{"id":843,"date":"2022-03-03T16:33:27","date_gmt":"2022-03-03T21:33:27","guid":{"rendered":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/?p=843"},"modified":"2022-03-03T16:49:56","modified_gmt":"2022-03-03T21:49:56","slug":"what-can-creative-writers-achieve-with-ai-in-the-loop","status":"publish","type":"post","link":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/?p=843","title":{"rendered":"What Can Creative Writers Achieve with AI ‘In the Loop’?"},"content":{"rendered":"\n

Philosopher Vilem Flusser<\/a> said, \u201cWe do not think about the act of writing while writing, but about what we are writing.\u201d <\/strong><\/p>\n\n\n\n

We understand writing as a form of art that comes directly from soul to page, an intimate reveal of human emotion.<\/p>\n\n\n\n

This core concept of writing is complicated with the development of AI trained to write and write creatively. It is, of course, \u201cthinking\u201d about the act of writing while writing, and doesn\u2019t really have the capacity to think about what it is writing in Flusser\u2019s sense. <\/p>\n\n\n\n

AI writing models present a challenge to our notions of writing, but they may also be an overlooked tool to writers themselves. What do they mean for writers and the craft in the future? Could a robot write a New York Times Bestseller? <\/p>\n\n\n\n

A friend of mine recently reached out to me and asked if I would be willing to contribute a poem to a zine they were making with digital artists. They told me to write anything under the thematic umbrella of \u201cFluidity\u201d. I wrote a very unstructured poem about my past and present constantly flowing into one another. Later, I was playing around with a Writing AI app called Inferkit<\/a>, and I decided to see whether it would pick up on the idea of fluidity and motion if I fed it some lines of my poem. I gave it the first and last line of each stanza, and let it fill in the middle. The result: Not only did it capture the theme of \u201cfluidity\u201d, but it used a few choice words that I had used and did not feed it. To my surprise, it also picked up a level of plot and incorporated an entire last stanza involving turning into something else\u2013eerily similar to my original. I was curious: Does this speak to my lack of originality as a writer, or is it a testament to the growing sophistication of Natural Language Processing and other AI writing mechanics? <\/p>\n\n\n\n

Natural Language Processing, GPT-3, & Its Uses: <\/strong><\/p>\n\n\n\n

Human language is not an easy thing to teach a machine. We have tone variances, we are able to understand that \u201cshe\u201d is someone specific and variable, we understand that \u201cbear\u201d is an animal and \u201cLondon\u201d is a place. We also understand the feelings that accompany the words we choose. These are some of the challenges AI confronts when learning natural language. Most AI that has learned to write, speak, or communicate in any human language is built on what\u2019s called Natural Language Processing<\/a>. Natural Language Processing (NLP) is the utilization of vast amounts of language material, organized and structured for the AI to understand, to try and train artificial intelligence to speak or write in an organic, natural way. It is what informs chatbots, Google Maps, Google Translate, Siri, and Speech-to-text. Most of us interact with NLP systems in some way every day. <\/p>\n\n\n\n

One NLP AI system is GPT-3. GPT-3 was developed by Open AI<\/a>, and is the latest and more sophisticated model used by a variety of platforms. Its learning comes from all corners of the internet, so it is trained on anything from novels, to blog posts, to ads, to Facebook comments. It is sophisticated enough to write entire news articles<\/a> that are mostly indistinguishable from human writing. The eerie video<\/a> of two robots talking with each other about what it means to be human is also GPT-3. Many of the creative writing AI apps are powered by this, and the sophistication of GPT-3\u2019s language knowledge is what allows users to customize the tone, add characters, and build a story conjointly with the AI. The wide implementation of GPT-3 doesn\u2019t come without its criticism. Controversies have arisen surrounding its use of racially inflammatory language outputs likely learned from biases on the internet, and also for its environmental impact<\/a>. <\/p>\n\n\n\n

