Artificial Intelligence - AI and the Art of Haiku

An Art-Tech Article on Generative-AI and Haiku

Example: Realistic digital art and its different versions created by Generative-AI (see below)

Generative Artificial Intelligence (AI) is an in thing or a buzzword in technology circles these days. The technology or Generative-AI is reaching out and being tried for all types of writings – poems, stories, blogs, and commercial marketing material, with varying degree of success. On the other hand, another type of Generative-AI is used for converting text lines to images and the visa versa as well. The article by Connie Gregory on AI and the Art of Haiku is thus at the interface of the technology and art or Art-Tech in general.

I began writing haiku several years ago as a way to annotate photographs that I had taken. Traditionally haiku are written as a way to evoke an image of feelings within the mind of a reader.  Initially my motivation was just the opposite, I had images or set of photographs and wanted words to describe those in unique way.  As a way of being consistent with my annotations, I chose to adhere to a strict 5-7-5 syllable in 3 lines format. The Modern haiku are often more free form than not, but traditionally they were constrained to this above specific syllabic structure. In free forms, for English writers, 17 total syllables are written in three lines with no counting of syllables in each line.  The coupling of a haiku poem with a visual image is called a “haiga”, which is a haiku plus a painting, a photograph, or an image to go with that.

Figure 1. A digital photo of a sunrise over a salt marsh, annotated with a suitable haiku.

After having produced hundreds of original haiga with my photos and haiku, I have arrived at a point where my objectives and limitations suggest a transition and expansion into new challenges.  Then looking at the reverse process, there are memories or experiences or ideas for which personal photographic documentation does not exist.  Creating a haiga prose continues to be satisfying but I am not able to produce an original painting or drawing for each poem. Even if I were a painter or a sketch artist, it will be time consuming as well. 

New and now widely available artificial intelligent (AI) technology offers a way to generate images or art from descriptive text. There also is software that will generate poems, stories, or even articles from prompts provided by the creators. Both are part of the branch of AI called Generative-AI, which is becoming very popular these days.  Personally, I find the challenge of using my own language skills to create the prose of a haiku extremely satisfying. The AI figures shown in this article were generated by the AI program by DALL·E 2, an open source AI program available from Open-AI, that can create realistic images and art from text base descriptions or descriptors.

Descriptor Sensitivity in Generative-AI Art Output:
Thinking about a visual experience, I had on a recent walk, led me to a simple example of the effect of descriptor choices on the nature of the AI-art produced.  On my walk along a dirt road, I encountered an attractive vista of a group of cedar trees arched over the road.  Reflecting on my recollection of the view, I composed this landscape inspired traditional 3 line haiku in 5/7/5 syllables format – “Above a dirt road”… “Nature crocheted cedar shawl,”… and “Shaded promenade”. In order to produce an artwork to complement the haiku, I began with the very basic text,  “cedar trees arched over a dirt road”.    Figures 2 a,b,c, and d illustrate the sensitivity of the AI output to various alternative enhancements where is (a) a realistic digital art , (b) in the style of Vincent Van Gogh, (c) in the style of Salvador Dalí and (d) in the style of Thomas Cole. Here the choice of which one to use finally becomes largely a matter of the personal preference.  In many instances, the choice will be obvious.

Figure 2 Affect of the descriptor on AI output on the art produced of cedar trees arched over a dirt road, (a) realistic digital art (b) in the style of Vincent Van Gogh (c) in the style of Salvador Dalí (d) in the style of Thomas Cole

Chop-Suey and Cream Puffs:
As a second example, thinking about my childhood, growing up in a small town of Avon, Massachusetts (with a 1950s population of approximately 2700), one of the” exciting” things, I looked forward to doing was the occasional family shopping trip to a bigger “city” of Brockton (with a 1950s population of approximately 60,000), a whole 5 miles away. I would eagerly anticipate our shopping in the BPM, i.e. the Brockton Public Market, for the then times, a rather “large” grocery store for us.  To this day, many years later, I have two vivid memories about that downtown market, a store worker in the window filling cream puff shells with whipped cream, and a Chinese cook stir-frying a batch of Chop-Suey in the produce section of the store. Best of all we got to take home some of both.

So, I wanted to create a Haiga about this memory. Writing a haiku that corresponds to my memories comes naturally – “Sweet and savory”… “Gastronomically fine”… “A matter of taste”.  However, I do not have any personal photographs to complete an original haiga of the memory.  Using stock or public domain illustrations does not work for me.  I feel that my work would be a diluted, i.e., become hybrid, as compared to being completely original. The availability of AI text to image technology provides a new and exciting way to create original art.  It is original because the creativity of my textual “instructions” determines the quality of the output. In AI jargon, it is also called “layering” of the input textual instructions.  At the present time, the process does have a degree of trial and error during the “layering” of the textual input, thus producing a satisfactory product may require multiple refinements of my own descriptive text.

Figure 3. Grocery Store On Site Cooking …Stir Fry Chop Suey and Home Made Cream Puffs  circa1950.

Wine and Al Capone
Lastly, recalling a story about my spunky maternal grandmother provides an opportunity to create a graphic to go along with a Tanka, i.e., a haiku expanded to 5-7-5-7-7 format with 5 lines instead of 3 in a haiku.  My maternal grandparents came to the US circa 1920 from Italy, from Benevento in Campania, the home of the Amalfi coast and Pompeii. They settled shortly after on a small farm in North Randolph, Massachusetts. The farm had a vineyard and thus the tradition of homemade wines followed in our family.  They also sold some of their wines to neighbors for extra income.  But it was the era of the prohibition, and one of the neighbors reported them to the Bureau of Prohibition as it was the time of the famous Eliot Ness and his team of Untouchables shown in the movie Untouchables.  As the story told to me by my mother, she remembered a visit to the farm by a couple of Prohibition police.  She was about fourteen at the time.  The family was at the diner table when they heard a knock on the door.   My quick thinking grandmother grabbed the bottle of wine off the table and hid it in her bosom.  The agents, not finding any evidence of a prohibition violation, had to move on and go after an easier target – perhaps some guy named Al Capone. Disclaimer, we were or are not related in any way to the Al Capone.
Bottle of Wine Tanka
wine on the table
revenuers knock at door
bosom expansion
busty italian lady
or dal seno prorompente
Figure 4  Italian lady with a bottle of homemade wine 

Figure 4.  Italian lady with a bottle of homemade wine 

Written by Constantine Gregory

Editor’s Note: The field of Generative AI, with programs like ChatGPT-3 and DALL E.2, is only a few years old, but getting popular by leaps and bound. We thank Constantine Gregory, a regular contributor in our group, for writing this article on our request.