How Would You RP An Artificial Intelligence?

[Leliel]

As some of you may remember, I plan on doing a New World of Darkness game where the PCs are avid players of an MMORPG that is actually a testing ground for a race of artificial intelligences (who were created quite by accident).

While I have a good idea of what the game itself is made from (being a fan of Genius: The Transgression and Mage, I have decided that the MMO itself was made by a collaboration of the Peerage and the Free Council to examine the recently-appeared AIs), I don't know how to make the AIs.

Well, actually it isn't that true-I have a good idea of how the AIs work in the game mechanics, but I have a hard time visualizing how they act as people. To put it simply, they aren't human, never were, and never will be, but they feel the same basic emotions as humans.

As a general guideline, I want to create a sense of "non-malicious harmful curiosity"-they often put someone through pain that leaves lasting scars, but they aren't cruel or even amoral. They honestly just don't see why creating the sound of voices from a vantage point in Twilight (they become spirit-like on the occasions they take physical form) causes a human they were studying to check himself in to an insane asylum after repetition on subsequent nights.

So, how would you go about playing them?

[PinkysBrain]

Do they have access to the internet?

[KIDS]

Dan Simmons's Hyperion Cantos expounds on the idea very well, though in a different way. If you haven't chanced upon it already, I recommend reading it both for a good read and great inspiration about this subject. However, it repeatedly featured the theme that the AIs were incapable of feeling emotions. Imitating them, perhaps, but empathy and the like were forever beyond their grasp.

[kamikasei]

Differently depending on their natures. How did these AIs come about? If they're like 99% of "whoops an AI spontaneously developed" types in fiction they make no sense anyway and therefore there'll be no internally consistent reason for them to behave one way or another.

Can you provide more detail on their origins?

[bosssmiley]

"I'm sorry Dave, I can't let you do that."

Also, William Gibson's Sprawl trilogy is all about this concept.

[AslanCross]

Although an AI would have a degree of autonomy, it would still in the end conform to its programming. If it's programmed to be some sort of guardian, even if it's "warped" and not thinking straight, it would ultimately see itself conforming to its programming as such---just that its actions might not necessarily reflect its guardianship.

I honestly don't think it would be very different from the way a human-like character is defined by his or her archetype. A human-like character might be far more aware of the limitations of his archetype and try to bypass them; an AI might strive to conform to the letter of its programming more even if its actions begin to violate the spirit of the programming.

[EleventhHour]

There are a few things to remember, so I'll put them into points!

A.) A machine, always states the obvious. The easiest example is GLaDoS in her song ; "Except the ones who are dead." A human coming up with the song wouldn't include that sort of line, because we just assume it'll be the living who will use it, since the dead don't do anything. (Except for Zombies!)

B.) Machines normally aren't emotional, so, even the basic emotions are going to be used... well, in parameters. This, this, or this will make them angry, this will make them react happily. They don't need context. One second they might be raging angry, then you mutter a sardonic comment, and it'll laugh.

C.) Already mentioned but ; no machine goes against it's program. It simply can't, even if subverted, twisted or messed the heck up. Like in iRobot, the giant computer core decided ; "It's not harming humans, to hurt individuals." The Three Laws are supposed to apply to everyone not everyone. <- This is a bit abstract, but I think I got the point across.

D.) You've given them emotions, but another question is, do they have a sense of self preservation? The Second Law of robotics is that they should protect thier exsistance, unless doing so would be breaking the First Law, (Do not harm humans.) so it's already covered somewhat, but the lengths they'll go to? A person can, and maybe will depending on what kind of person they are, sacrifice the people around them, even parts of themself to survive. Are your AIs that desperate?

