I was chatting with someone about whether it will ever be possible to create human-equivalent AI, and the difficulty that to be able to simulate what the human mind does it's really helpful if you can first figure out what it's doing. In particular we were discussing the fact that we don't know how people compose and read stories. We do know a bit about the different methods which can be used to generate prose, but there are two main things we don't know: Why people pick particular things to write about, and what constitutes a successful story (not to be confused with individual taste in stories). So basically I wondered if you all had any comments on why you write about whatever you write about, and what you think a story has to be made out of or do to be considered successful.

Well, stories can be written about anything as long as it's written well.
There's 2 main ways stories are considered good:
1) It can be experientially good.
2) It can be interpretively good.
And of course it can (and should) be a combination of both.

1- By "experientially good", I mean that the story is a vivid experience. It is descriptive, visceral, and brings the reader into the story. A story that excels in this respect creates a setting and environment that really makes the reader feel like the events are happening around him. This format is good because it carries the reader away to a world different from their own, which is a main reason we read stories.

2- By "interpretively good", I mean that the story carries meaning beyond just the words. It is symbolic, sometimes ambiguous, and gets the reader to think about the implications of the story. A story that excels in this respect has deeper meanings and proves a point about life, or raises questions about it. This format is good because it gets the reader thinking about how this story applies to his life, and affects the reader on a personal level.

Good stories are both, although Item 1 seems to be initially more important, since it is the one that draws a crowd and keeps the readers attention. It should hopefully draw things to Item 2: presenting a deeper meaning, which seals a lasting effect, keeping itself long in the readers' mind after it is read.

So an AI that can write good stories, as far as content goes, needs to understand metaphor and symbolism, and know which deeper meanings in life that humans care about. It also needs to know good literary conventions for good writing that keeps our interest. The only real way to know these things is to have everything told to it by the programmer, so in a sense, it's not really "knowing" these aspects, it's just mixing and matching ideas algorithmically, which may or may not turn out well. I guess we'll see.

What I think up and starting writing up (though I never finish) is just what I find interesting. One thing I think alot about and find myself doing on are the old games with a not so solid good story. Take Castlevania, the recent retcons, and the fact that the story is so "base" for every game. I like adding story to th lifeless part of it.

The thing is, what I think up ends up being so different from teh original that it can almost be my original story. The ideas that makes it so though, are just from influence I see in movies, books and other video games as well as just everyday life.

But as for AI writing human equivalent stories... Its possible, if you can make one that feels and understand whats going on around it. Then leave it to roam around a while to gather data. But if it can't do that, then no... not a chance. It'll just be ai writing what its fed and thats not the same as "human" level.

Stories are like an inner monologue, we use them to make sense of and order the world in ourselves.

We can express them to others, then they can use that to order themselves round a commonality, a shared view.

We created words, to define things and convey them to others, in doing so they were conveyed to ourselves.

I wonder what the first words were, and how much like words they might seem to us today.

We created sentences, to better convey. Its here the stories probably began, and our world was born. Realities, imaginings with words.

Anyway, thats what I think stories are.

As for how make a succesfull?(or do you mean a good) story you just write about a wizard who goes to school, or you could write about what you really know and understand, and let the words flow out of you that best express your knowing and understanding to others.

By 'successful story' I do not mean a good story since that's subjective, people often disagree about what makes a story good. A successful story is that which works. The interesting question is asking 'What work does a story do?' Some kind of teaching/persuading. And also, 'Does a novel do different work or more work than a myth (which is the most basic type of story)?' Also, when trying to figure out what stories do it's very useful to look at what pieces all stories seem to have, what shapes stories often follow, because the form of stories has evolved over thousands, maybe tens of thousands of years to better carry out whatever functions they carry out.

It's been a while since I've looked at story-based artificial intelligence, so I'm a bit rusty on all the theory. I'm assuming from your last post that a "successful" story doesn't necessarily mean a great story, but one that succeeds in actually being a story (as opposed to just a mess of words).

I'm of the opinion that intelligence can be modelled as a set of pattern matching abilities. Under this model, human intelligence is just layers and layers of recognising and repeating patterns that we've learned from constantly interacting with our world throughout our lives.

Stories themselves are encoded patterns. At the lowest level there's the basics of language and grammar. At the higher level there's the tropes that hold them together. The interesting question is exactly how big a set of rules you need to make a story.

I remember there was a computer science research project called MINSTREL by Scott Turner that explored story generation with an emphasis on creative re-interpretation of the predefined rules. In MINSTREL, the system was given a bunch of fairy-tale elements and sample stories of knights, princesses, dragons, potions and various themes of rescue, deceit, love and so on, which is would then use to generate new stories by trying to extend the rule set. Some of these new interpretations worked - for example, it took the element of a potion that causes the appearance of death to deceive someone from Romeo and Juliet, and turned it into a potion that causes the appearance of a dragon to deceive someone. However, occasionally it didn't: it took the element that dragons are violent and also eat princesses, and extrapolated that because knights are also violent they may also eat princesses. The system just didn't have enough info about how those kind of stories work.

From the story grammars I've read, the main conclusion is that it's extremely difficult to write down the complete set of rules to generate a story for anything more complex than the most basic of story types. It relies far too much on the subtle understanding we have of how the real world works. (Of course for human authors, we can assume that we've got at least a basic understanding of reality!)

A story is a speech delivered from a teller to numerous readers through natural language.
Like any form of human communication, in order to be effective, it requires the teller to have a behavioral model of its audience mind. Unconsciously, authors make a lot of assumptions about their auditors : "they are human, they have that cultural background, that much technical knowledge, these political biases". The teller has in mind a situation that it finds interesting and wants to introduce this situation into its audience mind using words. It does so by testing its sentences against its behavioral model. That's the author 'feeling'.

"Sandra just hit Brad" Was that shocking ? Was that funny ? Was that sad ? The author knows the situation and this piece of information is meaningful for him but he also has a standard behavioral model of his readers and knows that they would only be puzzled by such a piece of information out of context. Let's say the context made it sad. The author has enough information to understand that. Now it also needs to know what informations are relevant to transmit this feeling. "Brad's mother name was Mary" is not relevant, "Sandra loves Brad" is. It needs a model of what causes sadness to understand how to transmit it.

But a human is a complex interpretative machine. "Sandra loves Brad. Sandra hit Brad" isn't nearly enough to transmit feelings. It only explains why the author found the situation sad but it doesn't grant that the reader won't find it funny (or downright uninteresting) Then you have to explain that Sandra didn't know that this person was Brad, that she hit him with a lethal weapon, that she was trying to save Brad and mistook him for a serial killer. Then, you begin to have a 'working' story.

Then, there's the style. But we're not here yet, are we ? :-)