In the novel ‘The Vendor of Sweets’ by RK Narayan, later adapted into the beloved Malgudi Days TV series, we are introduced to Jagan, a traditional sweets vendor whose son Mali has returned from the USA with a novel-writing machine. This machine can create stories based on the user’s input, allowing them to choose the plot, the characters, and the emotions they want to convey, essentially making the creative process simple. However, Jagan, a strong follower of Gandhian principles and a believer in the sanctity of traditional storytelling, is not impressed by the machine at all. He thinks that it is a useless and expensive gadget that destroys the true essence of writing. He believes that stories should come from the heart and soul of the writer, not from a mechanical device.
In today’s world, where technology continues to advance at an unprecedented pace, the debate surrounding the use of Artificial Intelligence (AI) in creative endeavors remains relevant. Many people today have embraced AI as a tool for generating various art forms, whether it be stories, music, or visual art. However, for those who share Jagan’s sentiment, there’s a fear that relying on AI diminishes the authenticity and emotional depth of artistic creations. Nonetheless, for individuals who have experienced writer’s block, faced looming deadlines, or struggled to find inspiration, the necessity of using AI is undeniable.
For years, AI has been regarded as a tool for solving mathematical and scientific problems with a convergent solution approach. Creative works were once considered too complex for machines to address. However, with the introduction of Generative AI, one can now reconstruct literary works much like a creative writer. These models are specialized versions of well-known language models, most notably seen in the form of ChatGPT and Gemini (previously known as Bard) recently.
Poetry as an art form presents a unique challenge for AI. It provides a medium for the artistic expression of thoughts, often using lyrical words and employing figurative language and poetic devices for emphasis, beauty, and eloquence. A structured poem makes use of rhyming schemes (such as AABB or ABAB), meter (the rhythmic arrangement of stressed and unstressed syllables in a line), stanza structure (organizing lines into groups), and alliteration (repeating consonant sounds, often at the beginning of words) for composition. Various types of poems adhere to these rules, but not every poem follows the same conventions.
HERE ARE THE DIFFERENT FORMS OF POEMS BASED ON THE ABOVE TECHNICAL ASPECTS:
1. Blank Verse: Unrhymed poetry in iambic pentameter, often used in dramatic works.
2. Free Verse: Poetry without a fixed meter, rhyme, or structure, allowing for artistic freedom.
3. Couplet: Two-line stanza, often rhymed, providing a concise unit of expression.
4. Sonnet: A 14-line poem with various rhyme schemes, typically in iambic pentameter.
5. Haiku: Traditional Japanese form with three lines and a 5-7-5 syllable pattern, capturing a moment in nature.
6. Limerick: Humorous five-line poem with a specific AABBA rhyme scheme.
7. Villanelle: A 19-line poem with repeated lines (tercets ABA, concluding quatrain ABAA), often exploring complex themes.
8. Ghazals: With alternative rhyming pairs of lines.
9. Rubayee: With four lines in which the first, second, and the fourth lines rhyme.
However, the essence of poetry extends beyond the mere arrangement of words; it delves into the thematic depth, language precision, and the overall quality of its contents. Various forms of poetry based on these aspects include Epic, Narrative Poetry, Ode, Elegy, Lyric, Ballad, and Soliloquy, each offering a unique exploration of human experience and emotion.
Despite the vast array of poetic expressions, Generative AI has yet to fully master the intricacies of creating effective poems across these domains. However, through careful engineering of AI prompts and some user intervention, reasonably good poems can be generated. The prompts for AI-based poetry tools range from wildcard questions like “Please write a poem on a subject X” to requests for the agent to cross-check, improve, or rephrase a pre-composed poem.
There is also an ongoing debate about what constitutes a ‘good’ art, and whether AI-based art can truly be referred to as such. The general notion is that AI is best used as an assistant in the creative process, rather than being the creator itself.
