Ever since we had the notion of A.I. itself, we’ve continuously explored the conflicts between humanity and machines. It’s manifested in movies, tv shows, politics, and much more. But what if we focused more on cooperation between A.I. and people?
At first, my project was to experiment with Google’s TensorFlow to create original text based on the writings of famous people. TensorFlow is a set of tools made by Google to assist in the use of machine learning. I used GitHub user hunkim’s tool, word-rnn-tensorflow (Multi-layer Recurrent Neural Networks (LSTM, RNN) for word-level language models in Python using TensorFlow).
In an oversimplified way, a recurrent neural network essentially takes a sample text and spits out a new text that’s similar to the original text sample. It trains itself with the data by collecting the vocabulary of the starter text and creates statistical models of words. Then, it learns to group words together and recognizes sequences within the text. Eventually, it recognizes patterns such as common phrases and even sentences. To generate the output text, the algorithm picks a random word like “a” or “the”, and generates a word that’s most likely to come after. Then, it generates a word that might come after the second word. As it loops, sentences form and an incoherent mess of words that resembles the original text comes out. Pretty primitive, right?
At first, I trained it with every single tweet that Trump has ever made. By training it for eight hours with a couple hundred thousand words from the entire collection of tweets made by Trump, the algorithm returned an amalgamation of Trump that bordered on humorous parody to nonsense.
However, I took it one step further by feeding the algorithm a collection of poems created by E. E. Cummings. After a mere twenty minutes of training, the algorithm spits out a paragraph of words:
“to somehow might mile yes for comes your because; of joe hung of falsest gears God was body to deeplyness silence his look less those beyond birthday (quietly & father’s love 9 deaf and dumb thy before. sorry of more any smell a little about as by remember and not many perhaps. “sweet three acclaim the silver rooms dead, to dream,create and tell war mountains the sunly let i know all sunly a nothing (of dart but foolish no because not frisks) cohn) do hate which get offered stands killed also despite one dares of like be build and (with we mire, Smiles.he visited, and lover silver eternal invents each fades) so song and here’s to 8 its 35 My win a schoolroom this they it all here you you a not. murdering My asleep, and on less (love shall day her first on the green sea sing lifted go so. face are 11 ran a tree but inherit,all love of all said all of now should Love themselves as must my eyes mountains and pity like born their truth than spoke. i dancing is side the lean superfluous Progress 79 (wonderful) mountains; and made and born in skillfully chains again: to”
Obviously, this is not a poem. Instead, it is a collection of words randomly generated by a computer from another collection of words that were arranged more meaningfully. However, this isn’t the final step in writing this poem. I sent this block to text to my friend, who edited the text to create her own interpretation of the poem. By replacing words, adding enjambment, the very first iteration of the poem goes:
In this way, poetry is produced in a different manner than normal. The computer generates a proto-poem while a person refines it and adds meaning to make the poem more human. However, the computer can also fulfill other roles in creative writing, with editing and inspiration being just a few of the limitless possibilities. Ultimately, whatever the role the computer plays still makes the computer as a tool and not as a dominant component of the poem.
So through all of this, we can create a meaningful poem by collaborating with machine learning. But what does this all mean? It might mean that Elon Musk’s nightmare of an A.I. dominated future is the same as the backlash against machines in the industrial revolution (where workers initially thought that machines would reduce the need for workers). As history shows, however, it turned out differently; machines augmented the production abilities of workers, allowing for factories to produce more while retaining employees. Machine learning might work in a similar way. If companies can use machine learning to augment the abilities of workers, we might just have another industrial revolution.