By Sarah Yung
Machine intelligence is approaching the new frontier - creativity. Machine intelligence has been a game changer in healthcare (among numerous other fields), helping identify cancerous growths, protect patient records, and assist in surgery. Now, teams of researchers from all over the world are exploring other fields for computer intelligence to exercise its new skills. Much of this technology pushes the envelope in artificial intelligence. Human creators like Da Vinci and Tesla set a high bar for any software to match. Researchers are taking fledgling steps into this field, but their results show promise for powerful tools in the future.
Although computers that can create their own work are unprecedented, tools like Grammarly and SpellCheck are already frequently used by writers to make sure they can properly convey their ideas. Additionally, there are promising developments in programs that will protect authors and their work. Emma Identity, for example, is a self-learning technology that can detect authorship by analyzing writing style. Artificial intelligence programs today allow creators to focus on developing their ideas instead of nitpicking over syntax or protecting their work. New developments will hopefully continue to augment the ability of today’s creators in new and significant ways.
More recently, Microsoft is leading the charge into developing creative technology. Currently, its researchers are focusing their attention on poetry. Their recent successes are notable because poetry is an especially challenging form of information synthesis. Poetry involves a certain level of conceptualization and abstraction - instead of direct descriptions, poets make references to people, places, and objects with similarities. The AI system Microsoft XiaoIce has been trained to write poetry from keywords. The system was developed in 2014 and has continued evolving at an increasing rate. In 2017, a Chinese publishing company released Sunshine Misses Windows, XiaoIce’s first-ever poetry collection. This anthology included over 10,000 written poems accomplished in 2,760 hours, of which 139 were selected for publishing.
From XiaoIce, Microsoft researchers turned their attention to more challenging projects. Another project generates poetic language in response to images. To do so, experimenters fed the machine image poem pairs from a large poetry database. The researchers tested and refined the software’s poetry on over 8000 images, which were then evaluated by both machine algorithms and human readers. Researcher Bei Liu has her own favorite poem created during the study, paired with the image to the right:
Some researchers are focusing their attention to other forms of writing. Ross Goodwin and a team of scientists created a screenwriting software with Long Short-Term Memory. Long Short-Term Memory is an artificial recurrent neural network. A recurrent neural network has a feedback loop connected to past decisions, unlike feedforward networks, which feed information straight through the algorithm once. This computer wrote the short film Sunspring. Sunspring is the story of three people - H, H2, and C - living in a futuristic world, where they are entangled in a love triangle. Sunspring, directed by Oscar Sharp, was presented at Sci-Fi London, where it was selected as one of the 10 best short films.
Novels written by computers are also making ripples in the literary world. A team from Future University Hakodate in Hokkaido, led by professor Hitoshi Matsubara, developed an AI that entered a short story as a candidate for the Hoshi Prize. The Hoshi Prize is a Japanese science fiction award, one of the only competitions to allow entries from computers. The short story entered by Matsubara’s team - “A day when a computer writes fiction” - made it past the first round of judging. After that round, the judges decided it did not compare to its human counterparts. Researchers found that, while the computer could emulate Hoshi’s writing style fairly accurately, it could not create good plots. For the competition, humans handled the plot creation, then the AI wrote the story. Although the story passed as human writing, the software still has a ways to improve to match its human counterparts.
Google is also diving into artificial intelligence, founding the Google Brain research team in the early 2010s which took a slightly different path, diving into the field of music. Their Magenta Project is a research project that “explor[es] the role of machine learning as a tool in the creative process.” The Magenta Project utilizes machine learning techniques to develop a gallery of machine-made art and music, which is continually updated to this day. Magenta uses a combination of deep learning algorithms and reinforcement learning algorithms for its creative process. Deep learning algorithms train off existing data, improving with each cycle of new data, while reinforcement algorithms are trained repeatedly on the same set. After developing their algorithms, they released their models and tools as open-source. The team continues to collaborate with the public to modify and add to Magenta software. They recently worked on developing long-term coherence in music with patterns and themes and introducing more interfaces. for more people to interact with.
As more work is being done towards developing creative computers, some creators have expressed fear towards the development of such technology. Compared to their counterparts, computers can churn out work at a rate that far surpasses any human. For example, the AI system XiaoIce wrote over 10000 poems in 2760 hours. However, right now, and far into the foreseeable future, computers cannot genuinely feel complex human emotions, only mimic them. Some feel that because of this, instead of replacing authors, AIs should replace literary agents, editors, and publishers. Using repositories of published works, AIs could critique released work based on information in their databases. Even if AI doesn’t go into the field of publishing, the researchers behind these developments don’t intend to replace human authors.
Far from replacement, these teams intend for AI to augment creative activity. While full-fledged, creatively, uniquely thinking artificial intelligence is far in the future, the capability of today’s technology should not be ignored. In many ways, humans and robots are collaborating to put forth the best material possible. Humans are constantly innovating and changing the rules, and today’s software has a long way to go to match that. Right now, computers provide a medium to communicate concepts and ideas. Soon, they will become an integral part of the creative process. Let us approach this new frontier boldly - who knows what lies beyond!
“Artificial Intelligence In Creative Writing : A Curse Or A Blessing For Authors?” The Bookish Elf, 8 July 2018, https://www.bookishelf.com/artificial-intelligence-creative-writing/.
Duncan, Joe. “The Future of Writing in the World of Artificial Intelligence.” Medium, 10 Mar. 2019, https://writingcooperative.com/the-future-of-writing-in-the-world-of-artificial-intelligence-9ca9b6babb9c.
EMMA. Defining Writing Identity. Disrupting Plagiarism.https://emmaidentity.com/. Accessed 17 Aug. 2019.
Lewis, Danny. “An AI-Written Novella Almost Won a Literary Prize.” Smithsonian, https://www.smithsonianmag.com/smart-news/ai-written-novella-almost-won-literary-prize-180958577/. Accessed 11 Aug. 2019.
“Magenta.” Magenta, https://magenta.tensorflow.org/. Accessed 11 Aug. 2019.
Marr, Bernard. “Artificial Intelligence: What’s The Difference Between Deep Learning And Reinforcement Learning?” Forbes, https://www.forbes.com/sites/bernardmarr/2018/10/22/artificial-intelligence-whats-the-difference-between-deep-learning-and-reinforcement-learning/. Accessed 11 Aug. 2019.
Otake, Tomoko. “Japanese Researchers Take Artificial Intelligence toward the Final Frontier: Creativity.” The Japan Times Online, 19 June 2016. Japan Times Online, https://www.japantimes.co.jp/news/2016/06/19/national/science-health/japanese-researchers-take-artificial-intelligence-toward-the-final-frontier-creativity/.
Sunspring. www.imdb.com, http://www.imdb.com/title/tt5794766/. Accessed 11 Aug. 2019.
“The Poet in the Machine: Auto-Generation of Poetry Directly from Images through Multi-Adversarial Training – and a Little Inspiration.” Microsoft Research, 18 Oct. 2018, https://www.microsoft.com/en-us/research/blog/the-poet-in-the-machine-auto-generation-of-poetry-directly-from-images-through-multi-adversarial-training-and-a-little-inspiration/.