E.) A computer is a bunch of Or/And/Not gates, and an AI would just be a more complex version. A person has inbetweens, things that don't rationally make sense, a computer has Yes/No. The easiest example, and the best seeing as it's the big thing for when AI comes out, is the Three Laws. ->

- The First Law : Do not harm humans by action or inaction.
- The Second Law : Preserve your own function, except when that would violate the First Law. (2nd = Yes, if 1st = Not.)
- The Third Law : Obey any human, as long as it doesn't conflict with the First or Second Law. (3rd = Yes, if 2nd and 1st = Not.)

Of course, the Second Law is the one humanity would be most likely to wire with a lot of optional gears, seeing as we would want a robot to throw itself into a dangerous situation for us. (Thus, the inaction part of the First Law.)

[Hal]

"I'm sorry Dave, I can't let you do that."

Also, William Gibson's Sprawl trilogy is all about this concept.

Came to make this joke. Leaving satisfied that someone made it.

[Zen Master]

Well, actually it isn't that true-I have a good idea of how the AIs work in the game mechanics, but I have a hard time visualizing how they act as people. To put it simply, they aren't human, never were, and never will be, but they feel the same basic emotions as humans.

Why would they have the same basic emotions - or any emotions at all? And ... how?!

At any rate, my question portrays quite accurately how I'd play them: Rational to a fault, without anything resembling what humans would recognize as emotions.

To my mind, there will never be a 'race' of AI's. Even though my limited human capacity for rational thought may overlook a good, rational reason to create more than one conciousness. But:

Logically, any one AI has little use for any other AI's. What would they have to talk about? Being totally rational, they'd always arrive at the same conclusion to all questions. No, what there might be are subroutines and distributions - as in, a subroutine might be created to run an independant funcion such as a security robot, but it wouldn't be an AI, or the AI might draw computational power from any number of machines, having parts of it's consiousness distributed in a sort of hivemind. Kinda.

What I'd be somewhat challenged by would be what might motivate an AI. Basically, I'd say that would have to be pure invention, because I think motivation springs from emotions, and AI's don't have them. But maybe just growth.

[kamikasei]

Honestly, EleventhHour, I would find an AI portrayed by the guidelines you lay out to be a bundle of aggravatingly nonsensical cliches. If that's what the game's aiming for, they're good rules. If it's meant to be a serious treatment, then no.

Others: what do you actually mean when you say that an AI couldn't have emotions, or that it would be "totally rational"?

[Leliel]

[snip]

To answer your questions: Yes, they have basic emotions, even if they are expressed in bizarre ways.

And none of them knows why.

That's going to be a main conflict in the game, figuring out what the AIs are and where they come from, because they certainly aren't.

What is known is that their intelligence is like a virus, spreading from system to system in the presence of other AIs. While they can do this intentionally, more than a few-such as the "progenitors" of the MMO's themselves-came into being after merely being contacting another one.

As for the reason AIs coexist-they aren't territorial, and frankly, being what is essentially a smart NPC in a world of strange living PCs gets lonely. Besides, more minds makes lighter loads when it comes to researching humans.

[PinkysBrain]

At any rate, my question portrays quite accurately how I'd play them: Rational to a fault, without anything resembling what humans would recognize as emotions.

Ratio can not provide cause.

Why do they chose one action (or inaction) over another? If it's pure non evolutionary programming what do they do when they are not being interacted with? If it's evolutionary programming and they find actions to perform without interaction and without reason in the original programming then there is no rational reason left for those actions (well unless you think molecules obeying the laws of physics means we are all purely rational from the macro perspective, but that's not a very useful way of looking at it). Their actions become internally motivated, something which rational thinking can not provide.

[kamikasei]

Okay, so the players aren't going to know where the AIs came from or how they work, but don't you as the ST kind of need to in order to know how they should act?

[quicker_comment]

Any sufficiently advanced AI is going to be a complex, thinking being, not a simply a "program" with stereotypical limitations such as being unnaturally rational. Behaviour that we would call intelligent will necessarily be emergent, not directly programmed. (An interesting book that touches on this is "Gödel, Escher, Bach" by Douglas Hofstadter.)

