ReRead: Research Note

Poetry project 2020

This project is inspired by David Jhave Johnston’s works and the book, ReRites: Human + A.I. Poetry produced and collected by him. The book is a collection of poems generated by A.I. and edited, organized by humans. The book has selected a large number of edited poems written in a short period of one year, with a high level of creativity and literacy, almost indistinguishable from poems that are written completely by human poets. Yet the book itself is a presentation of the very inspiring human-machine collaboration process and workflow. In my project, I attempted to have the machine analyze the text from the book, deconstruct it and label it, and have the result returned to us as readers. Human audiences will then read it again, perceive it along with the instruction given. Throughout this process, I created an unfamiliar context for poetry to exist, with the absence of either human authorship or agency of perceivers. By deconstructing the components of poetry, I want to clarify the role that has been played by human consciousness in the process of creating poetics and possibly explore the essence of poetics.

The text analyzation program I made is a simple convertor of data. The sentiment model used for text analysis is based on a pre-trained model provided by ml5.js. It is trained to score a paragraph of text and produce a number between the range of 0 to 1, reflecting emotion from negative to positive. The neuron net utilized in the ReRites project is a much-complicated system with models that run based on very sophisticated logic, but essentially share a similar core idea with the most simple machine learning units. Though different levels of text data processing methods are used, which decide how “smart” the machine could be, machine learning units created to understand human language follow the same pattern while trying to imitate human creation. The algorithm looks for a pattern in data sets of source texts to deconstruct the components of a piece of true human writings and looks for the optimal solution by combining words that gives the best feedback. If we see the structure of human literature as a tree that starts from one point and spreads out, A.I. literature will be a million strains gathered to one point. There is an essential difference between the way a machine writes and the way a human does.

Through observing the poems created by A.I. and comparing them to poems written by humans, I created the chart below. The list on the left is a series of words that reflect approaches and ideas used by machines while creating poems, while the other list is pointing out corresponding features in human poetry traditions. Words in the second list are not necessarily the opposite of their corresponding words in the first list. Words in the same row sometimes share the same origin or motivation, sometimes presenting different methods used to achieve the same goal.

The raw output section in Jhave’s ReRites has shown us that the A.I. poetry generated without human editing is nothing like a conventional poem. Though they seem to follow the logic, sometimes even containing certain forms of narrative, they are yet just compiled words selected based on a binary standard. We can say that these raw outputs are not as good as human writings because the machine hasn't yet fully understood what poetry is, that there are too many factors to be considered in order to recreate a human consciousness. But on a deeper level, the essentially different approach could also be a reason that A.I. cannot and might never be able to write like a human.

The human experience is a very critical component of the process of generating poetics. Factors that we believe are what make a good poem, such as metaphor, imagery, association, and self-reflection exist relying on the present of experience as a physical body. In a poem of traditional meaning, the poet will construct words and sentences around his or her personal experience to send a coded message to readers and have the readers decode the message with their own experience of their life. When the poet's intention of expression matches with others’ experience, merged and resonated to a certain extent, poetics exist. On the other hand, machines certainly will not have experience of their own for they have never lived through life like a human, but that doesn't mean it is impossible for machines to learn about the human experience. Though such technology doesn't seem to exist yet, theoretically, it is possible for the machine to study human narratives and extract a collective experience of all humans and use it as a pattern to construct a narrative around it. With the same mechanism, machines could also find out a general equation of association made when human readers read.

If a human mind could create poetry based on their experience, and have it perceived by others with their very own experience. How could a machine, with no experience that is unique to itself, take the place of either the author or the reader? At this point in technology development, we could say that machines are experiencing the world through data sets, as all machine learning models are trained with information collected one way or another. In most cases, they are designed to digest contents collected on the internet or obtained through an archive. The data set is the worldview of the machine. The human mind behind the data, ones who produced content, are providing bits of clues from their own experience to form a collective experience for the machine to take as its own. It is reasonable to believe that in the very near future, a powerful machine will be developed to gather data from every corner of the world in real-time with no bias. Then will this machine be able to experience the world with the mind of all human beings on earth, and predict the perception of a poem for everyone? The question then led me to another part of this project.

Through the long history of poetry and humanism, we tend to believe that art and poetry are inviolable territory for the human mind. Therefore, we can assume that there will be only poetic production in the process where human agency is involved in the start. The diagrams below represent four different cases of a poem that could be produced and perceived with or without Machine.

The writer and the reader are two individuals that are responsible for creating poetics in the process of reading a human written traditional poem. The writer encodes their experience into the poem and predicts the way it will be decoded by readers, in order to take dominance in the process of creating poetry. Reading poems from the book ReRite is a very similar process as reading a human written poem, but yet very different. The editor and the reader are still responsible for creating poetics in this process, but the dominant source of experience that poetics rely on becomes unclear. During my experiment of analyzing poems and having them read by human audiences, I found I have taken away the authority for readers to insert poetics into the poem, instead, they are following the given instruction, perceiving the poem partially without responding to their own experience. For comparison, I programmed the machine to create a barcode in my experience, and switched the input to the raw output from the book as material. I created a theoretical vacuum of poetics, for there is no human involvement beyond the programming level, and only to have the machine talk to the machine, speaking in a language that no human could understand.

However, after reading all the outcomes of all combinations, I found that the presupposition of my experience can be overturned completely, while the essence of poetics remains unclear. In all the poems I have identified a certain level of poetics and sensed the existence of someone's experience and consciousness. Can we then say that poetry is just a combination of words that optimizes its effect on readers by corresponding to the experience of them as close as possibl?We could only answer this question if we have a poet that writes with absolutely no experience and intention. but that poet does not exist. Whether the creator of poetry is a machine or a human, or both. The editor was also perceiving poetics and experience while editing the raw text instead of simply selecting words to construct meaning from scratch. The edited poem is not just a reflection of the editor's experience, but a process of mutual inspiration between the machine and human, a complex that can not be defined. At this point, there is no answer to what the future of poetry will be like. Though there is a possibility for machines to embrace collective experience in writing, it is hard to define and make clear how personal a poem should be to be a reflection of human spirit and free-well. Like Jhave said at the end of his book, the creativity of humans is ephemeral, but the machine will never stop evolving. Perhaps that is the very poetics of poetry, it is the very selfishness of our ego to see ourselves in all that we see, hear and read; and the perfect poem would be found within this realization of human nature, whether it is in us as individuals or collection.

To view the program created for this project, click here.

ReRead: Sentiment analysis program:

Output pdf:
Shuyuan Zheng Deisgn & Digital Assets
© 2021 Shuyuan Zheng.