zoë's itp blog

Learning Machines: Final Project Proposal

Teen girls occupy an interesting place in Western cultural imagination. It’s a fetishized state where women are beginning to see that the rest of society sees them as objects and sexualized creatures. The obvious example is Lolita, but other narratives such as Thirteen and even Clueless and Mean Girls come to mind.

In 1999, Tiqqun published a monograph, ‘Preliminary Materials for a Theory of the Young Girl’ which details how capitalist society is oriented towards the teen girl as ur consumer: ” Tracing consumer society’s colonization of youth and sexuality through the Young-Girl’s “freedom” (in magazine terms) to do whatever she wants with her body, Tiqqun exposes the rapaciously competitive and psychically ruinous landscape of modern love.”

I took Allison Parrish’s class ‘Reading and Writing Electronic Text’ in the spring and got to experiment with writing programs to craft text from multiple corpuses. My final project for that class was a re-visiting of my high school blog, sugaredlemon (still my handle online). I created a next iteration of the blog that was populated with new blog posts, written using six years of my own writing and markov chains. One of the things that I found most interesting was the linguistic peculiarities of both internet-speak of the early aughts, teen girl speak, and my own particular phrases.

 

PROJECT CONCEPT:

How does what exists online about teen girls conform to or break away from a vision we maintain? Is it possible to create a teen-girl’s voice and actions and what does that say about teen culture and visibility? What is the style and aesthetic moment that we’re in that draws inspiration from 90s teenage girls? What power do they hold? What do their words and actions say about our society? How does mess and hormones play with sleek technology? How does confession play a role?

My first idea is to create a fictional teen girl, instantiated through an auto-generated online diary or My ideal would be using the form of the diary, which while sites like Xanga and LiveJournal are relics of a different web though it seems teens still keep pen and paper diaries. Websites for woke teen girls like Rookie actually have a diary section.

My second idea is to make it more relevant by utilizing platforms that are currently popular with the demographic, like Snapchat and Instagram. Would involved more of an image generation, but comments would also be important.

Of course, I can’t go any further without a mention of Tay, Microsoft’s glorious fail of a twitter bot.

Following many others, Tay is a female-gendered AI/bot. What’s interesting for me about Tay is the mystery with which she was developed. Microsoft simply says that she was trained using “public data” and from reading some of her tweets, they were at least using some actual teen speech (“new phone, who dis?”). For more, this Quora article does some detective work plus Microsft released a platform based off of what they used to create Tay to allow others to build chatbots.

What would it mean for it not to be a space of interaction but rather render it as this somewhat fleshed-out online model? Could she be programmed to create certain amazon wish lists? Spotify playlists? Basically, what generative potentials exist beyond text?

 

EDIT: turns out, shopping bots are a thing and Darius Kazemi made a random Amazon shopper  and Nike Shoe bots are apparently a real problem.

TOOLS:

Depending on what direction I take…but most likely TensorFlow in any case.

If I deal with images, I’ll be using Clarifai’s API (particularly image concepts). As well as looking into this MIT research that has come up with some really interesting generative videos/animations.

If I’m dealing with text, diving into natural language processing such as NLTK or Stanford’s Core NLP (though it’s in Java…). Google has some interesting white papers, as does Microsoft.

 

CHALLENGES:

(Many…?)

Finding an automated way of scraping the web/texts (web crawlers?) to create a corpus of visual or textual content

Simplifying, choosing one form of output (Snapchat? Amazon?)

What algorithms to choose

 

Other random project ideas:

VR and machine learning (Google: deepdream in daydream)

Same as above, but with internet trolls

Generative intimacy images (generate video/images trained on pictures of people holding hands, hugging, kissing, etc.) (teach a robot how to love)

White-ness

07. November 2016 by zoe.bachman.itp
Categories: Learning Machines | Tags: generative, girls, machine learning, nlp, teens | Leave a comment