Artificial Intelligence Personality

[DaFool]

What if you made a game about trying to design and teach an android? You could program her personality and potentially become attached to your creation in the process!

VERY GOOD IDEA.

and potential players can expect to be some screwball conversations (which might turn off some players if the A.I. were non-transparent)

[Taleweaver]

What if you made a game about trying to design and teach an android? You could program her personality and potentially become attached to your creation in the process!

VERY GOOD IDEA.

Which is why I'm currently writing a FRIGGIN RAISING SIM based on that idea >.< and I wanted to keep this secret too...

[mikey]

What if you made a game about trying to design and teach an android? You could program her personality and potentially become attached to your creation in the process!

VERY GOOD IDEA.
and potential players can expect to be some screwball conversations (which might turn off some players if the A.I. were non-transparent)

Hmmm, I don't know if this is similar to what monele thought when saying it would be better if it were scripted (2 posts above), but - the fundamental problem from my point of view is indeed that you only become attached to something you can't control. The AI needs to be non-transparent. It's easy to get attached to a maid in a maid-training sim when you're just giving her general orders (something like The Sims, or Black and White). But actually seeing through the process, even if you use randomizing or other tricks just isn't the same. I'm sure it looks great on paper, but I really have a lot of doubt if it would in practice.

EDIT:
Upon reading Taleweaver's response - this was the point I was making, it's a raising sim, you're directing a girl's general tendency - awaking her abilities, but you're not actually creating the girl and her potential to develop those abilities.

[DaFool]

*scratches head

I don't really see the potential problem...oh well, there's only one way to find out...

2 VERSIONS:

VERSION A: A.I. Version
VERSION B: Scripted Version (although slightly randomized)

Player can set which style of gameplay to use. Then they can vote and then we'll have our answer!

The only thing will be that the version that wins depends on whether the Writer is more competent than the AI engine or vice versa.

[monele]

whether the Writer is more competent than the AI engine or vice versa.

Is there really a point in wondering? :p

Tales weaver : did you plan an AI based one or is it just the same story concept?

DaFool : I really think you're biting a huge chunk with this AI thing... If you're really serious about this, I suggest making at least a little test. Start small on that one and try to publish a little something in less that 2 months. By then, you'll know wether it's possible or actually hellishly impossible.
In the second case, use your new knowledge to make a regular VN but based on this experience (I just hope it doesn't clash with Tales Weaver's idea... ^^;...)

[miriam]

Papillon, yes!

If the AI is playing the part of an AI then I think the human could be much more forgiving of blunders... if written right.

It would be especially useful to have a backstory that encouraged the desire to help -- for instance if she is escaping from some threatening situation in which she was damaged and needs your help rescuing the human child she is supposed to be looking after...

I can imagine any number of such backstories that would make the player want to help and forgive blunders, which could be explained as the result of damage sustained during trauma or incomplete reconstruction.

It could be made even more compelling if she became upset and depressed when she "malfunctions" and the player doesn't understand what she's said ("I'll never get to save my human at this rate. Who am I kidding? I'm just one damaged android against them."). The human would have their sympathies triggered instead of their scorn. That's part of what I meant about using the limitations of the AI to advantage.

DaFool, even though I have a heap to do (gahh!! -- I can't believe I'm offering this) I'd be interested in tinkering with such a project. Prior to this thread I would not have considered fiddling with a top-down AI, but this is getting very interesting.

mrsulu -- your post was amazing! I am still following up on the goldmine of data in it. Galatea looks really cool. The article in the "What is an agent" piece was interesting too. I've encountered a lot of bots in the 3D worlds I've visited. One helpbot used Alicebot as its program. I managed a 3D world for a while that used some very simple bots to help people. Another 3D world had several bots sprinkled about the world to give clues in solving a mystery. You would engage them in conversation and they'd mention stuff if you asked the right questions, otherwise they'd just chat.

[mrsulu]

Miriam: Oh, don't mind me. I just spent a long time in graduate school.

Mikey was talking about how there needs to be some hidden information in an AI. At a very high level, and I'm going to abscond with his thought and talk a little bit about the field of "believeable agents". Believeable agents is what the graphics/AI fields agree to call things like Sims or Catz. A chatbot is essentially an unbelieveable agent almost immediately, while Cookie (the host from You Don't Know Jack) is trying very hard to believeable.

Speaking of YDKJ, Jellyvision has a neat collection of rules about how they structured Cookie's commentary, and it's a great resource for text-generation. It's good reading if you're trying to design conversation interfaces.

http://www.jellyvision.com/ici/jp/jackp ... sshort.pdf

The monsters in Black and White aren't staggeringly complicated internally (a series of interlocking hand-edited decision trees, if I recall correctly), but, as Mikey points out, they were surprising and unpredictable (such as being able to accidentally get a monster in a state where it can only poop if it first finds a boulder and tosses it over its right shoulder). It's this goofiness that gives them personality. (In fact, players complain that the monsters lose their personalities once they become well-trained for helpful play---they become a faceless cog in the war machine.)

