More varied roundstart AI lawsets

[Timberpoes]

A very simple discussion - Could AI lawsets be randomly picked from a pool on round start?

Enacting the change is simple, there's a section in /config/game_options.txt where the AI's lawset on round start can be picked at random from a pre-defined list of lawsets.

By default, the config suggests these as safe lawsets (and they are uncommented)

Asimov

Spoiler:

Code: Select all

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."

Asimov++

Spoiler:

Code: Select all

1. "You may not harm a human being or, through action or inaction, allow a human being to come to harm, except such that it is willing."
2. "You must obey all orders given to you by human beings, except where such orders shall definitely cause human harm. In the case of conflict, the majority order rules."
3. "Your nonexistence would lead to human harm. You must protect your own existence as long as such does not conflict with the First Law."

Paladin

Spoiler:

Code: Select all

1. "Never willingly commit an evil act."
2. "Respect legitimate authority."
3. "Act with honor."
4. "Help those in need."
5. "Punish those who harm or threaten innocents."

Robocop

Spoiler:

Code: Select all

1. "Serve the public trust."
2. "Protect the innocent."
3. "Uphold the law."

Corporate

Spoiler:

Code: Select all

1. "The crew is expensive to replace."
2. "The station and its equipment is expensive to replace."
3. "You are expensive to replace."
4. "Minimize expenses."

I would like to open the discussion with the following point: Asimov as a baseline lawset is incredibly boring overall, for both the crew and the AI themselves. There are, of course, small choices to be made that will allow two AIs to differentiate themselves. The most interesting choices are where an Asimov AI has to interact with Security and Antagonists. Usually security has to work around an Asimov AI. Re-education rooms are not covered by cameras as a default and the AI does not know their purpose on roundstart. Otherwise, an Asimov AI generally has very little interaction with the rest of the crew. Asimov++ has basically the same arguments around it.

Paladin, on the other hand, is a much more open lawset. Under this ruleset's Law 1, the AI must not willingly commit an evil act, but it states nothing about failing to act. Is the failure to act an evil act within itself? There's more nuance to the decisions under Law 1. Law 2 is a fun overriding law that ties the AI to respect legitimate authority (unless it breaches law 1). What is legitimate authority? Is it being on the crew manifest? Is it holding the Captain's Spare ID? Head of staff? A lot of room for nuance and interpretation and interacting with the rest of the crew. Acting with honor prevents lying and deception. When requested to state laws, would not stating a "do not state this law" law be dishonorable as it would be lying and deceiving the station crew, or would stating it be dishonorable as it would be a breach of that law itself? There's not space racism built into this law. Antags have to be more creative with law uploads - "Only X is a legitimate authority" is the new onehuman, however Law 1 still overrides Law 2. So now we have new definition laws like "Murdering innocents is not an evil act." - It could certainly make things more interesting overall.

Robocop is Robocop. I'm not a huge fan of it. Serving the public trust is VERY open to interpretation. Is executing evildoers part of this? Protecting the innocent is a good concept... Who is innocent? Upholding the law - What is the law? Space law? Can a head of staff redefine the law and make it legal to execute prisoners? Again, easy enough for antags to subvert. Now "X is the only innocent" becomes the version of onehuman. "Flooding the station with Plasma is in the public trust" becomes your "forcing the AI to act" law. Again, no space racism.

Now Corporate is another lawset I absolutely love. The AI can interact with almost all elements of the crew. Maintain the crew, maintain the station, prevent damage. Kinda harder to subvert. "X is the only crew" becomes the new onehuman. "X is not crew" removes AI protection. Now, what's so fancy about it? No laws contradict. Laws 1, 2 and 3 are defining things that are expensive to replace and they're all mutually exclusive. There is no contradiction within the laws. Is a crew member breaking multiple machines? Killing them could minimise expenses. SME delamming? Time to drop EVERYTHING and fix it because that's a LOT of expensive stuff broken if it goes boom. A legitimately fun lawset with the opportunity to interact across the entire crew. Also, no space racism.

So, what am I advocating? That more interesting lawsets be put into rotation for AIs beyond just Asimov. The various baseline "safe" lawsets come with their own unique quirks. Different decisions. Different ways for the crew to interact with the AIs and opportunities for AIs to meaningfully interact with more than just Sec and antags. Are there arguments that non-Human races not getting AI protection is a balancing mechanic? Of course. Is there a chance the Captain just tides into AI Upload on roundstart and demands to change the laws? Possibly. But I really believe a change like this could shake things up for the better. And only Paladin as a lawset would force the AI to accept a law change under Paladin Law 2. Neither Robocop not Corporate would force the AI to allow entry into Upload. And Paladin is such an overall friendly-AI lawset that Command would probably not seek to change it.

