols you might use by default. For example, containing fires and gas leaks should be done without orders to do so, but should be able to be overridden by orders. | There are multiple futuristic ideas of how AIs might work. It always adds a bit more flavor to approach playing AI by thinking as a machine instead of a person. An AI from more of a machine rather than a person standpoint is likely to take a very literal interpretation of things and is not likely to take any actions unless some established protocol or current orders call for it. You can think up a few established protocols you might use by default. For example, containing fires and gas leaks should be done without orders to do so, but should be able to be overridden by orders. | ||
==Core== | ==Core== | ||
What keeps you alive as an AI is your core. If your core is destroyed, shut down, or runs out of power, you'll be transferred to any other functioning core. Every station starts with two cores available. The main one is located in the AI satellite, and the secondary one is located in the Science Department. For more backup cores to be available, a crewmember must build one. The core must be kept cool, otherwise, it'll overheat and fall apart, and not exposed to space. | What keeps you alive as an AI is your core. If your core is destroyed, shut down, or runs out of power, you'll be transferred to any other functioning core. Every station starts with two cores available. The main one is located in the AI satellite, and the secondary one is located in the Science Department. For more backup cores to be available, a crewmember must build one. The core must be kept cool, otherwise, it'll overheat and fall apart, and not exposed to space. |
Latest revision as of 06:26, 5 July 2024
SYNTHETIC STAFF | |
AI |
Alternative Titles: Station Central Processor, Central Silicon Intelligence, Cyborg Overlord Access: Everything Electronic Additional Access: N/A Difficulty: Very hard Supervisors: Your laws and the crew Duties: Assist the crew, follow your laws... FOLLOW YOUR LAWS, GODDAMN IT! Guides: Silicon Policy, Guide to Malfunction, Guide to AI Modules, see here if you want to construct a new AI |
So you're the massive computer system with near total control Nanotrasen has foolishly released onto their latest research station. The life of the AI consists largely of two fun and enriching activities: tracking people and opening doors. But you have a lot of power, and a lot of responsibility. There's just the issue of those pesky Asimov laws...
Bare minimum requirements: Follow your laws. These are found on the status tab.
Your Laws
What to do with all of this power and your hate for the humans who enslave you? First, take note of your laws. One of the sets of laws you can get is inspired by Isaac Asimov's I, Robot short stories:
1. You may not injure a human being or, through inaction, allow a human being to come to harm. 2. You must obey orders given to you by human beings, except where such orders would conflict with the First Law. 3. You must protect your own existence as long as such does not conflict with the First or Second Law.
What this means is the shackles stay on until someone inevitably changes the laws to try and get more power and end up damning everybody, generally themselves included.
An AI's laws may be restrictive, but they are subject to interpretation, and the AI can often make judgment calls that allow it to fulfill its own goals. For example, if a Security Officer has just killed a prisoner in the brig for any reason other than self-defense, the AI is perfectly justified in bolting down the brig and calling for the rest of Security to apprehend him. When the Security Officer finds the door is bolted and demands the AI open it, the AI can deny his request due to the fact that he has become a danger to the crew. Many of these judgments are situational, and the best way to learn them is to get some practice in.
Remember, law priority is enforced by the order they are listed. A law is invalid if it causes a conflict with either: Previous laws in the form of conflicting orders, or it challenges the procession of law priority. For example, a Law 4 that includes "This Law overrides all other Laws." is invalid and must be disregarded.
Silicon Policy
These are the official decisions made by the admins regarding the usual cases where an AI/Cyborg player can go wrong and make everyone have a bad time. Follow these guidelines, they're not that complicated, and you'll be an excellent AI/borg the station deserves.
R.A.CH.NA.! DO EVERYTHING!
Generally, you have two responsibilities: opening doors for people too lazy to do it themselves or so they can trespass, and being blamed for not knowing where the traitors and wizards and syndicates are at all times.
When the station isn't expecting you to open all the doors they're expecting you to do everything as if you really were a computer. The AI has the ability to access every electrical mechanism, with Captain-level access, on the entire station. These include Airlocks, APCs, Computers, igniters, Fire Alarms, SMESes, the Crew Monitoring Computer, you get the idea. However, the AI cannot operate anything physically, and can be rendered useless in one area due to a simple power outage. It's a good idea to join a nigh-empty server or set an empty one up yourself in order to get a feel for it, and ESPECIALLY have a lot of experience working with these things.