There are structural and thematic challenges for AI-based systems to write high-quality poetry. This includes the ability to create complex rhyming schemes. For example, simple rhymes occur when the last words of two consecutive lines have similar sounds. The matching sounds are typically the final stressed syllables of each word. Complex rhymes on the other hand involve a more intricate matching of sounds, extending beyond the final stressed syllables. This can include matching vowel sounds, consonant clusters, or even multiple syllables. Words like ‘moon’ and ‘croon’, or ‘grumble’ and ‘stumble’ showcase complex rhymes due to similarities in both vowel sounds and consonant combinations. Experts in AI-based poetry have been grappling with compiling databases and computational algorithms to enable machines to write text using simple and complex rhymes with different structural organization.
HISTORICAL BACKGROUNDS AND STATE OF THE ART
Many of us are aware that the first questions regarding man versus machine intelligence were posed by Alan Turing in the form of his famous test, named after him. Machines of that era patently failed the tests of intelligence in his experiments. However, modern machines have evolved to become intelligent and can often beat the Turing test. Researchers have expanded upon the concept of human intelligence inherent in the Turing test by designing the Turing test for poetry. The Turing Test Poetry Fest is a poetry competition where humans pretend to be AI machines, and AI programs, in turn, pretend to be humans. The Turing Test Poetry Fest welcomes all kinds of poets, both human and non-human alike (https://turingtest.mit.edu/).
Another similar initiative is called OULIPO. Oulipo, short for “Ouvroir de littérature potentielle” (‘Workshop of Potential Literature’ in English), is a literary movement founded in 1960 by Raymond Queneau and François Le Lionnais. Writers associated with Oulipo use systematic approaches, including mathematical principles and linguistic constraints, to explore the vast potential of literature. Notable examples of Oulipo’s influence include Georges Perec’s novel La Disparition, written without the letter ‘e’, and the use of techniques like ‘N+7’, where words are replaced with the seventh word following them in a dictionary.
Image Courtesy: Flickr
Formal attempts to write AI-based poetry are relatively new but numerous. For example, in the work Hafez: an Interactive Poetry Generation System (http://dx.doi.org/10.18653/v1/P17-4008), the authors presented an interactive poetry writing system that can incorporate rhythm and rhyming schemes provided by the user.
In another work, authors introduced Deep-speare: A Joint Neural Model of Poetic Language, Meter, and Rhyme(https://arxiv.org/pdf/1807.03491v1.pdf) that captures language, rhyme, and meter for sonnet modeling. They assess the quality of generated poems using crowd and expert judgments.
Another tool for generating sonnets can be found at https://www.poem-generator.org.uk/sonnet/. Additionally, the AI Poem Generator (poemanalysis.com) offers analysis and interpretation of various poems, along with a free AI poem generator capable of creating poems on any subject. Users can input their desired poem topic and click generate, with options to customize tone, form, themes, length, and language. This tool quickly produces a unique and creative poem tailored to the user’s specifications.
Apart from these specialized tools exclusively dedicated to writing poetry, generic language tools such as ChatGPT by OpenAI and Bing Copilot by Microsoft are very helpful in generating de novo poetry or fine-tuning a pre-composed poem. Bing Copilot has three modes: creative, precise, and balanced. Each mode suggests things to the user differently, depending on their needs and preferences. The creative mode is for when the user wants Copilot to be more imaginative and inventive. It generates content for the user, such as stories, poems, jokes, or images. Gemini (formerly Bard) by Google and Grammarly AI are less powerful in these tasks but do provide alternatives that are likely to improve.
Image Courtesy: Shutterstock
Machine-based poetry and its future potential are being viewed with appreciation and fear. Will this impact the professional work and employability of some of the best brains of our times? As of now, AI-based poetry is only attracting more audiences to this form of art, and the poets adopting these tools ahead of others have an advantage. It will take a long time before machines can replace the best poets if they come anywhere close at all. The first casualty of AI-based poetry may be the lyrics used in movies and albums, but large-scale adoption is still too far. Moreover, human oversight and collaborative work with AI will probably remain there forever.
*Richa Mishra holds a master’s in Medicinal Chemistry and Drug Design, and is a PhD student at SciWhyLab, SCIS, JNU. Prof Shandar Ahmad specialises in the field of Bioinformatics and Data Science. He is a former Dean and currently a Professor at School of Computational and Integrative Sciences (SCIS), JNU where he also leads the research lab called the SciWhyLab. He can be contacted at shandar@jnu.ac.in