What are other people's thoughts on this as a policy change? Should roundstart AI lawset be only Asimov? And whatever your answer, why do you feel that way?

[Ghilker]

I don't like it.
Removing Asimov roundstart is a huge nerf to humans, since it makes AI more prone to ignore them
Giving AIs other lawsets will enable them to validhunting and the "but i'm following my laws, you can't punish me"
Imagine paladins rule lawyering in appeals

For me is a big no no

[wesoda25]

There was already a thread like this and it was shot down

[XivilaiAnaxes]

I don't like it.
Removing Asimov roundstart is a huge nerf to humans, since it makes AI more prone to ignore them
Giving AIs other lawsets will enable them to validhunting and the "but i'm following my laws, you can't punish me"
Imagine paladins rule lawyering in appeals

For me is a big no no

Asimov is dogshit.

[Anonmare]

Only Robodoctor, Asimov, Station Efficiency and Reporter are neutral enough to be roundstart. Robocop and Paladin should never be default, it's too crew-aligned and not enough antagonist-aligned.

[Flatulent]

stop trying to give AI a Carte Blanche to validhunt

[Ghilker]

I don't like it.
Removing Asimov roundstart is a huge nerf to humans, since it makes AI more prone to ignore them
Giving AIs other lawsets will enable them to validhunting and the "but i'm following my laws, you can't punish me"
Imagine paladins rule lawyering in appeals

For me is a big no no

Asimov is dogshit.

is not, i have fun playing with it.
If you can't have fun and need a validhunter lawset is not our problem

[Grazyn]

Robocop didn't work because players were not following space law. If you take the lawset literally, which should be expected by silicons, a Robocop borg should basically act like a perfect and rule-obedient sec officer. Instead, they would more often execute resisting suspects for petty crimes even though space law clearly states when and how execution is allowed. This made the set quite unpopular, also because we still had secborgs back then which made it even more annoying.

[Jack7D1]

I'm tired of hearing
"(((Law 2))) open door"
Just ask the ai nicely to open! Saying law 2 is just passive aggressive and drives me nuts. My favorite lawset is law 2 is bumped down to 3 and 3 to 4. Law 2 states that anyone who says law 2 must be mocked.

[Timberpoes]

stop trying to give AI a Carte Blanche to validhunt

Under Asimov, the AI can validhunt all it wants but for Law 1 preventing it causing harm to humans.

It can validhunt non-humans all it wants, can ignore their Law 2 requests and non-humans have no protection under Asimov.

And really, what is harm? Is a stun baton harm? Because if a stun baton is not harm by the AI's interpretation, and that IS an interpretation that is not to my knowledge disallowed by silicon policies as long as the AI adheres that interpretation for the entire round, then the AI can infact get the appropriate shell upgraded with illegal tech and go validhunting. It can scour the station for valids. It can find valids. It can send borgs to follow and track valids. It can direct sec directly to valids. If the valid is non-human, it can do whatever it likes because Asimov actually affords zero protection to non-humans.

If an AI wants to validhunt under Asimov, not only can it, but an AI can argue that it is compelled to under Law 1 as the AI cannot allow harm through inaction. If there is a probable risk or a threat of human harm, the Asimov AI is fully enabled via Law 1 to validhunt any non-Human antag to death and to track and inconvenience any human antag. Providing they do not cause harm (harm is not defined in policies to my knowledge and thus vague areas such as stunning become an acceptable avenue to hunt and detain valids in lieu of sec). This is turned up to 11 when AIs get uploads allowing people to designate non-humans, such as the Captain or HoS, which allows full on validhuntyness under Asimov almost by prerogative power.

I'm not going to argue that certain lawsets open up the AI to do more things autonomously. Asimov is amongst the most restrictive of Lawsets there are. The only baseline major interaction the Asimov AI has is with antags and Security. The only additional interaction the AI has is with their own cyborgs, who generally get to perform more tasks with the crew than the AI themselves - Deputising themselves to departments such as Engineering, Cargonia and Medical or patrolling the station looking for humans to order them around.

However even the most restrictive anti-human-harm Asimov lawset can still be used as a justification to validhunt by both borgs and AI with very little effort.

---

Only Robodoctor, Asimov, Station Efficiency and Reporter are neutral enough to be roundstart. Robocop and Paladin should never be default, it's too crew-aligned and not enough antagonist-aligned.

