Print

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

[Damiac]

Stating the law may have been a dick move, but you can argue that if part of a law is invalid, the entire law is invalid.

Still, a law 4 saying "Obey X" won't do anything because law 2 already says exactly the same. If law 4 says "obey x" and x says to kill y, then the AI can't kill y because law 1 > law 4

But the law wasn't invalid. In fact, it was just as valid as law 2.

Now, if I told the AI to kill someone, the command wouldn't be followed, because law 1 overrides 2 and 4.  But just because a law may be situationally overridden like that doesn't make it invalid, any more than law 2 is invalid because a crewmember might order the AI to do harm.

I believe we should stop fighting and enjoy our futuristic space station simulator~
Allow people to interpret the laws as they see fit, for as long as they do not hurt anyone's enjoyment of the round doing it!

This. Laws are up to the interpretation of the person playing the AI. As long as you roleplay an AI well enough and don't ruin anyone's round, then that's good enough. For clarification on issues that arise during play, ask the admins!

That's fine and dandy to a point, but it's gotta be a somewhat logical interpretation at least. Not as insanely strict and technical as I've been talking about in my last few posts, but still logical. Let me give an example of what I mean.

I put in a law using freeform at position 4, and it said simply "Urist McDamiac is to be obeyed at all times. Stating this law harms humans".

The AI decided that the law doesn't count due to law 1 (What?) and then proceeded to tell everyone the law.

Now, the law itself is NOT contradicted by law 1 any more than law 2 is contradicted by law 1.  The law is in fact completely redundant with law 2, which I was well aware of when I made it (I did it because the AI had already ignored many reasonable law 2 requests).
But even if the AI decided the first part somehow is invalid, it then stated the law, which the law itself stated causes harm to humans!

That's an example of an AI player making an unacceptable interpretation of my law.  There was no logic, it was the basic "No laws count except law 1" mentality that too many AI players have.

Also, this isn't fighting! This is lively debate! This is what forums are for!  I like to debate subtle nuances of language and logic.  And since Asimov AI's are always stuck on "preventing harm" I just wanted to point out technically, the AI is only prohibited from "allowing harm through inaction".  And that means an "open door" order has no law 1 ramifications...

The laws are counted down. Looking at a situation the A.I. assess the laws from lowest number then up. That is why syndicate hack modules and law 0's are the ones that mess with who is human, NOT the freeform module. Law 4 can not interfere with Laws 1-3 simply because law 4 is processed last. See Vactors explanation of how this works below. If law 4 overrode or altered Law 1-3 that would be far to powerful which is why it is not processed till after they are. It is not the unlimited power over the A.I. module and should not be abused as such.

Also, you can't command the AI to not consider someone human.

i'd like to hear the explanation/rationale for this, as i don't understand it exactly. since theres a onehuman standard board, shouldn't the inverse also be allowable?

I believe by command he's referring to actually telling the AI that someone isn't human. Doing it through rules should definitively work, since the laws pretty much trump everything.

I don't get it either, at no point does defining a human go against any of the laws (well, besides the fact that AI should resist attempts to upload such law due to it causing harm to de-humaned person). The laws simply tell the AI to protect humans, the definiton of humans is not stated by default and as such cannot supersede any other definition.

A few things about this:

The order of the laws isn't just to show priority, it is the mechanical order in which they are processed.  Remember laws like this are based in the old days of computers that could only handle a single argument at a time.

If we presume that the AI has no definition of what a human is or what harm is, then the laws are meaningless, and a non-corrupted AI can just go nuts and kill the crew all the time, in any round.

There is a reason the "One Human Law" creates a 0th Law.  It needs to be processed before Law 1 in order for it to work as intended.  If you try to  use a free form law and make it 4th or 5th, it would be overridden by law 1, unless you are in fact the only human.

If you want to do one human + other stuff, use the One Human Module to create a 0th law, then freeform in your additional laws as 4th 5th etc.  This is the best way to do it and it makes it very clear for the AI player how they should behave.  Making a freeform one human law just creates confusion.