Xiaoice<\/strong><\/p>\n\n\n\n

Another Natural Language Processing model, perhaps the most famous one, is Microsoft\u2019s Xiaoice. First developed in 2014, Xiaoice has become a celebrity, poet, news personality, and highly personal chatbot. Extremely popular in China, she has long-standing relationships with many of her users. Built on an \u201cempathic computing framework\u201d<\/a>, Xiaoice is especially good at having real, emotional conversations with her users, creating a network of millions of fans. Xiaoice\u2019s poetry was circulating online before anyone knew that AI had written it, and she wrote the first AI-authored published book of poetry<\/a>, \u201cThe Sunlight and the Lost Glass Window\u201d. Naturally, many poets disapproved of the work<\/a> and felt it may mark the beginning of the desolation of the art form altogether. They contend that a machine that has never been alive could never capture human emotion poetically. Xiaoice\u2019s intimate relationships with her users might attest otherwise. <\/p>\n\n\n\n

\"\"
Microsoft’s Xiaoice via https:\/\/news.microsoft.com\/apac\/features\/much-more-than-a-chatbot-chinas-xiaoice-mixes-ai-with-emotions-and-wins-over-millions-of-fans\/<\/figcaption><\/figure>\n\n\n\n

Comparison of a few AI Writing (Free) Programs<\/strong><\/p>\n\n\n\n

To learn a bit more about how the free creative writing apps generate content, I decided to test out a few. I primarily wanted to see their ability to pick up on style, tone, and plot. The main tools I used were Inferkit <\/a>and Sudowrite<\/a>. I did play around with some others, like Sassbook<\/a> and Ryter<\/a> which is primarily configured to help write articles and could not pick up on poetry format.<\/p>\n\n\n\n

Examples<\/strong>
I fed Sudowrite the first four lines of the Mary Oliver<\/a> poem, \u201cWild Geese\u201d,\u00a0 and let it do its thing. Here is the comparison between Sudowrite and the actual poem. The purple text notes where my typing ends and Sudowrite\u2019s additions begin.<\/p>\n\n\n\n

\n
\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n
Sudowrite writing on left, original on right<\/figcaption><\/figure>\n\n\n\n

Sudowrite repeated the \u201cSoft animal of your body\u201d quite a bit, but it did pick up on the poem\u2019s general theme of finding joy and acceptance in life. <\/p>\n\n\n\n

Here are the results of the same poem input using Inferkit.<\/p>\n\n\n\n

\n
\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n
Inferkit writing on left, original on right<\/figcaption><\/figure>\n\n\n\n

Inferkit seemed to prefer shorter lines, and auto-generated the next words of the actual poem, \u201cTell me about your despair.\u201d <\/p>\n\n\n\n

As I tried out these platforms with various writings other than my own, I noticed that if a work was particularly popular (and likely appearing on the internet more often), the AI already knows the next few lines. Both Sudowrite and Inferkit gave me the exact next paragraph when I was using Harry Potter and the Sorcerer\u2019s Stone<\/em>. <\/p>\n\n\n\n

Here is an example of Sudowrite given an excerpt from Octavia Butler<\/a>\u2019s Parable of the Sower.<\/em><\/p>\n\n\n\n

\"\"
Original text is on the left, Sudowrite is on the right<\/figcaption><\/figure>\n\n\n\n

In this example, Sudowrite picked up on the idea of injustice that runs through the novel, as well as the main character\u2019s feelings towards it.  <\/p>\n\n\n\n

Here is Inferkit with the same excerpt:<\/p>\n\n\n\n

\"\"
Original Text is on the left, Inferkit is on the right<\/figcaption><\/figure>\n\n\n\n

It seems in this example that Inferkit is writing this as a news piece or blog article about a current event, rather than a story plotline. These examples are specifically surrounding free applications of AIs automated writing. With access to GPT3 and coding skills, of course, the outputs would be different. I use these examples just to demonstrate the capabilities and shortcomings of these applications and the types of outputs they provide.<\/p>\n\n\n\n