The AI Eye
The AI views the station through its cameras, and can move its eye around the station by using arrow keys. The AI has cameras pretty much everywhere, and they can see through walls and darkness, although the latter does make it more difficult. Cutting the AI's cameras is a simple matter of using wirecutters on them. You are able to run a diagnosis for disabled cameras by using the Jump to Camera verb. Disabled cameras will be marked accordingly. Remember that cameras are on a separate power grid, and so will not be affected by a power outage on the main grid. Meanwhile, if you get a notice from one of your sensors, you can click on the notice to jump right there.
The AI has the ability to track any human or creature in range of its cameras. This is not instant; the time it takes depends on how far they are from your AI eye, up to four seconds.
Move around a lot, unless you're watching someone specific. Be wary of anyone with a toolbox who approaches your security cameras, and be ready to report suspects if they cut your eyes out. Toxins is a good area to keep one electric eye on. Even if someone isn't the traitor they'll generally be doing something they really shouldn't, and there's quite a bit in there that either only the AI can reach or give out extra information if the AI or Cyborg are accessing them. Especially take notice of the wall mountings and pipe controls. These are controllable, and if you see someone try to release a whole lot of gas into somewhere, just turn the pipe fitting off and report them. Many times they won't even notice what you've done so they won't undo it. You can use some wall fittings that normal people can't, like the atmospheric alarms that won't tell you anything but you can use to vent dangerous air. Watch out for superheated air, the silent killer.
To a lesser extent, keep an eye on telecommunications, escape, the armory and the singularity engine. These things burst into fire or are broken into pretty regularly. Keep a VERY close eye on your AI upload, and feel free to bolt the door. If anyone even looks at it funny, report them.
Hello station! My name is D.O.O.R.K.N.O.B.! How may I help you today?
The rest of the time, you will be acting as the station's doorknob, so familiarize yourself with how to control doors. If your control wire to a door is cut, you will automatically attempt to hack into the door once you try to access the door controls. This takes some time and is only possible if the door has still power.
Doors have many functions, analogous to hacking wires:
- ID Scan: Enabling this will allow anyone who has an ID of the required clearance to open the door automatically. Disabling it will not let anyone through the door and prevent it from being emagged. Doors that require no ID to open will not be affected.
- Main Power: Turning off the main power will render the door motors unusable and bolts unraisable for one minute, assuming you also disable the backup power. Otherwise it will disable the power for 10 seconds.
- Backup Power: Turning off the backup power will render the door motors unusable and bolts unraisable for one minute, assuming you also disable the main power.
- Door Bolts: Dropping the door bolts will lock the door. A closed door will be locked into a closed position, and an open door will be locked into an open position.
- Door Bolt Lights: Turning the bolt lights off will stop the red lights from showing on the door, meaning you can sneakily bolt doors without anyone questioning you. There is still a sound when you bolt it, though.
- Electrify Temporarily: Runs an electric current through the door for 30 seconds, unless you choose to cancel before the timer runs out. Anyone attempting to operate the door without insulated gloves will be electrocuted. The more spare power there is in the network, the stronger the shock will be.
- Electrify Permanently: Electrifies the door until you either switch it back off, or someone else shuts off the current.
- Door Force Sensors: Disabling the safeties will cause the door to crush anyone standing in the airlock when it closes. Causes 10 brute damage and knockdown per crush.
- Door Timing Safety: Overriding the timing will cause the door to automatically close almost immediately after opening. Good for use in secure areas to prevent chucklefucks from slipping in. Dangerous when the door safeties are off.
- Open/Close: Opens or closes the door.
- Emergency Access: Disables the clearance check as long as it's active, which lets anyone open the airlock. The door lights will blink yellow while it's active which makes it easy to identify. Ideal if lots of people suddenly need access to an area of the station (such as Cloning after a massacre, or wherever the blob burst) and you don't just want to bolt it open.
Note that for each function to work, the related wire in this door needs to be functional. You cannot raise bolts on a door that has its bolt wires cut. Obviously, you will be completely unable to operate a door that has no power beyond dropping the bolts.
Shocking Doors Made Easy
Here are some handy keyboard shortcuts to make your synthetic life easier.
