In broad terms, the concept I am interested in is ‘Digital Gentrification’; the idea machines will begin to move into our abandoned virtual spaces. This project hopes to capture the feeling of returning to these spaces to find that they have been taken over by another intelligence. The space is now alien, we are no longer relevant or necessary. I like to think of the analogy of an abandoned theme park that is now home to a swarm of insects.
I am interested using web scraping and open API’s to appropriate content that has been generated by people on the internet. I am creating a database of this information and have began using various machine learning technologies to create distortions in the data that reflects a sense of a machinic ‘perception’ of us.
One example of the algorithms that I have invented to generate content is NASDAQ GOOGLE WORD ASSOCIATION:
In this example: The algorithm downloads the first Google image that relates to the NASDAQ stock code, then performs a kind of ‘Word association’ game on this image with Google’s ‘im2text’ algorithms.
AETHLON MEDICAL, AEMD
Man Holding a baseball bat on a field
Man in a suit and tie standing in a field
NABRIVIA THERAPE ADS, NBRV
Woman holding a pair of scissors in her hands
VARIABLE RATE INVESTMENT GRADE PORTFOLIO, VR IG
Person holding a pair of scissors in their hands
PACIRA PHARM INC, PCRX
People standing next to each other
EGAIN CORPORATION, EGAN
Man holding a tennis racquet on a tennis court
PATRIOT NATL, PNBK
Different colored signs on a pole
I am comfortable with the high level of ‘randomness’ that this algorithm generates - the point here is scale of content rather than quality. These ‘associations’ are not made by myself, they are made by the inherent biases and limitations in Google’s machine learning tools.
Another algorithm I am working on is a DCGAN of generated meme images. The DCGAN is trained on 40,000 memes that I have scraped from instagram. It will ‘generate’ a memes based aesthetic based on a representation learned from training.
Here is an 8x8 grid of ‘memes’ generated by the algorithm after about 12 hours of training.
And another from later in the training:
The DCGAN aesthetic is always cool to me, although I’m a little disappointed that the ‘meme’ references seems to have been lost. I am considering getting a database of ‘meme phrases’ and then programmatically adding the text to these images with the iconic Impact typeface.
I am also working to scrape tweets from twitter for a char-rnn network that will generate text as a representation of the noise on the internet.
I wish to mask and pervert the basic sense of reality that the internet is a human space. While networks gains an understanding of who we are, we are incapable of comprehending the nature of the machines intelligence. Our definition of intelligence is anthropocentric - we evaluate how ‘intelligent’ something is based on how similar it behaves to a human. This project seeks to subvert this sense of anthropocentric and give sense of a ‘networked other’.
The original idea for this project was to take the content and automate a series of ’bots’ that would ‘take over’ an open source message board (flaskBB). However, I am in the process of exploring other options regarding how this virtual space could be represented.
Mat DesLauriers has generously shared a generative node.js script that could be used to generate a base landscape. This could be a basis for experimentation. In addition to this, Three.JS could be used to develop an asset that could be procedurally placed on this generated landscape that could function as screens that will display the hallucinations from the machine learning content database.
It would be wrong not to mention the work of Design and Technology theorist Benjamin Bratton whose ideas about Artificial intelligence and the Anthropocene have influenced me greatly.