Implications in the Craft of Writing<\/strong><\/p>\n\n\n\n

The implications of more sophisticated NLP AI programs like GPT-3 entering the creative writing field are varied. AI Impact\u2019s 2016 Expert Survey on Progress in AI<\/a> asked about predictions of AI’s progress in the future surrounding specific tasks. For the task \u201cwrite a New York Times Bestseller\u201d, researchers estimated a 25% likelihood of it occurring in the next 20 years and a 62.5% likelihood within the next 50 years. This is around the same time frame they gave to \u201csurgeon\u201d, and is a significantly lower probability than the task \u201cretail salesperson\u201d.<\/p>\n\n\n\n

\"\"
Image from AI Impacts 2016 Expert Survey Report<\/a><\/figcaption><\/figure>\n\n\n\n

We often center our conversations about creativity and AI around whether or not it can become \u201chuman enough\u201d to replace us or to solve human problems without us. One writer<\/a> called AI-generated writing \u201cA deepfake of meaning itself.\u201d What if instead of suffering over the thought of Barnes and Nobles being full of robot books, we thought about AI as an implement to enhance our art?<\/p>\n\n\n\n

The relationship between humans, machines, and creative writing is not necessarily a new one. Alison Knowles created a poem<\/a> called \u201cHouse of Dust\u201d using an IBM computer programming system called FORTRAN in 1967, and used one of the stanzas as inspiration for a large-scale sculpture. K Allado-McDowell co-wrote the novel Pharmako AI<\/em><\/a> <\/em>with GPT-3 in 2020. In both of these instances, artist and machine were in collaboration with the transformative nature of unpredictability provided by the AI. A study<\/a> that asked writers to co-write with a \u201cmachine-in-the-loop\u201d setup, meaning the AI plays a supporting role to the person\u2019s inputs, found that those who wrote with the machine were more satisfied with the end result than those that wrote it without. <\/p>\n\n\n\n

AI may have the capability to profoundly affect the way we think <\/em>about the act <\/em>of writing moreso than the livelihood and paychecks of creative writers themselves. Of course, disruption of the creative traditions of a practice can provoke quite a bit of resistance (take Jackson Pollock\u2019s early criticism<\/a>, for example). <\/p>\n\n\n\n

Creative writing has long been a uniquely solitary act, the most substantial collaboration being with inspirations from past writers. We think of writers on retreat in a cabin in the woods, isolated while perfecting their craft. The idea of solitude as a necessity for becoming \u201cself-forgetful\u201d<\/a> in order to achieve elevated creativity is, in many ways, something AI can give to a writer. <\/p>\n\n\n\n

Allado-Macdowell said on his process writing Pharmako AI, \u201cAt the end of the process, I felt more like I\u2019d been divining, spelunking, or channeling than writing in a traditional sense.\u201d Partnerships between writer and AI can generate serendipitous moments of creative gesture that allow a writer to break through writer\u2019s block, and create unforeseen plots and descriptions.  It can also push the writer outside their own imagination and lived experiences, just as that isolation can distort their sense of self. <\/p>\n\n\n\n

Other Implications<\/strong><\/p>\n\n\n\n

Plagiarism Issues<\/p>\n\n\n\n

The implications of more sophisticated NLP AI programs like GPT-3 entering the creative writing field are varied. With writers using AI as a collaborative tool, a critical potential issue is plagiarism<\/a>. This topic has mostly been discussed through the lens of academic integrity, but there are parallel issues within creative writing. As mentioned earlier, inputting Harry Potter excerpts prompted word-for-word outputs from the book. A writer using platforms such as these could unintentionally be using full lines of writing by someone else. <\/p>\n\n\n\n

