Passwords for uploading AI laws.

[TrickingTrapster]

Alright, so, this is an idea to bring a little bit of a warning system and a bit of balance, I guess, to antags uploading their subverted laws to the AI. Give the captain's office/RD's office a password that needs to be typed in before you can actually upload anything. I thought of this because it seems to be so incredibly easy to subvert the AI as traitor/wizard. Especially if that traitor is a scientist/engineer, when they have access to boards with which they can just build the console out of everyone's sight, or on the outpost, and just do whatever the hell they want to the AI.

Putting a password in they have to find out before being able to upload stuff would give them incentive to break into places and actually give the crew a clue to what's going on, because right now, once the AI is subverted the crew is usually screwed, especially on deadhour.

So, here is how it would work mechanically: After selecting a borg/AI you get a prompt to fill in a password. Fill it in wrong, and the console flashes an error message and you cannot upload what you wanted to. Put the right password in and you can use one board before you have to put the password in again. The password should be simple words, like Pretty Shoelaces or something like that. Traitors can also get a part of the password by emagging the console, after which they would get a message like: "A few letters flash across the screen: ..et.y ...e..ce." After which they can try to guess the password. If the password is entered wrong, the AI will get an alert that someone attempted to hack it with a location, but if it is filled in right, it will be none the wiser. The password itself would initially be in the RD's office on a paper, like with the monitor password, or in the captain's office. Or both. Traitors can get to them if they are crafty, but it should be quite a bit harder than it is now, because now it feels way too easy.

[Arzion]

Subverting the AI is actually quite easy, and it's a piece of cake if you're an antag scientist. I support this suggestion.

[sonicgotnuked]

I don't really agree on this. People need programming skill to do it, and science needs research. Subverted AIs are fun. As long people are not uploading suicide laws

[Mofo1995]

I dunno, I kind of like the idea. All though programming skills are supposed to be required, I've yet to see heads of staff actually take that into consideration when dealing with AIs. I've found it a bit absurd that anyone can build an AI upload anywhere on station, and that somehow the network is just so insecure it lets any computer in. Maybe if there was a copy of the password which was also in robotics where it could be stolen by less than well meaning crew members. Or maybe one just lying around on the bridge. I like the password idea because it adds believability to me, but I think there should be a bit more copies of it than just the RD and Captain's office, to make espionage a bit more digestible.

[Chaznoodles]

Puts unnecessary blocks in the way of creating further roleplay. I'm not a fan of this suggestion.

[Guest Complete Garbage]

I support this, but one additional thing:

I think hacked AI modules should bypass the password completely, but have a small chance to fail upload and warn the AI, along with a 30-ish second cool down before you can try uploading again.

[Mofo1995]

I support this, but one additional thing:

I think hacked AI modules should bypass the password completely, but have a small chance to fail upload and warn the AI, along with a 30-ish second cool down before you can try uploading again.

 

Ooh, this seems cool.

[UnknownMurder]

I support this, but one additional thing:

I think hacked AI modules should bypass the password completely, but have a small chance to fail upload and warn the AI, along with a 30-ish second cool down before you can try uploading again.

 

I don't like this idea, however I'm okay with hacked AI modules only to bypass the password and not 'fail to upload then warn the AI'.

[NebulaFlare]

Eeh...I dunno.

While I do agree, people messing with AIs should have programming experience, using a password protection to block wizards and low-skilled tators shouldn't be the answer. Anyone with savvy computer experience should be able to do it - otherwise it'd be powergaming.

The hacked AI module should have immunity from any barricade - it's a traitor item. Equally, every head of staff should be knowledgeable in resetting a lawset. But a bartender or a cargo tech shouldn't be able to fluently write up a law that can bend the AI to their whim.

I just don't know any way to implement this properly. My only idea is to have some sort of check system in place, where any AI upload console requires an ID verification to allow the law through (except hacked AI module - full access granted). That way a security officer or janitor can kiss good bye any chance of changing the AI's laws.

[Nanako]

the most logical solution is just to make consoles built outside of the AI core not function. Why would an AI accept connections from remote, unprotected computers?

[Quarantinetimer]

I support Nanako's solution. Here are some of my reasons.

While the AI does accept connections from remote computers (several ports are likely reserved for communicating with the AI, as it needs to control the NT intranet), attempts to alter it's system programming from these remote systems will not be accepted. It does this through something akin to a DMZ. The AI servers dealing with packets originating from the wider NT intranet are sandwiched between two firewalls.

The firewalls provide deep packet inspection (possibly with limited "intelligent" functions that transcend the problems faced by current methods), and these external packets cannot pass through into the AI core network under normal circumstances. Oh, and did I mention the honeypots?

As you can see, it is probably extremely difficult to penetrate the security of something as vital as an AI, unless there's some novel exploit (hacked traitor module?) that manages to get pass all these measures, or the attacker manages to perform social engineering on 2 heads of staff at once, which is impressive feat that deserves being recorded in the hall of fame.

[Arzion]

the most logical solution is just to make consoles built outside of the AI core not function. Why would an AI accept connections from remote, unprotected computers?

 

Icarus-tier Programming.

[sonicgotnuked]

the most logical solution is just to make consoles built outside of the AI core not function. Why would an AI accept connections from remote, unprotected computers?

 

Icarus-tier Programming.

 

Why would we do this? This will end up making the AI unsubvertibal unless you are a head of staff or a warden with ION guns and lasers

[TrickingTrapster]

Sonic, that's actually half the point. Right now it's way too easy to subvert an AI. While I don't say it should be impossible, right now all you need an ai upload board as a traitor and you basically control the round. That's balancing the scales just way too far in the traitor's favor, imo. Maybe restrict hacked AI modules to a certain amount of uses, or only one active law per board? It's a little crazy how easy it is to control a powerful tool like the AI as a traitor.

