It is currently 20 Feb 2018, 05:39

Anything GoesCreating simple AI to help writers with guidelines (Showcase | Slideshow)

Anything, almost anything at all
(even advertisement if it's decent, spammers will be banned)
Post Reply
User avatar
tim

Administrator
View: Showcase | Slideshow
Topic Posts: 2/2
Posts: 2162
Joined: 26 Sep 2016, 00:16
Cash on hand: 7,946.37 GLP
Bank: 51,356.25 GLP
Gender: Guy
Location: Victoria, BC
Website: https://gimplearn.net
Operating System: Windows
GIMP Version: 2.8.22
My Personal Blog
My FineArtAmerica
x 1207
x 256
Contact:
Canada

Creating simple AI to help writers with guidelines

Post#1 by tim » 08 Nov 2016, 16:24


The idea is this.
You take a bunch of stories/tv shows/movies (media) and for each one you answer YES/NO to a list of features. List of features can be as big as you'd like, ideally large enough to cover all the media.
Ones you have answers YES/NO to all the features for these media and rating score for each media.
Just simply turn YES/NO answers into 1s and 0s.
TV Shows_Stories_Movies_Features.png
TV Shows_Stories_Movies_Features.png # (8.42 KiB) Viewed 328 times (Watermarked | Thumbnail)
You write a simple neural net that you feed input as features, and the output as ratings.
You then train the neural net with all the data.
After it's trained, it'll be able to identify between good features that go well together from bad ones.
You now generate a bunch of randomized media data with random features turned ON/OFF (YES/NO) (1/0). Then feed that through your trained network.
And output the ones that the the neural net thinks it'll get high rating for.
Then what you'll have is a simple guideline, or suggestive guide for media writers/creators to base their story/movies/tv shows on.

I had a chat with my cousin, and she doesn't think there's room for creativity in that. I think the opposite. I think there's plenty of creativity, it's only a guide to suggest what might work as higher rating, the details of the story still needs to be created/imagined to fit that suggested guide.
But i could be wrong, because i am a little machine biased.
I believe that computers can imitate creativity given large enough features list, it'll be able to identify good stories from bad stories. The ones it generates could be so random and that people didn't think of the possibility from before, but if the neural net is able to say it has good ratings from its learned experience from past media features, it would have great possibility of being a good skeleton to a good story. The meat of the story still needs to be created. How the bones are placed together still need to be decided. It's like a giant puzzle in the land of imagination.

But the hard part for me (the main part, the only part) that would stop me from writing such a program is the data (the YES/NO of features and the ratings from movies/tv/shows) as right now there seems to be no such data. Movies and books have ratings but they're not listed with exhaustive features like i need. You might get a short plot description. Writing the neural net is simple enough using Python but the data...is unavailable to me.

I was thinking we could potentially do something similar for music as well. With notes, we can teach a neural net to identify catchy songs or popular ones from unnoticed ones.

We are years behind in AI being able to write their own stories, but to identify simple YES/NO combinations? I think we've reached that point years ago.
0 x



aka Tin
Image gimplearn.net                                              

Registered members enjoy ads-free content
User avatar
tim

Administrator
View: Showcase | Slideshow
Topic Posts: 2/2
Posts: 2162
Joined: 26 Sep 2016, 00:16
Cash on hand: 7,946.37 GLP
Bank: 51,356.25 GLP
Gender: Guy
Location: Victoria, BC
Website: https://gimplearn.net
Operating System: Windows
GIMP Version: 2.8.22
My Personal Blog
My FineArtAmerica
x 1207
x 256
Contact:
Canada

Post#2 by tim » 09 Nov 2016, 00:43


The neural net doesn't have to be fancy. But the data is the key.
If i had access to such data i'd probably just use Python library neurolab like how i used here --> Neural network to predict Powerball white numbers by predict i mean for fun quickpick as it is unable to predict something that has no dependency on previous draws (disclaimer).
0 x



aka Tin
Image gimplearn.net                                              

Post Reply

Last 20 Members Who Visited This Topic

User avatar gnerd (1), User avatar Steve Reid (1)

  • Similar Topics
    Replies
    Views
    Last post
• Featuring Members' Works (Image Attachments Only) •
1: Ella
2: Issabella
3: Pat625
4: dinasset
5: Alan
6: tim
7: Conbagui

Who is online

Users browsing this forum: No registered users and 2 guests