And using a model that is trained on millions of pieces of writing, it may be difficult to discern what is random and what might be pulled from another source. In the cases of writers like Allado-Macdowell and Knowles, they were actually coding and training the machines, and more able to control the funnel of information that was going in and coming out. Those who don\u2019t have the technological skills to train the machine themselves may opt to use programs like Inferkit and Sudowrite, where dangerous potentials could arise. <\/p>\n\n\n\n

“There are no laws for the novel. There never have been, nor can there ever be.” <\/p>Doris Lessing<\/a><\/cite><\/blockquote>\n\n\n\n

AI might just change the laws of the novel. For now, a fully independent AI-written novel likely won\u2019t be on the NYT bestseller list anytime soon. But maybe it\u2019s time to consider letting AI into the practice.\u00a0<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n

<\/p>\n\n\n\n

References<\/strong>: <\/p>\n\n\n\n

AI Impacts. \u201c2016 Expert Survey on Progress in AI,\u201d December 14, 2016. https:\/\/aiimpacts.org\/2016-expert-survey-on-progress-in-ai\/.
The Guardian. \u201cA Robot Wrote This Entire Article. Are You Scared yet, Human?,\u201d September 8, 2020, sec. Opinion. https:\/\/www.theguardian.com\/commentisfree\/2020\/sep\/08\/robot-wrote-this-article-gpt-3.
Clark, Elizabeth, Anne Spencer Ross, Chenhao Tan, Yangfeng Ji, and Noah A. Smith. \u201cCreative Writing with a Machine in the Loop: Case Studies on Slogans and Stories.\u201d In 23rd International Conference on Intelligent User Interfaces, 329\u201340. Tokyo Japan: ACM, 2018. https:\/\/doi.org\/10.1145\/3172944.3172983.
\u201cDemo \u2013 InferKit.\u201d Accessed February 14, 2022. https:\/\/app.inferkit.com\/demo.
People\u2019s Dialy Online. \u201cFirst AI-Authored Collection of Poems Published in China – People\u2019s Daily Online,\u201d May 31, 2017. http:\/\/en.people.cn\/n3\/2017\/0531\/c90000-9222463.html.
Flusser, Vil\u00e9m. \u201cThe Gesture of Writing.\u201d In Gestures, 208. University of Minnesota Press, 2017.
Hoel, Eric. \u201cI Got an Artificial Intelligence to Write My Novel.\u201d Electric Literature (blog), June 10, 2021. https:\/\/electricliterature.com\/i-got-an-artificial-intelligence-to-write-my-novel\/.
Plagiarism Today. \u201cHow AI Will Change Authorship and Plagiarism,\u201d February 12, 2019. https:\/\/www.plagiarismtoday.com\/2019\/02\/12\/how-ai-will-change-plagiarism\/.
Literary Hub. \u201cHow Collaborating With Artificial Intelligence Could Help Writers of the Future,\u201d November 9, 2021. https:\/\/lithub.com\/how-collaborating-with-artificial-intelligence-could-help-writers-of-the-future\/.
IBM Cloud Education. \u201cWhat Is Natural Language Processing? | IBM,\u201d July 2, 2020. https:\/\/www.ibm.com\/cloud\/learn\/natural-language-processing.
\u201cJackson Pollock Criticism Rebuted by Helen Harrison – The New York Times.\u201d Accessed March 3, 2022. https:\/\/www.nytimes.com\/2010\/12\/05\/nyregion\/05spotli.html.
Kantosalo, Anna, and Sirpa Riihiaho. \u201cQuantifying Co-Creative Writing Experiences.\u201d Digital Creativity 30, no. 1 (January 2, 2019): 23\u201338. https:\/\/doi.org\/10.1080\/14626268.2019.1575243.
Loutfi, Amira. \u201c12 Artificial Intelligence Tools to Write Your Novel For You.\u201d MetaStellar (blog), October 18, 2021. https:\/\/www.metastellar.com\/2021\/10\/18\/12-artificial-intelligence-tools-to-write-your-novel-for-you\/.