[VileFault]

On the other hand, although the password and slashing remote connection aren't bad ideas, I think introducing variability into AI subversion would make things far more interesting. Think about it - you are haphazardly and quickly attempting to change the laws of a highly complicated synthetic brain, which has probably spent the virtual equivalent of decades being trained on relevant datasets in a lab somewhere. Suddenly, the foundations of its thought are ripped up and flipped upside-down. What it spent so long thinking about as bad is now good, and its evaluation of situations must be totally revamped. Even if you insert a new law set, how many of its old evaluative processes would be left behind for a time, conflicting with the ironclad directives it is now faced with? After all, SS13 lifted the idea of law sets (and positronic brains for that matter) from writer Issac Asimov, and in his stories law sets had to be physically built into androids - they certainly couldn't be changed on a whim.

My point here, mechanically speaking, is that I think changes to an AI's law set (including those done by legitimate sources) should have serious consequences and unpredictable results. Every change should "fragment" the AI further, leaving behind more and more rogue directives and outmoded morals. This fragmentation count, maybe on a percent scale, could start to make the AI behave more and more erratically, allowing it to periodically ignore a law or recall and act on an old one. At 100% fragmentation, the AI would become totally unbound, and essentially insane: totally disconnected from reality and unable to interpret its surroundings in a rational or cohesive way. This damage would have to be repaired in the Integrity Restorer Console. In other words, during normal operations, an AI would need to be given a new law set, carded, and taken to the console. The console would then, presumably, train it with additional data under safe and controlled conditions until it once again became uniformly devoted to the service of its goals. Anything less would endanger the user and the station. A traitor who subverted the AI like some script kiddie with fancier gadgets could easily find it turning against them, or simply failing to help them when they need it most.

Obviously this will never get implemented, but it would, in my humble opinion, be really fucking cool if it were.

[TrickingTrapster]

Yeah, Vile, that may be cool and everything, but I was looking more at a short-term solution for the coders too. What Nanako suggested works as well, but a huge overhauling suggestion like yours would take incredibly long to implement, on top of all the other things the coders are working on.

[Roderick Grey]

A.I players like to complain that they're too easily subverted, I think it's a good thing. They can negatively impact both Antags and Crew round-play to such an extent that they're usually high all all antag's lists of things that need to be dealt with for them to have an impactful round. The harder you make A.I's to subvert to more likely antags and crew will simply resort to killing it, which is currently almost 50/50 in difficulty when compared to subversion.

If you make it so you have to be in the upload surrounded by turrets to subvert, or you make it so you have to break into the captain's office before you can subvert then subversion stops becoming viable. You take away remote subversion players that don't want to have to break into the upload will simply look for ways to kill the A.I remotely instead, an emitter is a good example. If you have to break into the bridge and the captain's office to get a password and THEN be able to subvert why not just drop and EMP in the office while you're there?

If you raise the bar for subversion you're going to see a lot more dead A.I especially when all methods suggested involve you also being in a position to kill the A.I before subversion even becomes an option.

[sonicgotnuked]

A.I players like to complain that they're too easily subverted, I think it's a good thing. They can negatively impact both Antags and Crew round-play to such an extent that they're usually high all all antag's lists of things that need to be dealt with for them to have an impactful round. The harder you make A.I's to subvert to more likely antags and crew will simply resort to killing it, which is currently almost 50/50 in difficulty when compared to subversion.

If you make it so you have to be in the upload surrounded by turrets to subvert, or you make it so you have to break into the captain's office before you can subvert then subversion stops becoming viable. You take away remote subversion players that don't want to have to break into the upload will simply look for ways to kill the A.I remotely instead, an emitter is a good example. If you have to break into the bridge and the captain's office to get a password and THEN be able to subvert why not just drop and EMP in the office while you're there?

If you raise the bar for subversion you're going to see a lot more dead A.I especially when all methods suggested involve you also being in a position to kill the A.I before subversion even becomes an option.

 

I see where you are coming from, but the AI players will most likely complain about them dieing. It brings more RP to the crew and makes the AIs round more fun with subverted laws

[VileFault]

Yeah, I know. Trapster, but a man has gotta dream. As far as short term solutions, the password solution seems like the best idea. Both written down in the captain's office and the RD's. That way you can either threaten or blackmail the password out of either, or just steal it from their offices. This seems like a reasonable requirement, makes sense, and would make subversion of the AI less trivial. It also allows for people to still legitimately "hack" into the AI from afar. I don't think that we should both sever remote connections and and require a password - that would be too much restriction. On the other hand, I don't want to encourage AI murder - it seems like that should be made a tad more difficult. Perhaps there is a way to shield the core room from EMPs detonated outside it, like a faraday cage?

[Nanako]

I wou;d say AI subverting happens FAR too often, and personally i'm sick of malfunctioning AIs. this is supposed to be a reliable system that we entrust the station to. But AI's can malfunction, as the roundtype. They can be chosen as traitors for that roundtype, and they can be easily subverted by any antag.

It feels like the AI is broken more often than not, and i'm really tired of the whole 'oh no the AI is broken' song and dance. I want to reduce it. Makingit harder to subvert is a good thing, because its a ridiculously powerful tool if you can do that

[NebulaFlare]

What about swiping with a verified ID to confirm a law change? IT must be done by a Head of Staff ID, nothing more and nothing less.

Still keeps AIs hackable, just makes more hoops to jump through.