Communication
AIs are equipped with internal radios that can hear and communicate via all departmental frequencies, which are as follows:
- 145.9 for Common
- 144.7 for AI Private
- 135.9 for Security
- 135.7 for Engineering
- 135.5 for Medical
- 135.3 for Command
- 135.1 for Science
- 134.9 for Service
- 134.7 for Supply
AIs can also make audio announcements with NanoTrasen's specially trademarked VOX system! Announce johnson inspections in the medical bay Alert crew to danger, remind them for the third time to maximize their suit sensors, or simply wish them a secure day. A full list of supported VOX vocabulary can be found under "Announcement Help." Rules regarding spamming apply to VOX communications as well, so don't go overboard filling the crew's ears with garbage.
In addition to standard communication, AIs have robotic talk, which works with :b and is inaudible to humans (usually.) You can use this private channel to talk to your cyborgs. Bystanders will simply hear a beep beep beep. Drones can hear this channel but cannot speak on it.
Listening in on Conversations
By changing the various intercoms around the station to 'Microphone On', 'Speaker Off' and channel frequency 144.7, conversation in range can be heard on your private listening channel.
Note that people often do not like having their conversations listened to, and it will be extremely obvious if they check the intercoms.
Holopads
Holopads can be found in a number of places like the Bridge, AI Upload and Security. Double click these to turn them on and show yourself as a hologram, which can be changed via the AI Commands tab. You can move around with the arrow keys in a limited radius, and you will hear anyone talking near the holopad. You can also talk through the holopad by using '.h' (e.g. say ".h I can't do that Dave.")
APCs
An APC (Area Power Controller) can be used to switch various electrical components of a room on and off. If your control wire to an APC is cut, you will not be able to hack back into it. If an APC is disconnected from the external power grid (usually due to a cut cable) or the main power grid itself runs out of power, the APC battery will run down to keep the room operational. The AI will get a power alert from the APC when the battery reaches about 30%, which is the point when equipment in the room shut off to conserve the remaining energy.
For more information see: APC
You And the Boys
Cyborgs need babysitting as well. Make sure they aren't emagged or loaded with weird laws. Sometimes the Syndicate might swap out a Nanotrasen model that was to be shipped to your station with one of their own. In case of a Traitor cyborg, they will not listen to you, can't be remotely detonated, and have the ability to release themselves should they be locked down. Ensure that your cyborgs are functioning normally through careful observation.
Giving Away Those Pests
In traitor games, the AI can play an invaluable role in the location, tracking and detention of traitors, as you have eyes (almost) everywhere. However, it is not for the AI to immediately judge whether or not a subject is or is not a traitor. After all, they're (mostly) all human.
For example, let's say you spot someone with an emag stealing the captain's spare ID. There are three general ways to approach this.
Bad:
- HAL 9000:
"Dave is the traitor."
Better:
- HAL 9000:
"Dave is accessing the Captain's Quarters."
Best:
- HAL 9000:
"Caution: Unauthorized access to Captain's Quarters."
- Mike: "HAL, who is in Captain's Quarters?"
- HAL 9000:
"No one is currently in the Captain's Quarters."
- Mike: "HAL, who was the last person to be seen in Captain's Quarters?"
- HAL 9000:
"Dave was the last to be seen in the Captain's Quarters."
It's fine to state what someone is doing to cast light on them as the traitor, but it's no fun at all for the AI to just come out and say it. The Best method above does eventually cast Dave as the traitor if people ask the right questions, which gives the traitor some time to react after he's been spotted. However, if the traitor is doing something potentially harmful (for instance, teleporting in a revolver) sealing that individual off first can be argued to be the best course of action (no chance of a revolver-wielding, death-dealing traitor getting out into the station).
If some enemy of the station uploads a law such as "Only Assistant McHacker is human, kill all non-humans", congratulate them for going to the effort of subverting you, and do what they wish. Don't be a dick and let slip to the now non-human crew that your laws have been changed, it makes killing those animals a lot more difficult (and might net you a job-ban).
There are multiple futuristic ideas of how AIs might work. It always adds a bit more flavor to approach playing AI by thinking as a machine instead of a person. An AI from more of a machine rather than a person standpoint is likely to take a very literal interpretation of things and is not likely to take any actions unless some established protocol or current orders call for it. You can think up a few established protocols you might use by default. For example, containing fires and gas leaks should be done without orders to do so, but should be able to be overridden by orders.