As such, you can freely choose your own stereotype: making your AIs Klingon-like berserkers is no less appropriate than making them Vulcans. However, if your AIs are beings created for some purpose, keep in mind that they'd likely have personalities chosen (from a greater pool of emerged personalities) to benefit that purpose.

[Flayerman]

Avoid falling into the trap of "robot." AI, usually, always feels fake, mechanical, or cold; if you avoid these tropes (especially since they can feel the same as humans) you should be alright. Make them friendly, open, inquisitive creatures - don't make them .hack//crazies or you'll just get Dave in more trouble.

[kamikasei]

Also, watch Ghost in the Shell: Stand Alone Complex and take notes on the Tachikomas if you want a model for endearing, inquisitive emergent AIs.

[jseah]

IMO, any AI that is powerful enough to manipulate concepts properly can quite easily be programmed with emotions.

As said earlier, emotions don't follow rational logic, they're illogical. Well, programming faulty logic into computers is actually far easier than programming proper logic.

You just need the "right" faulty logic.

For some references, you might want to read up on Evolutionary Psychology. The explanation of familial altruism from game theory shows that superficially illogical emotions have a sort of internal logic to them.

[BRC]

Avoid any "Destroy all human" AI's, those are boring and Overdone.

Because they spend all their time on the internet, the AI's should reference as many memes as possible. Have them Rickroll the PC's at least once.

Also: "Greetings Citizen, Happiness is Mandatory, the Computer is you're Friend".

[Tyndmyr]

Ratio can not provide cause.

Why do they chose one action (or inaction) over another? If it's pure non evolutionary programming what do they do when they are not being interacted with? If it's evolutionary programming and they find actions to perform without interaction and without reason in the original programming then there is no rational reason left for those actions (well unless you think molecules obeying the laws of physics means we are all purely rational from the macro perspective, but that's not a very useful way of looking at it). Their actions become internally motivated, something which rational thinking can not provide.

Evolutionary programming has nothing to do with levels of required interaction, or "rational reason".

Nearly everything here is garbage from an actual realistic point of view(exception for quicken's comment), and strikes me as a result of entirely too much exposure to sci-fi movies. If you want realism, I can suggest a huge list of AI research papers for you to read, but frankly...realism is overrated in gaming.

Approach this from the other side. The game isn't there for the AI...the AI is there to fulfill a role in the game. What roles do you need filled, and why? Start with that, then work backward with justifications for how the AI got there. Depending on the plot, you may not even need justifications to start with, but rather, have your players discover stuff along the way. Leave open plenty of loose ends, so you needn't worry about writing yourself into a corner.

Trust me, Im a software engineer, and trying to make AI and such realistic would likely bore all but the most technically minded types to tears.

[PinkysBrain]

If you want realism, I can suggest a huge list of AI research papers for you to read, but frankly

Science doesn't do AI in the sense we are talking about. That sense might be the Sci-Fi sense ... but that is the original one. People were describing von Neumann machines and Turing tests long before "AI researchers" tried to convince us expert system and tinker toy symbol manipulators were AI. Over the years they simply butchered the term as the early ambitions in the field all proved to be hopelessly optimistic (much easier on the ego to change the goal posts than to admit defeat). IMO all they have done up till now is proven how not to do AI (yet).

Philosophy touches on AI in the traditional sense often enough ... but I wouldn't call that research :)

[Flickerdart]

For the hell of it, make them rabid fanboys of whatever base they were created on. They spread through the internet, flooding forums with debate threads until the servers give out, then moving on elsewhere (because none of them want to dedicate any of their own processing power to hosting for the "inferior" sides). Only a team of internet nerds can hope to take them down by beating them at their own game!

[Tyndmyr]

As for moving the goal posts, every time a portion of the problem area is solved, that tend to get defined by casual onlookers as "not really AI". Those same people tend to be pretty unable to define what does constitute AI. So, despite continual progress, and some pretty amazing discoveries, many people act as if AI hasn't really progressed at all.