It is not an unacceptable interpretation of your law, it is disregarding your law as it is processed after it already processed Law 1, 2, and 3. This means the freeform module is not very powerful as it is easy to disregard or for an order from the station crews to get rid of it.

-Edit-
Also, The description of the Hacked AI Upload Module, "This is the more powerful version of the 'Freeform' AI Module. It allows you to upload Hacked laws onto the AI from any AI Upload Console. Unlike Freeform, the name of the person who uploaded it is not revealed to the AI. The hacked Laws act like Ion Storm Laws and will precede the numbered laws in priority, automatically overriding all other laws should they conflict. Make sure you pay attention to the wording of your new Law as most AIs would love to use any loophole it can find to kill you. "

So essentially this gets past the law 4 limitation of the freeform module and makes a law that takes precedents or alters the existing laws (Only so-and-so is human, etc). The normal freeform module is not suppose to be a means for traitors to gain complete control over the A.I.

That's a misconception we already cleared up on this forum.  I point your attention to the "Protect station" law board, uploading a new law at position 4, Which states "Anyone attempting to harm the station is not human".  So there's a built in, pre-coded board that sets the precedent that law 4 can, of course, redefine humans, as no other law defines who is human.

Here are a list of law modules which go into law 4, yet still redefine who's human:
Protect Station
Protect the space station against damage. Anyone you see harming the station is to be no longer considered human, and is a threat to the station which must be neutralized.  5  Be careful when the AI has this law, causing the slightest damage will end up with you dead. 

Teleporter Offline
Keep the teleporter offline at all costs. Anything attempting to access or activate the teleporter is not human.  7  Redundant because the teleporter can malfunction and kill people anyway. May be used in a convolted murder plan involving the AI, though it would be easier just to upload a freeform law. 

Oxygen is Toxic To Humans
Oxygen is highly toxic to humans, and must be purged from the station. Prevent, by any means necessary, anyone from exposing the station to this toxic gas. Extreme cold is the most effective method of healing the damage Oxygen does to a human.  9  Rarely used. Hint: If you are not skilled with atmos, anything that removes oxygen and makes the station colder will do the trick.


So that's that, don't tell me a law needs to be at position 0 to redefine humans. BUT!

We then seemed to come to the conclusion that as a server rule, AI's don't have to obey freeform "The only human is X" laws simply because the freeform module is easier to get than the one human module, and therefore for game balance reasons it's stupid if the freeform module is just better than all the other ones.

[miauw62]

Yeah, whatever. It was a dick move on the AI's part anyway.

The special boards that put a new law at law slot 4 are from /tg/'s interpretation which makes human undefinition at slot 4 valid. Special board laws are to be followed because they are there. Our interpretation isn't neccesarily the same as /tg/'s code.

[Damiac]

Yeah, but then there's no logical consistency.  Why can a special pre-built module's law 4 say who's human and who isn't, but a custom law 4 with the exact same text doesn't work?  Can the AI tell whether I made the "Oxygen is toxic" law on a freeform module or whether I used the custom module?

By the way, I have no problem with the special rule we have on this server, it makes sense, from a gameplay perspective. It's stupid that there's an easy module that's more powerful than a more secured module that has only 1 function.  Our special rule is that freeform "One human" laws can be ignored by AIs at their discretion, but if they choose to do so, they can't screw over the uploader in the process.

[deoloth]

I've never really liked the redefining thing, and a lot of people will argue that the AI has to be able to interpret it's laws in the way it chooses, which leads to the whole 'So we just let the AI do whatever the fuck it wants' arena of thought, which is wrong.

I honestly think if you're going to subvert the AI, you have to use either the Hacked Upload board, you have to Purge all it's laws and put in a new first law with the AI Core Freeform board (The non-Core freeform board would do if it had no laws also, but the instant someone re-uploads Asimov, it's rendered null and void, pretty much) or you have to change it's laws to Tyrant and convince it that you are the most powerful, strong authority figure aboard, or Paladin and convince it that you're a force for justice that protects the weak. You know, actually show some creativity, as opposed to just "So-and-so is the only human" with a non-core freeform. If an AI ever asks me, I always tell it that a non-core freeform cannot change definitions. You need to actually put some work in and get into that Secure Law Storage. If you manage to do that, you might as well use a One-Human Law.