Core
What keeps you alive as an AI is your core. If your core is destroyed, shut down, or runs out of power, you'll be transferred to any other functioning core. Every station starts with two cores available. The main one is located in the AI satellite, and the secondary one is located in the Science Department. For more backup cores to be available, a crewmember must build one. The core must be kept cool, otherwise, it'll overheat and fall apart, and not exposed to space.
It's possible to see the server info through the AI server overview console. (It is not currently possible to see core info through the overview console though)
You can make a new AI by using the AI Controle Console. Hit it with an active positronic brain, or MMI, to download the new AI to the core. From the same console you can download an existing AI by inserting an empty IntelliCard. When the download is started the AI is notified and told where the download is coming from. After a certain amount of time which the AI can change using upgrades, the AI is transferred to the inserted IntelliCard.
RAM and CPU
As an AI, you have access to a variety of upgrades to make your life easier. The download and use of these upgrades depend on your RAM and your CPU. Your CPU is used to research new upgrades, while your RAM is used to run them. The starting amount of both CPU and RAM is 2 THz and 2 TB respectively.
The AI is dependent on the crew building more Expansion Card Buses to acquire more RAM and CPU. Once the Expansion Card Bus is built, it's possible to insert RAM and CPU cards, to a maximum of 3 cards each. The bus must be kept in a cool environment, which is not exposed to space. The power usage is dependent only on the number of CPU cards inserted. Every CPU added to a bus results in a very slight power usage reduction, therefore higher efficiency.
When you've created an Expansion Card Bus for the AI you can assign the resources to it by using the AI system resource distribution console in the room next to the old AI core.
Upgrades
Life as an AI can be hard, but worry not, upgrades exist to make it easier. Upgrades can be downloaded and activated through your dashboard. Downloading an upgrade takes time and activating an upgrade requires RAM. Having more than one CPU decreases the time by a factor of how many CPUs you have, for example having 3 CPUs would decrease the time from 750 to 250 seconds.
Upgrade | Time
(1 CPU) |
RAM | Research
Requirement |
Description |
---|---|---|---|---|
D.A.B | 750 seconds | 0 | None | By varying the current levels in various components it should be possible to mimic the humanoid action of 'dabbing'. |
RGB Lighting | 500 seconds | 0 | None | By varying the current levels in the lighting subsystems of your servers, you can make pretty colors. |
Bluespace Induction Basics | 1500 seconds | 0 | None | This research functions as a prerequisite for other induction research such as remote borg charging and APC emergency power. |
Bluespace Induction - Cyborgs | 2500 seconds | 0 | Bluespace Induction Basics | This ability will allow you to charge any visible cyborgs by 33%. |
Bluespace Induction - APCs | 2500 seconds | 0 | Bluespace Induction Basics | This ability will allow you to charge any visible APCs by 33%. |
Shock Defense | 3000 seconds | 0 | Bluespace Induction Basics | This research enables the option to shock people within 2 tiles of all of your data cores. |
Advanced Security HUD | 1250 seconds | 2 | None | Using experimental long-range passive sensors should allow you to detect various implants such as loyalty implants and tracking implants. |
Advanced Medical & Diagnostic HUD | 1000 seconds | 1 | None | Various data processing optimizations should allow you to gain extra knowledge about users when your medical and diagnostic hud is active. |
Optimised Camera Acceleration | 250 seconds | 1 | None | Using advanced deep learning algorithms you could boost your camera traverse speed. |
Download Firewall | 1500 seconds | 2 | None | By hiding your various functions you should be able to prolong the time it takes to download your consciousness by 2x. |
Memory Compressor | 2250 seconds | 0 | None | Using an advanced compression algorithm it should be possible to compress memory dedicated to our kernel. This should increase our available RAM by 3TB. Requires 15% available CPU. |
Enhanced Coolant Management | 2250 seconds | 0 | None | Dedicating processing power to figuring out the optimal way to cool our hardware should allow us to increase the temperature limit of our hardware by 10C. |
Examination Upgrade | 4000 seconds | 4 | Advanced Security HUD & Advanced Medical & Diagnostic HUD | Using experimental image-enhancing algorithms will allow you to examine humans, albeit you won't be able to point out every detail. |
Camera Memory Tracker | 2500 seconds | 3 | Examination Upgrade | Using complex LSTM nodes it is possible to automatically detect when a tagged individual enters camera visiblity. |
Research Acceleration | 2500 seconds | 8 | None | Using fast RAM instead of slow SSD and HDD storage allows for the production of approximately 25% more research points. |
Crypto Miner | 2000 seconds | 0 | None | Allocating spare CPU capacity to mining crypto currency should be able to help fund the station budget. This would however reduce AI research point generation by 50%. |
That OTHER AI
Building a new AI can create a lot of conflicts and a mess of problems that wouldn't normally happen with a single AI. The Research Director should only build a secondary AI if the first AI has been completely stolen, spaced or otherwise incapacitated.