So, if you have an agent in your game that has lots of hidden state (emotions about others, needs, desires), the player can have fun determining those needs and desires by observing the character interact with an environment and then deducing what makes them tick.

Then, the player can apply torque to the character and see how it makes them change, and those changes will be rich and personality-driven because of this hidden information.

In the long run, this hints to me that the "camera", the level of abstraction in such a game, would be very distant, very abstract, so that players will fill in gaps in the narrative with their own ideas.

The Sims works this way---you hear the characters talking to each other and their body language and tone of voice, but the words are just gibberish. The other cues are enough to let players fill in their own story for what happened, and it's extremely effective because human beings love to guess what makes people tick.

[F.I.A]

If anything, that last game I played which allows full interaction is that old Starship Titanic. However, you are interacting with robots and any hard subjects are stuffed with the game plot of that "A short circuit caused the robots to go haywired, thus losing their common 'sense'".

Currently, such features never see light due to their complexity, since the possibilities are way too vast. Thus most dating sims have those various commands(Talk, See, Touch, etc).

But if it is done... that maker will be RICH.

[miriam]

F.I.A -- I remember Starship Titanic... I still have it here. Pretty amazing game. Douglas Adams spent a couple of years (I think) working on that. It was all hard-wired, if I remember rightly, which is what made it so darned complex. That is, it was set up pretty-much like a standard IF piece.

The cool thing about using a chatterbot is that it might remove a lot of the programming complexity, while at the same time adding infinitely to the variability of the game.

DaFool -- I've been so inspired by this thread I've been writing a story which I'll send to you when I'm done. It isn't so much a story to be turned into a RenPy piece (though perhaps it could), but something to indicate how the viewer's sympathies could be used.

[absinthe]

DaFool -- It occurs to me that one way you could conveniently excuse some verbal blunders by an Alicebot is to fabricate a backstory that the AI character is foreign and is still learning English.

I wrote part of a game using a shtick like this once. In it, you're kidnapped by aliens and locked in a cell with a girl who doesn't speak your language. The goal was to learn to communicate (either teach her your language, or vice versa), work together to escape, and eventually fall in love.
It was pretty fun to write the language code, if I remember rightly, but I got bored after that was finished.

[miriam]

Sounds like a lovely story idea.
Funny how sometimes the challenge of making something can be the greatest source of fun.

[DaFool]

Looking forward to your stories!

I agree, a foreign / alien / cyborg element will be plenty of excuse to make the player forgive any weird interactions.

[Koichi]

You guys might probably want to know about AGI, which stands for "Artificial General Intelligence".
I would recommend developers to check out these links:
wikipedia: Strong AI
wikipedia: Artificial General Intelligence Research Institute (AGIRI)
AGIRI.org
Email Discussion Lists maintained by AGIRI.org

[chronoluminaire]

What if you made a game about trying to design and teach an android? You could program her personality and potentially become attached to your creation in the process!

That sound just like that bizarre Japanese Xbox game, Natural Ultimate Digital Experiment. Which sounded fascinating and engrossing, possibly disturbingly so. I recommend reading that review, as it's very interesting.

On a slightly different note, I've mused longingly several times about the concept of conversational AI, but always concluded it's way too much work and has too many holes in it at the moment.
However, something that one of the early posts reminded me of is how in Elven Relations, there are a number of places where the dialogue will change, just slightly, based on how the player's done so far. Little tweaks to the dialogue, or even just to a character's expression, that vary from one playthrough to the next. I'd put them in because I liked the idea, but I was surprised at how many positive comments on it I got. It seems it's the kind of thing that people really notice.

It'd certainly be possible for games to make more of an effort to insert these slight variations - just minor differences in how a girl reacts. It doesn't have to affect the future plot, just reflect how things are going. For example, if the plot says the girl accepts your invitation, just have her response vary from "I suppose so" through "Yeah, I guess" and "Okay" up to "Sure", "Definitely!" and "Yay!", based on the things the player's done (and/or other things like the weather).

Of course, that's nothing like the proper AI that's being discussed here, but it is a way to add variation and replay-appeal to a game without too much extra effort.

[PyTom]

1) Looking at the web site, there doesn't seem to be much there in terms of content. You'd probably have a better chance looking at chatbots, which seem to have accomplished something.

2) My gut feeling is we don't actually want AI in our games. Visual novels are a form of fiction... yes, somewhat interactive fiction, but I think that one of the advantages is that we get a writer planning the story out, and writing clever/stirring/thought-provoking/tearjerking/sexy/etc dialogue. While I guess one could have a bot come up with _story_ and _dialogue_ rather than just _conversation_, I'll point out that the number of humans who could come up with a really good story in realtime is quite small.