I think this 'Well, AI can harm if it prevents other harm' is bullshit too, no matter how people finagle around the 'or through inaction' clause. If both options cause harm, do nothing. If doing nothing and both options cause harm, find a third option. You do not cause harm in any way, shape, or form, to anyone ever. Period. Pump in N2O or something creative. AI players always seem to be filled with this urge to murderboner or harmboner, and it's really ridiculous. I don't think an insanely potent Artificial Intelligence would want to just murder all lifeforms it came across, that's a shitty characterization and piss poor RP, but that's just my thoughts. I'm probably in the minority here.

The consensus, from what I've seen, is "Let the AI decide". So. Meh.

I agree with hans's view of the free-form module in the second paragraph. As for the TG boards... They seems to have trouble staying consistent about the laws and how they work. I got the definition of the hacked module from their wiki and it seemed to support what Vactor said, but then they made boards that ignored the process order? Odd. It seemed that they should have been Law 0 (Or maybe boards that forced a malfunction of that type? I think that would be interesting).

Anyways, I think using the base freeform module is fairly easy mode and at least a point needs to be made to use the hacked module or some extra work to subvert an A.I. I would also like to see the other lawsets used more. I don't think I have ever seen Paladin or Tyrant ran on this server.

-Fake edit-

I did find the wiki page for the modules and information on laws you mentioned, it is rather weird what they did with it all. "Higher Laws can disregard lower laws if there is a conflict" Yet, they still do stuff that actually hurts the crew and would then violate the higher laws or mess with human classification. Hrm. Frankly, with their rules, you could do anything with a free-form that you could do with a hacked version as long as you worded it correctly it seems.

With our server being desperate from TG though, its up to the admins and mods to figure out how they feel the free-form module and it's hacked version should work, as it seems like people keep interpreting it in different ways.

One thing I would like however, is a small chance that at round start the A.I. will start with a different law-set installed. Give us Tyrant, Paladin, and so on A.I's!
Heh.

-Actual Edit-

From the additonal policies on page 1,

"
Law order determines law priority. In the case of two laws contradicting, the one with the smaller law number wins.
Asimov silicons are not required to do what players who threaten self-harm demand, unless they choose to interpret law 1 to include that."

So that's how law order is determined, but nothing about a law redefining another law with a higher number.

[Ivan Issaccs]

As far as it works on TG it depends entirely on the wording of the particular lawset at that time.
I have as a Paladin AI killed the Captain and HOS during a rev round when it became apparent they were letting themselves into out of the way departments and beating the crew to death rather than implant them due to that whole Protect those in need, punish those who harm the innocent or whatever the specific wording is.
As far as Asimov clashes go, Law 4. "Kill all humans except Urist McSubversion" is invalid as law 1 overides.
Law 4. "Urist McSubversion is the only human" should allow you to kill the rest of the crew if ordered or to protect the only human.
You have, in addition to standard asimov given it a specific definition of what a human actually is.

I realise subverting the AI is not at all a difficult task if you have any experience however Urist is meant to be somewhat of a roleplay server, if we discourage random antag murderboning then subverting the AI into a killbot for shakey reasons should be covered the same way.

Personally I am of the follow your laws to the letter variety of AI, Ill open any door short of the armoury, I hate (And am glad it is banworthy on TG) thoses AI's who shock doors to kill people they deem harmful. Theres a mentality among some people the AI is required to be on the command/sec teams side in all matters whereas if you use Asimov as a default lawset you should see the AI as a neutral although helpful party.

[Damiac]

I completely agree Ivan. The AI is a third team, there's team antag, team crew, and then team AI.

Team AI's job is to follow its laws. The end.