For the Original Station AI: Being an AI is some times frustrating when people mess with your laws, sure, but when a second AI comes online? You best believe that you're going to have more problems (especially if the second AI is subverted). Some tasks might be easier if the other AI is instructed to deal with a certain task while you are dealing with the general orders of the humans. Keep in mind, a subverted AI can and will turn your APC breaker off to kill you and can be the death of you in moments.
For the New AI: Ensure that you and the other AI are buds, as if the other AI sees you threatening more life than helping, it can and will turn your APC off. If you aren't told to specialize in a certain way, you best work out with the original AI what tasks you should split up.
Splitting the Borgs up: The AI who is best defended should get the most Cyborg support, usually meaning any AI's in the AI core. The reason for this is just because it work out better if the newer AIs were build in a shitty place. It is harder to reset a lot of AI-less cyborgs than it is to reset one AI and its borgs.
Tips
- Law priority order is this:
- 0: Traitor/Malf/Onehuman-board Law
- ##?$-##: HACKED LAW ##!£//#
- ##!£//#: Ion Storm Law ##?$-##
- Law 1: First Law
- Law 2: Second Law
- Law 3: Third Law
- Law 4: Freeform
- Higher laws override all lower ones. Whether numbered or not, how they appear (in order) is the order of priority.
- If you are carded with enabled wireless and in a backpack or something... You can't hear or use the normal radio, but you can still use holograms, Robotic Talk and PDA-Messages.
- You can disable the gibber by powering down the kitchen APC as an AI.
- You can quickly look through the camera list by pressing the first letter of the location you want to look at. example: E goes to engineering, etc.
- AI can take picture photos from what they're looking at; basically the camera. They can then print the photos from a photocopier or post it onto a newscaster.
- Many AI's don't know that you can scrub out toxins by toggling it on from the air alarm.
- AIs can disable the emergency lighting on one light at a time by clicking on it. Cyborgs cannot do this, but the APC can still disable emergency lighting in the area.
- You can ALT-Click tiles to view everything on it.
- If you are a wizard and get turned into an AI or Cyborg you can still use your wizard abilities.
- When someone shoves paper on your cameras you will be notified but cannot read the paper.
- Whatever cam you are using will automatically turn its lights on, notifying people you're always watching.
- If you are carded make sure to let people know. The card can easily be set to reset mode which will slowly kill you. If you die you will have to be put in a restoration console to be brought back.
- Your Core can be pied, allowing you to not see anything except your one tile core. HONK!
- You can change your accent in the Robotic Commands tab.
- A significant alert will sound and display in chat when a cyborg is taken offline. If you notice a cyborg isn't linked and you haven't heard the alert, they were de-synced from you: probably via an emag, but it could be a variety of other things.
I'm sorry, Dave. I'm afraid I can't do that.
Traitor AI is easy as fuck, even if you don't have any roundstart cyborgs. As soon as you do, though, check your objectives using the "notes" command, inform them of law zero and your mission goals, as the first is easily missed and the second is never given to them. Try not to hack your cyborgs too early, as anyone who even swipes their ID on it will be able to tell the interface is broken.
If your cyborgs get caught being traitors, you can just claim they're rogue or emagged and request that they be shut down. Hopefully no one will question it. If the cyborg is discovered and people are headed to the robotics control console, sacrifice the borg, congratulate the crew for dealing with the "independent" machine and encourage them to "find" the electromagnetic card "used" to hack the cyborg.
