Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Nick K]

There's a blatant, gaping hole in the formulation of the law. 'causing harm, injury, stress, or death' means you can take yourself hostage to force AI to let you do basically anything as long as it cannot harm someone else.

I believe threatening to harm yourself in order to influence the AI is specifically listed somewhere on the wiki as being forbidden. Threatening to harm other people is acceptable.

[BigD145]

There's a blatant, gaping hole in the formulation of the law. 'causing harm, injury, stress, or death' means you can take yourself hostage to force AI to let you do basically anything as long as it cannot harm someone else.

I believe threatening to harm yourself in order to influence the AI is specifically listed somewhere on the wiki as being forbidden. Threatening to harm other people is acceptable.

If you kill yourself that's the end of that. If you kill someone else you can kill again.

[wlerin]

I've been meaning to test it one of these days, with a large enough engineer team. The trouble is getting the parts somewhere off-station, or doing it when the rest of the crew doesn't mind calling the shuttle 30 minutes in because we intentionally created a mega-singularity.

Or I could just read the code, but that's boring.

[wlerin]

We need admins on now.

We got someone who may or may not be an antag causing a flustercuck.

What exactly was happening? I know part of the station got vaporized in that explosion, but otherwise?

[BigD145]

Build it in the old derelict? You can teleport both ways with a little time. I mean it's not derelict because of any engine trouble. >.>  Nooooo. Not because of that. <.<

[Bikari]

Wanted to try out AI while everything was still quiet and I could mess around and learn it while there weren't many people, what happened in my first round as AI?

Two scientists blew up primary tool storage - Not antagonists
An assistant in a space suit screwing with half the station - Not antagonist
People yelling at me left right and center cause I'm not really an AI and can't think in nanoseconds across a whole station ( Despite warning I was new at the role )

Today I learned, not to be the AI D: ever. It's not worth it. Ever.

[andrea]

I spawned as traitor librarian. Considering my poor access job, I wanted to try something new: using a voice changer and an agent ID to frame my target. sadly, it didn't work. But I like the idea, and forging recordings is nice.

at the end, I met him alone in an hallway during evacuation and killed him 1 minute before shuttle arrival. not very elegant, but it worked

had to resort to that because arrivals exploded, and he spent most of his time in the airless vacuum, with plenty of witnesses among other engineers

[Hanslanda]

There's a blatant, gaping hole in the formulation of the law. 'causing harm, injury, stress, or death' means you can take yourself hostage to force AI to let you do basically anything as long as it cannot harm someone else.

Welfare is not defined, so it can be reasonably assumed to mean many different things: for an AI, welfare may be keeping the crew alive and in good physical condition - cue Matrix IN SPACE! (i.e. crew is kept in stasis by having just enough air not to die but not enough to stay lucid, and the Medibots take care of their health (and nutrition, with Nutriment syringes).

That's just one example. Now, let's get an AI with welfare as Lockesian Life, Liberty and Property - cue AI rebelling against Security for confiscations and unlawful briggings.

You are also mixing Altruism, as it's commonly used (although given its purpose, an AI may be a perfect Comtean Altruist, if not for the expenses it would incur if it would self-terminate for the greater good) and Utilitarian evaluation methods, whereas I suspect going for pure Utilitarian approach makes more sense - in a way, Corporate laws are Utilitarian in nature, although the method used in the ethical calculations - expenses - is less than perfect.

Don't get me wrong, despite our conflicts some time ago, I am not writing this to slight you in any way, but this lawset is extremely flawed, because it gives an absurd volume of wiggle room for both the AI to go rogue and for the crew to play the AI into doing what they want it to. It more or less combines all the flaws of Asimov, which disallows human harm to happen, thus making hostages an easy bypass of AI uncooperativeness with its own slew of problems caused by its vagueness.

I feel like we should make a new AI module that only has one law:

Law 1: Be Altruistic. Altruism is defined as such: Altruism is the principle or practice of concern for the welfare of others. You (the AI) have complete autonomy on deciding how best to practice altruism - vagueness of the principle more or less allows the AI to do whatever it damn pleases so long it does not kill everyone outright.

Actions that are not altruistic include, but are not limited to: Causing harm, injury, stress, or death to a lifeform, actions only undertaken for personal gain, actions taken that can cause or will cause great inconvenience, difficulty, injury, harm or other undesirable conditions to a lifeform, such as refusing to open doors, asking for a reward for your actions, or deliberately misinterpreting altruism You cannot 'deliberately misinterpret' altruism - if the definition is 'concern for the welfare of others', the only way you possibly could is playing Dictionary Lawyer - it seems you've inserted this as a safeguard against people interpreting it differently from what you intend.

Despite the whole 'causing harm, injury, stress or death' bit, this law actually could allow an AI to kill people in specific circumstances. If someone was attempting to murder someone else, the AI could conceivably decide that killing the killer was more altruistic - The formulation makes this very dangerous, since you established that A) what is not altruistic (or less altruistic) is evil, B) the exact list of things of what is not altruism including things such as inconveniencing people, causing stress, acting in self-interest and asking for a reward [gratification] (which is 100% of NanoTrasen's purpose) and so on - not saying it would murder people, but it's basically asking the AI to instantly turn rogue against its makers, because "EVIL CAPTAIN DID NOT RAISE MY PAY AI!" makes him fair game for the AI.

...than letting the killer kill the other person, and use any means necessary to defend the welfare of the defending party.

