Social media permeates every facet of modern life, from our connecting with loved ones, social groups, to how we gather and make news. With humans being the social creatures that we are, it’s no surprise that this form of communication would become who we are. We want to connect with others as much as we can and with the technology that we have today, this is possible. In our need to connect we often end up exaggerating and lying, both consciously and unconsciously, in order to paint a better picture of ourselves to show to the world. Bad actors of the world have noticed this and seek to exploit our biases and gullibility towards what we see online in order to feed us fake news and sow dissent amongst ourselves to further their agendas.
Using varying methods of machine learning in order to create mock societies within the internet, such as markov chains to produce text and generative adversarial networks to produce images, I seek to expose these holes in our global internet society and bring it to the forefront of the conversation around how we conduct ourselves on the internet. With a conscious mind we may be able to reject the superficiality of the internet and use it to supplement and support, rather than taint, the reality in which we all live.
Congressional Lunch is a website that generates fake tweets from real politicians using a Markov chain, a process where random variables are transitioned from one state to another based-on assumptions and probability. All the most recent tweets from each politician's Twitter pages are put into a pool of their own party. From there, tweets are generated and published periodically according to the party they are accredited to. With three screens, the Republican tweets are shifted to the right, the Democratic tweets shifted to the left, and the middle acts as a feed for all of them. Although the tweets are affected by their party, they all utilize the same platform that brings them both together while still being separated.
The Book of Markov
With the Christian Bible being the best-selling book of all time and translated into hundreds of languages, errors and mistranslations are bound to happen in the almost 2000 years of its existence. In fact, Leviticus 18:22 states, "do not have sexual relations with a man as one does with a woman; that is detestable." Yet it is argued that the word "man" has been mistranslated from Hebrew and a more accurate reading would be "underage boy." Therefore, the detestable act would be referring to pedophilia, not homosexuality. With this in mind, how do we know how much of this sacred text has been altered and misused by man to be shared and preached to the whole world?
The Book of Markov is a website that places artificially generated Bible verses in front of stock images. Using a Markov chain, a process where random variables are transitioned from one state to another based-on assumptions and probability, created from the entirety of the King James Bible, these holy words are misappropriated into a new space and ready to be shared onto your Facebook wall. With these new biblical passages, we can create a new never-ending version of the Bible for the internet age, The Book of Markov.
As our ability to access the internet has grown, so too has our access to pornography. Every day, hundreds of thousands, if not millions, of pornographic images and videos are posted to social media sites to be shared for whoever can view them. As we scroll through our feeds, we're bombarded with graphic images that desensitize us from the thrill we once experienced when viewing them. These images get blended into each other in our minds and turn the explicit into the abstract.
Artificial Flesh is a series of images created using a Generative Adversarial Network (GAN), where a generator learns to create real images and a discriminator learns to tell apart real images from fake one. With roughly 100,000 pornographic images scraped from Reddit, the network goes through numerous cycles to learn what the binds these images together in order to create new images that can be best described as an average pornographic image. Although the original images feature people in various poses and acts, the one attribute that makes up all of them is the flesh on display and the generated images reflect this, creating an abstracted and artificial flesh with each passing cycle.