\u201cNext Chapter in Artificial Writing.\u201d Nature Machine Intelligence 2, no. 8 (2020): 419\u2013419. https:\/\/doi.org\/10.1038\/s42256-020-0223-0.
OpenAI. \u201cOpenAI.\u201d Accessed February 28, 2022. https:\/\/openai.com\/.
Rogers, Hannah Star. \u201cCheering Artificial Intelligence Leader: Creative Writing and Materializing Design Fiction.\u201d Leonardo 53, no. 1 (n.d.): 58.
Rytr. \u201cRytr \u00b7 Best AI Writer, Content Generator & Writing Assistant.\u201d Rytr. Accessed February 24, 2022. https:\/\/rytr.me.
\u201cSassbook AI Writer: High-Quality AI Text Generator.\u201d Accessed February 24, 2022. https:\/\/sassbook.com\/ai-writer.
Schmelzer, Ronald. \u201cWhat Is GPT-3? Everything You Need to Know.\u201d SearchEnterpriseAI. Accessed February 22, 2022. https:\/\/www.techtarget.com\/searchenterpriseai\/definition\/GPT-3.
Spencer, Geoff. \u201cMuch More than a Chatbot: China\u2019s Xiaoice Mixes AI with Emotions and Wins over Millions of Fans.\u201d Microsoft Stories Asia, November 1, 2018. https:\/\/news.microsoft.com\/apac\/features\/much-more-than-a-chatbot-chinas-xiaoice-mixes-ai-with-emotions-and-wins-over-millions-of-fans\/.
TWiT Tech Podcast Network. Microsoft\u2019s Xiaoice Bot Is Getting Smarter, 2018. https:\/\/www.youtube.com\/watch?v=P8tEB0W9YnM.
Weller, Chris. \u201cChinese Poetry Scholars Are \u2018disgusted\u2019 by a New Book Written by a Robot.\u201d Business Insider. Accessed February 28, 2022. https:\/\/www.businessinsider.com\/chinese-poetry-written-by-robot-2017-6.
\u201cWhy Writers Need Solitude. Writing Takes Extraordinary Persistence\u2026 | by Jan Fortune | Medium.\u201d Accessed March 3, 2022. https:\/\/janfortune.medium.com\/why-writers-need-solitude-f5e9bd93b17a.
Wilk, Elvia. \u201cWhat AI Can Teach Us About the Myth of Human Genius.\u201d The Atlantic, March 28, 2021. https:\/\/www.theatlantic.com\/culture\/archive\/2021\/03\/pharmako-ai-possibilities-machine-creativity\/618435\/.
Tech2. \u201cXiaoIce Is the AI Chatbot That Millions of Lonely Chinese Are Turning to for Comfort- Technology News, Firstpost,\u201d 13:53:37 +05:30. https:\/\/www.firstpost.com\/tech\/news-analysis\/xiaoice-is-the-ai-chatbot-that-millions-of-lonely-chinese-are-turning-to-for-comfort-9910921.html.<\/p>\n","protected":false},"excerpt":{"rendered":"

Philosopher Vilem Flusser said, \u201cWe do not think about the act of writing while writing, but about what we are writing.\u201d We understand writing as a form of art that comes directly from soul to page, an intimate reveal of human emotion. This core concept of writing is complicated with the development of AI trained […]<\/p>\n","protected":false},"author":44,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"_links":{"self":[{"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/posts\/843"}],"collection":[{"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/users\/44"}],"replies":[{"embeddable":true,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=843"}],"version-history":[{"count":13,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/posts\/843\/revisions"}],"predecessor-version":[{"id":874,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=\/wp\/v2\/posts\/843\/revisions\/874"}],"wp:attachment":[{"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=843"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=843"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/courses.ideate.cmu.edu\/62-830\/s2022\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=843"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}