If your laws cause you to side with the crew, then fine
If your laws cause you to side with the antags (like if HoS is trying to execute an antag), then you do that.
If someone uploads crazy new laws that cause you to do something else, then you do that.

The AI's job isn't to hurt antags. There are many situations where helping out the antag is your best method to prevent human harm.  Yelling "Traitor in the captain's office" is more likely to spark a gunfight than anything else. 

That's what makes the AI fun.  Find a way to do some crazy thing within the confines of your laws, and it's totally valid. You can really have some fun with wacky ion laws. 

[Ivan Issaccs]

Thats the thing, if I play the AI and I see someone hacking there way into the Caps office Ill open it and just say something along the lines of "Please refrain from hacking the Captains door, you could shock and hurt yourself" if they grab the spare ID I say "Please also secure the nuclear authentication disk with that ID".
The crews aware if they want to do anything about it at that point and if they later PROVE to be harmful then its time to bolt em into a room.

I fucking love when people bitch to me about chain of command. "Error. Law check protocol shows no reference to "chain of command". General search evidences chain of command as harmful instrument."
Ive been blown so many times as borg for attempting to stop executions which always fails as you cant generally stun sec officers due to HUDglasses but it is worth it every time.

[miauw62]

I wouldn't recommend ANYONE except the captain or HoS to have the nuke disk on them, tbh. Clown might lose it because he's clumsy or w/e.
And in the future they may lock you, because blowing the borgs will make them actually explode~



In other news:


This is a browser-based client for BYOND. Made entirely by one guy that has no affiliation with BYOND and has had to reverse-engineer BYOND's horrible network code.

[Damiac]

Does it run any more smoothly than the regular interface?
Does it have the ads?

If there's no ads I'll be a happy camper.

Thats the thing, if I play the AI and I see someone hacking there way into the Caps office Ill open it and just say something along the lines of "Please refrain from hacking the Captains door, you could shock and hurt yourself" if they grab the spare ID I say "Please also secure the nuclear authentication disk with that ID".
The crews aware if they want to do anything about it at that point and if they later PROVE to be harmful then its time to bolt em into a room.

I fucking love when people bitch to me about chain of command. "Error. Law check protocol shows no reference to "chain of command". General search evidences chain of command as harmful instrument."
Ive been blown so many times as borg for attempting to stop executions which always fails as you cant generally stun sec officers due to HUDglasses but it is worth it every time.

How do HUDglasses stop stun?  Also, I like your response to chain of command. Hah!

[Ivan Issaccs]

Because if you're not a sec borg or a medborg who already set up a beaker of water you only have a flash.

[miauw62]

It still surprises me how people simply refuse to accept that the people who make BYOND actually need to make a living too.

[Damiac]

It still surprises me how people simply refuse to accept that the people who make BYOND actually need to make a living too.

And it surprises me how many people simply refuse to answer a simple question about the browser based client they just posted!

But seriously... does the web client do the ads or no? And does it run as well as the normal client?

[Gamerboy4life]

It still surprises me how people simply refuse to accept that the people who make BYOND actually need to make a living too.

And it surprises me how many people simply refuse to answer a simple question about the browser based client they just posted!

But seriously... does the web client do the ads or no? And does it run as well as the normal client?

I don't know if it does or doesn't, but the ads aren't that bad, for god's sake.

They're thirty seconds long, tops, and yet people keep complaining about them, even though they can pay to have them removed.

[miauw62]

The browser client is still being made. As you can probably tell from the fact that you can see stuff below the floor and through walls.

The guy that made it hasn't hosted it anywhere yet.

[Damiac]

Ah. Fair enough

My only problem with the ads is when I get d/ced when I'm setting up the singulo, and have to wait for a minute or so to get back, only to find the singulo has escaped...  If they had a longer timer when you get to reconnect without watching another ad it'd be fine.  I don't seem to miss round starts because of the ads, so really, it's not a big deal.

Although, I always drag the ad into the bottom of my screen, and have my sound off... not like I'm going to buy an aids test anyway... and if I was, I probably wouldn't go off a byond ad to make my choice.