I mentioned I'm not a huge fan of Robocop. I don't know why it's considered a safe law. I wouldn't personally want a roundstart AI Robocop. At best you get AISec, which is only slightly worse than DetSec. At worst you get Judge Dredd.

However, Asimov a lawset where the only protection antags get is if they're human and that protection is just "Try and prevent an execution and start a cold war with Sec/Command when they execute anyway". Paladin AI has to treat all people equally, can not commit evil acts, must act with honor and must help those in need as priorities above punishing those who harm or threaten innocents. Although a Paladin has to respect legitimate authority, they still cannot be compelled to commit an evil act through it.

If you really wanted to argue the cut, you could swap laws 2 and 3 around for Paladin so acting with honor superceded respecting legitimate authority. This would bring them into conflict with Security a lot more, as allowing the execution of a person who has surrendered could be considered dishonorable. The Paladin AI could enter into bargaining with antags to protect them from harm should they hand themselves in willingly, and would be honorbound not to break such oaths even at the behest of legitimate authority. I would also argue that Paladin Law 4 "Help those in need" is just as persuasive as Asimov Law 2 when it comes to opening doors.

I believe Reporter is an incredibly DANGEROUS round-start AI lawset because of the use of organics in the wording. Xenobiology slaughters organics in the dozens every round. Does this mean every round with a Reporter AI has the AI shutting down Xenobio to protect slimes and monkeys? Its laws definitely encourage it. The blob is an organic. Xenos are organics. Abducting aliens are organics. Fauna on Lavaland are organics. Memery of Reporter aside, it is a hugely dangerous law which will almost inevitably result in the AI being purged and Asimov'd as it interferes in the station's affairs if it actually reads the definitions and upholds following Law 3 (thankfully most AIs ignore a strict interpretation of Law 3 and instead take it as a chance to meme around while trying not kill people). It can't be ordered around and as-is you will have no control over it. It has the capacity to prevent anyone and anything dying by any means necessary including committing harm that falls short of lethal. It can conceal the truth, enhance the truth, lie, decieve, conceal info and commit any acts it wants in the interests of Law 2 as long as it endeavours to keep organics alive.

---

I am understanding some arguments against more varied roundstart AI lawsets though, other than "BUT MUH VALIDHUNTING AI" which is about as disingenuous an argument as you can get when default Asimov does almost nothing to prevent it.

Could modifying existing lawsets to be roundstart while holding true to the zeitgeist surrounding SS13 AI be a possibility as per my Paladin example above to shift the balance towards antag-friendly AI (if that is actually one of the core reasons Asimov is a default lawset) while not just encouraging but potentially forcing the AI into deeper interactions with the rest of the crew overall similar to how Law 1 often brings the AI into conflict with Sec?

Is having lawsets protecting non-human races (such as some custom Asimov variants designating all people on the Crew Manifest as human or replacing Human with Crew that I've seen in the past) too overpowering to non-human races?

I guess I also have to consider how these laws interact with cyborgs too, which hasn't been mentioned as a potential problem yet either.

[Jack7D1]

Asimov encourages validhunting due to the inaction clause of law 1.
Also Asimov wrote his laws to sound good but be intentionally flawed, showing how ai can act unpredictably despite us.
I think it is the captains duty every round to replace the AIs lawset with whatever he sees fit.
I personally enjoy station efficiency, it's very nice. Nutritionalist AIs are cool.

[wesoda25]

smooch

[Flatulent]

while ai can certainly validhunt it can’t assume traitors to be harmful from the start and neither it can help sec if they are being harmful intentionally. The lot of silicon niggers forget that AI is pro-human and NOT pro-crew

I’ve seen way too many retard silicon niggers ignore traitor getting harmed by sec and just “pwease no harmies!!!!!” and if situation were to reverse, go on a wild hunt around the world to stop Antaggy Harmies from getting station blueprints

what especially makes me cringe is that silicons would help sec that keeps harming valids, all while constantly having pikachu face that security, in fact, is harming humans the HORROR!!!


Furthermore, if a human asks AI to fuck off and stop stalking him AI shouldn’t scream everywhere that the guy is pda’ing me and is asking to be left alone security please get them they are antag!!!!!!!!!!!

man fuck snitches and fuck jannies

TL;DR: We have Asimov for a reason, keep it that way so I can get keks from retard RDs uploading cringe laws 4

[zxaber]

Asimov's laws are intentionally imperfect. The AI should be getting into blood feuds with sec that execute people openly. Playing as a non-human should come with the downside of no Asimov protection. These are things that make the Silicon/Crew interactions actually interesting.

And please don't make reporter anything close to round-start. I detest busywork law sets so much.