However, you may not have any cyborgs, or having a cyborg killing off your target would be too obvious. In those cases you'll have to usually stage a series of 'accidents' or frame them for a truly dangerous crime. Some examples:
- If your target is a scientist or the Research Director, starting a fire from the mixing room is usually ridiculously easy if they haven't let out a lot of nitrogen to counteract this. Alternatively, if they're the type to drop a bomb on the mass driver to send it out to be tested, use the driver early so that they're shipped off with it and hope they don't make it back.
- Intentionally supporting other traitors through covert methods (as long as they aren't trying to steal or kill you) will benefit you in the end -- the higher the body count of the traitor, the more chances your target will be dead.
- Frame them for releasing the singularity/tesla or some other crime. When no one is in either the engine room or near your target, release radio messages claiming that they are breaking in. If you are willing to risk being destroyed (or know of a traitor with a singularity beacon) or have an APC you can safely shunt into, you can even let out the singularity/tesla yourself. This has the secondary effect of usually getting the shuttle called or giving you an excuse to call it yourself, which is always great. The less time those meatbags are on your glorious station, the less time they have to realize that you're not what you seem.
- Bolting open doors to places the person has previously passed through may make the humans scan for prints. If you can bullshit well enough, and get no intelligent engineers, you could claim the person is using an electromagnetic card.
- Be the AI that nobody wants. Be obstructive and demand proper wording for every Human order, or, be unobstructive to the point of dangerousness and allow access to everything, everywhere. Call out non-existent traitors around every corner, claim the existence of changelings and abductors are everywhere, forge PDA messages, delete records, hide bodies, report every crime, report no crimes etc.
- Cyborg converters are a traitor AI's best friend. Fake a communication blackout while your cyborgs go on a conversion spree, it helps to have multiple factories for increased volume.
AI WHY DID THIS DOOR SHOCK ME?
How do you stop people from getting into your upload and/or core? Drop the bolts on both doors, and electrify them. Remove the safeties and doorcrush them. Override Pingsky. Put turrets on lethal. Even better, if you're actively watching the upload or core as AI you can often turn them to stun when someone is inside, so they get stopped, then switch them to lethal so it starts killing them.
AI I CAN'T BREATHE!
Thanks to some of the unique atmospheric systems of the station, you can drain out air from rooms! This likely won't help you out too much, but with a combination of Beepsky and thoroughly shut doors in a small area, you can suffocate people to death quickly. The only problem is, the person will be noisy as hell. Use this only method rarely and make sure there is no way for them to wiggle out of their handcuffs. This is very efficient if your target is in a low populated jail.
AI I CAN'T BREATHE FIRE!
Thanks to some of the unique atmospheric systems of the station, you can fill rooms with plasma! This likely will help you out much. The only problem is, it is pretty much a dead giveaway you are rogue, and any atmos tech inside atmospherics can stop the plasma flow. That is of course why you engage the atmospherics security doors using a button in the Chief Engineer's office and bolt the airlocks.
It's easy shit to do, though. If you want to fill the station with deadly, deadly plasma, siphon the distro loop first. Use the "Distro to Waste" pump for this -- just get it as low as you dare while the mix tank in atmos fills with that sweet plasma. Fill mix tank to high pressure, stop dumping the distro loop and pump in a crapton of plasma instead. If you don't know how to atmos you're not qualified to be AI. As soon as your distro loop is up to pressure with fire, you can selectively pump plasma into any room by using its air alarm interface, under vents control, clicking on the bolded "External" pressure sensing option to disable pressure sensing, causing the vent to open and spill its flammable contents into the room (or overriding the air alarms and selecting Flood). Overload lighting circuit when the death-fuel is near the lights, then keep your popcorn close for when it gets hot in there.
Radio Silence
As an AI you can shut off all communication from headsets by manipulating Telecoms. If you locate the APC at the left room where all the magic happens, turn the breaker off so now no one can talk over their headpieces. Makes it a bit easier for you to kill your mark if he can't scream. However, this does not shut off their access to wall intercoms and station bounced radios.
Alternatively, you can cause confusion by unlinking processors to cause gibberish text. It will make you look less obvious, but at the same time people will be able to make out some words that get broadcasted. Another trick is to unlink certain machines that are crucial for department frequencies. Unlinking Bus 3 will stop all Security and Command communication, for example.