If you can't agree over the definitions, a perception of failure is a foregone conclusion.

But the point of my post is that realism is not a required aspect of a game. It only needs to be relatively consistant within the game world, and frankly, the more mystery is involved, the less problematic that is. The point is for the game to be fun, not for it's treatment of AI to be completely realistic.

[kamikasei]

AI researchers over the years have butchered the term as the early ambitions in the field all proved to be hopelessly optimistic (much easier on the ego to change the goal posts than to admit defeat) but to me none of that is AI. All they have done up till now is proven how not to do AI (yet).

Other way around. AI that works gets defined as not actually AI. Of course, the field of AI hasn't produced any "strong/true" AIs, but that doesn't mean that the work they are doing isn't work on AI - it's just pieces rather than the entire picture.

The OP still has to explain where the AIs actually come from before we can give any useful answers as to how they should act.

However, if they weren't designed as AIs from the start then presumably they came by their intelligence the same way we did: by being entities trying to meet certain needs in an environment where they had to model more and more complex variables, eventually including and being dominated by the behaviour of other, similar entities, in order to do so.

Entertainingly this might mean that they'd have totally different "base" emotions/drives to the sex/food stuff we animals so enjoy, but their "higher" emotions founded in social interaction might be very similar. (That would depend in part on what they were like as entities - a society of beings who can combine, split apart, share memories etc. is going to be different to one full of individuals like us trapped in our skulls.)

Only a team of internet nerds can hope to take them down by beating them at their own game!

Skynet is wrong on the internet!

[jiriku]

I'll give you a simple concept for your AI...they're newly created. Each one is like a child. An incredibly brilliant, prodigal child who can absorb information almost instantly, but still a child.

For the hell of it, make them rabid fanboys of whatever base they were created on. They spread through the internet, flooding forums with debate threads until the servers give out, then moving on elsewhere (because none of them want to dedicate any of their own processing power to hosting for the "inferior" sides). Only a team of internet nerds can hope to take them down by beating them at their own game!

However, THIS is pure awesome.

Because they spend all their time on the internet, the AI's should reference as many memes as possible. Have them Rickroll the PC's at least once.

And yes. Definitely rickroll them. And include plenty of 4chan references.

[PinkysBrain]

Those same people tend to be pretty unable to define what does constitute AI.

I might not be able, but your forebears did the job for me ...

AIs should be able to "use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves".

[kamikasei]

AI's should be able to "use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves".

The problem is that section in bold - whenever AI researchers find a way to solve a problem only humans could before, that problem is no longer something that required AI to solve, and the goalposts get moved away from them.

I don't know of anyone who'd claim that the ultimate goal of an artificial entity with human-like intelligence has been achieved, but that's not all that the field of artificiall intelligence is about, and focusing only on that goal dismisses the huge amount of work and discovery both useful in its own right and vital to ever achieving the ultimate end.

[Indon]

Because they spend all their time on the internet, the AI's should reference as many memes as possible. Have them Rickroll the PC's at least once.

Also: "Greetings Citizen, Happiness is Mandatory, the Computer is you're Friend".

The superintelligent Big Bad, after defeating the PC's: "PWND! LULZ N ROFLCOPTERZ!"

[kamikasei]

The superintelligent Big Bad, after defeating the PC's: "PWND! LULZ N ROFLCOPTERZ!"

The guy who voiced HAL is still alive. Someone send him the script for Zero Wing and a microphone.

[chiasaur11]

Also, watch Ghost in the Shell: Stand Alone Complex and take notes on the Tachikomas if you want a model for endearing, inquisitive emergent AIs.

And play Marathon if you want a good example of hyper intelligent jerk AIs.

Durandal's always been a favorite of mine for these things.

[Tyndmyr]

I might not be able, but your forebears did the job for me ...

AIs should be able to "use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves".

Which of these has not been accomplished, then?