Visual Journal: Entry 5

 To help me create a piece of generative art, I began by looking at the work of Surrealists from the 1920s. One technique, called “Cut-Up”, was pioneered by Tristan Tzara, an avant-garde poet. He would create a poem by pulling words at random from a hat. I thought it would be interesting to experiment with this technique to create my own surreal poetry. However, I wanted to update the technique by using a contemporary means of production to generate the words. On my smartphone, I noticed that there’s a keyboard feature called “Predictive Text” that activates whenever I type a word. It attempts to save time by predicting what I’m going to type next and suggesting three optional words to use. What’s interesting, is that these suggested words are taken from a custom dictionary of my most commonly typed words and phrases. Whenever I type on my phone, an algorithm constantly scores each word by the probability that I’ll use it or need it again and gradually learns my typing habits.


 For my first experiment, I wanted to see what kind of sentences the algorithm would generate if I chose words that it had suggested. I clicked the middle button many times, and found that it was capable of composing a semi-coherent sentence all by itself:


 This was a great start but I noticed that since it’s based on my most commonly used words it would suggest them so often that after a few lines it would fall into a feedback loop and get stuck repeating itself, with only minor variations. Since the interface is capable of suggesting three words at once, I decided to assist the algorithm by choosing whichever word I felt made the most sense and would allow the sentence to continue:


 This gave much better results, since the algorithm can only predict the next word, so I was able to guide it towards creating a more coherent sequence of words. Whilst doing this, I realised that I was effectively playing a digital version of “Cadavre Exquis (Exquisite Corpse)” with an algorithmic version of myself. This technique was invented by the Surrealists in 1920, who would fold a piece of paper into sections, and then take turns drawing on each segment, whilst only being allowed to see the end of what the previous person had contributed.
 I also noticed similarities between my results and those created by Brion Gysin’s “Cut-up Technique”, which was extensively used by William S. Burroughs. However, my sentences repeatedly contained similar words, despite their varying compositions. Since these words are sourced from my personal dictionary, I decided to ask a friend to repeat my experiment on their own smartphone. They used the same method to choose words but, since their writing style is naturally different to mine, they produced completely different sentences.


 So far, my experiment had successfully produced a series of interesting results. Nevertheless, I wondered whether the generative process was too constrained by my initial word, since the first three suggested words by the keyboard were always “I, You, and I’m”. To investigate this, I tried one final variation of the method by using an online random word generator, to choose the first word of my sentence. The website chose “Houndstooth”, so I used predictive text to generate a sentence and tried to keep it relevant to the original seed word. An old person is sometimes described as being “long in the tooth” and my sentence came close to describing that. To produce an artwork, I copied my sentence into Photoshop and pasted it on top of a houndstooth background pattern.


References:
1. https://lifehacker.com/how-predictive-keyboards-work-and-how-you-can-train-yo-1643795640
2. https://www.moma.org/learn/moma_learning/max-ernst-levade-the-fugitive
3. http://www.bbc.com/news/magazine-33254672
4. http://watchout4snakes.com/wo4snakes/Random/RandomWord