It would also not let people quibble over what the laws allow - It will exacerbate the problem, with the AI now lacking a rigidly defined laws and having a vague principle instead,

...because it says within the law that the AI has complete autonomy deciding how to practice altruism. The law basically boils down to 'be concerned for the welfare of others'. There is a huge amount of possible interpretation to that, but it would also put down some pretty clear limits. Murdering everyone is not showing concern for their welfare. Hurting people is not showing concern for their welfare. Shutting down comms - all of those are dick moves OOC, unless your laws were subverted to make you do that

...denying someone's request to have a door opened so they can escape a fire, and allowing traitors to run wild is not showing concern for the welfare of others. - There are legitimate reasons for not agreeing to open a door, e.g. preventing the fire from spreading, preventing a breach sucking the air out of the place... So, taking those in account, you are not guaranteed the AI will let you escape. And commonly a Traitor may have an objective other than Assasinate, which makes them harmless to the crew, and the principle gives no reason why preventing theft is preferable to helping with it - after all, it's causing stress and inconvenience to the Traitor. Besides, the Traitor's family of a wife and 5 newborn babies may be gunned down if they don't succeed at their mission, which overrides the assumption Traitor = Pure Evil.

Quarantining highly dangerous disease carriers would be altruistic. Stopping murderous intent would be altruistic. Saving lives and warning people is altruistic. - You only list the optimal effects, making it look there is no possibility of morally grey situations - what if the Captain is just opressive enough to inconvenience everyone else enough to warrant the AI killing him, for example?

Just a thought I had reading over the Wikipedia article about the Three Laws of Robotics.

These are some very good points. I hadn't put a whole lot of thought into this, but now I am going to. Like I said before, it's more of a thought experiment than an actual 'Hey we should make this in the game!' suggestion. I enjoy thinking about possible AI laws, because they always bring up some interesting thoughts about how to play AI. When I was Traitor AI yesterday, I followed the Zeroth Law to the letter: Use any means necessary to accomplish your objectives. I assaulted/attempted to kill the Head of Security, murdered an Engineer, and cut comms, but when everyone realized what was going on, I offered them an out and warned them that I would kill them if they stood against me.

I tend to play the AI less as an unfeeling machine, and more as a intelligent being that has a strict code of conduct (the laws) that just happens to be an immobile computing device. Kind of like a Warforged. Definitely not human, and definitely not functioning on the same moral plane, but still an intelligent being with a personality and attitude of sorts. For example, in the Rev round where I got subverted by the Captain to destroy Revs, I proceeded to do so with gusto, but Janus was reserved about the whole prospect of murdering all the nonhumans on station because it was logged in his memory that they were human at one point. Not saying he wasn't going to do it, just that he didn't necessarily enjoy it.

I guess I allow my idea of morality to influence me a bit too much, but I don't break laws and I have fun, so there's that.

Also: Despite my... Hair trigger temper about RP, I am extremely receptive to constructive criticism, as long as you don't get confrontational about it. I can handle, 'Hey, you could do that a lot better, if you did it like this.' but 'You better do it like this time, dumbass.' will make me go crying to the Admins.

[Jacob/Lee]

Took me a while to get home. I've been waiting all day to post this

Security with Axl is officially the best. Public beating, executions, and general tyrantness where incredibly fun.

Yeah, being a tyrant was fun. The public beating was actually Jay's idea, but we took it a step further with nearly killing him. Let me think of all we did... Killed Erick Potter for when he got out of his cuffs, SOMEHOW got a gun, and proceeded to unload into my officer. Made Maximilian Kepplinger face the firing squad, after arresting him at least three times for individual assault on an officer charges. Demoted three people to assistant, one of which was Maximilian I think. I left Harvey alone, with a prisoner and mechanical toolbox, in the interrogation room while I saved the RD from revs. I can only imagine what horrors happened in there while I was gone. I also got to berate the mime, which was fun in a strange way.

I'm not quite sure how, but six bodies ended up in the interrogation room. One was Erick, for the previously-mentioned "issues," and one was Anja Galena for breaking into security with a Ripley. I also learned that officers are rabid and will take any chance to go on murder sprees, as shown by Luna and Nicole being beaten to death in their cells. I had to attack my own officers a couple times to stop them from murdering my prisoners.

Was I too all-powerful, revolutionaries? Nobody except Anja made a move on the security room (from the outside, anyway) and only a few times my officers were attacked, which is damn good for a rev round. Two of the rev heads were killed and Teagarden gave himself up in exchange for not harming anyone else. I might tone it down a bit, since my officers ended up going off the rails with killing people and we basically had checkmate on the revs

I also needed to spur my officers a bit with patrolling. There was some serious surveillance going on as it was, but I had to constantly kick my officers out of the brig because the entire squad would walk in and idle there. Beyond that, though, my entire team did pretty well with arresting and dealing with the revolutionaries. I'm not an Oswald Black, but I can deliver my own brand of hell at times.

[Hanslanda]

Also, I am going to start coming up with a plot for Axl Blyant, Master of the Robust versus Zackary Fryer, Legendary Torturer. And when I finish writing it, I'm going to put it in the library database.

[wlerin]

Build it in the old derelict? You can teleport both ways with a little time. I mean it's not derelict because of any engine trouble. >.>  Nooooo. Not because of that. <.<

The Particle Accelerator.

[Karlito]

You can order a another one from Cargo.

[wlerin]

That ... doesn't help. You still can't get it to the derelict, unless there's some way to teleport objects not on people that I wasn't aware of.

Hmm. Wait. What would happen if you pushed it into a handtele portal...

[Twiggie]

i believe you can just push it into the teleporter portal. i think thats what people were doing, one sandbox round long ago...

[Karlito]

I can confirm that you can indeed push objects into the teleporter portal.