0031 - Pattern analysis

[dr pepper]

[dr pepper]

I've done that sort of thing a few times, myself. So have a few other people i've known. If you know enough stories, you begin to recognize elements of them in other stories, especially when the writers are trying to be cutesy, like sticking the word "lyre" in. That doesn't necessarily mean poor story telling, though. And even if you can predict the ending, it can still be entertaining to see how the plot gets there.

[Deepbluediver]

I'm really REALLY wondering how much of Zoa's behavior is programmed in and how much of it is via learning-and-imitation...which I guess would go a long way towards determining if he/she/it should be classified as human.

Here she's seen crossing her legs, which is the kind of thing that COULD be the hard-coded response to a social situation, or could have just been picked up from watching humans. But in comic #30 she awkwardly rubs the back of her neck in panel 5, which I don't think anyone has ever said "this is proper social behavior". Its just a nearly-involuntary reaction when people are feeling just slightly embarrassed and don't know what else to do with themselves.

[bargamer]

Alright, unless Lee's Crowning Moment of Awesome involves fighting to give all AI equal rights and all that jazz, in memory of Zoa getting scrapped for not meeting her quota because she hung out with Lee instead of sucking dick, I'm just gonna give up on MY foreshadowing skills.

[Merle]

I hate to say it, but...this is sounding more and more like a show I'd like to actually see.

[Killjoy]

If Mars is supposed to be producing entertainment, then it sounds like they're failing.

[KrenkoMobBoss]

I'm way late to the conversation (mostly because I only just started reading Forward) but I really want to discuss the things the title text is talking about.

in post-labour futures, Art is presented as one of humanity's few remaining pastimes. (I, as an artist, certainly prefer to pretend that this will be the case.) It is not impossible, though, that AI will inevitably surpass us in creative pursuits as they will in manual labour and customer service - in certain artforms and genres, if not in all of them - and I doubt that audiences will insist on only watching human-created art if AI-created art is a) cheaper, b) more plentiful, and c) engineered to be more pleasurable to consume.

I won't deny that it's "not impossible," obviously. Especially since almost nothing conceivable is totally impossible. But even if it's coming, I think we're a long long way away from that.

I've listened to a little procedurally-generated music before. Not much, but some. My impression of it has always been "It sounds incredibly generic, flat, and lifeless." It's possible this is me being harsh on the robots; it might be worth doing a blind taste test at some point. If anyone wants to set that up, let me know. But I digress.

What I do have a lot of experience on is games. I have played many many games with hand-crafted content, and many games with procedurally-generated content. Once again: the rando stuff almost universally feels lifeless. Compare the dungeons from Zelda with the caverns of Minecraft. It's no contest: Zelda's dungeons just feel more alive, more animated, more interesting, even if you're running the craziest mods in the Minecraft market.

It is entirely possible for machines to one day generate art, but I think it's a far-off dream for them to produce anything that would be recognized as meaningful. I believe you when you say an AI can fabricate a bassline and a drumbeat and a melody that sound pleasing to the ear, but I would not believe you if you said an AI had randomly generated a story. For example. Or perhaps, I can imagine an AI generating something like abstract art, making pleasing combinations of shapes and colours. I could even see one generating pinup pictures arbitrarily.

But, for example, I don't think an AI could ever generate Forward. : P

[GruntBlender]

Keep in mind that the procedurally generated stuff of today is basically result of toys. Current Markov chains and "deep learning" is to AI as Hero's aeolipile is to the locomotives of yesteryear or rocket engines of today.

Then there's the conflation between procedural generation and neural net output. Procedural generation is done with a relatively simple algorithm that uses randomness and a few predefined options to assemble something from essentially modules. A neural net learns what to make from observing a bunch of similar things (a training set). Both are currently limited in that the procedural algo has to be completely human-created and the neural nets don't understand what it is they're trying to mimic. You add a few analysis layers that could extract characters and themes from novels and you'll get a training set that could potentially lead to an actual interesting book.

There's no reason a sufficiently complicated machine couldn't generate a compelling story that makes you feel things you've never felt before, even without sapience.

[KrenkoMobBoss]

Maybe someday. But from what I'm seeing so far, AI is only capable of generating bland, generic things. I think by the time it's generating things like meaningful stories, first it's going to need to document the lives of so many actual people that it can randomly generate a hypothetical person, with life experiences and friends and enemies and influences, all the way to the point in their life when they could plausibly drawn on those experiences and write a story. I just don't see any other way it could happen, with the current technology, or with extrapolations therein.

[strange7person]

Maybe someday. But from what I'm seeing so far, AI is only capable of generating bland, generic things. I think by the time it's generating things like meaningful stories, first it's going to need to document the lives of so many actual people that it can randomly generate a hypothetical person, with life experiences and friends and enemies and influences, all the way to the point in their life when they could plausibly drawn on those experiences and write a story. I just don't see any other way it could happen, with the current technology, or with extrapolations therein.

Well, consider how it went with chess. First there's the hurdle of making a computer that can play chess at all, knowing how big the board is, which moves are legal and which aren't. Chess-playing algorithm one-point-oh is going to be absolutely pathetic, outwitted by children. Then somebody throws datasets and supercomputer cycles at it until it can think more than one move ahead. A few years later, it's competing with masters. A few years after that, trouncing them.

Current AI-generated music is bland and generic because it's based on some statistical distillation of everything that has ever been both described as music and recorded on digital media. This is the pathetic 1.0 model. Next step is scoring: sort the dataset according to, say, commercial success, or hire some test subjects and plug in some electrodes to get hard data on how humans feel about different combinations of sounds, and that intangible qualia of 'meaningfulness' will start ratcheting upward.