Conjuring Dragons with AI

By Gary E. Richardson, Y62 Communications Team member

The explosion of interest in artificial intelligence (AI) over the past few months has been phenomenal, especially since the release last November of Open AI’s ChatGPT, which has already been superseded by GPT-4. These pioneering AI neural networks are fraught with inaccuracies—confident responses that don’t seem to be justified by the chatbot’s training data. Like many, I’ve had mixed results with Chat GPT: I’ve learned a few things I did not know or had forgotten. The bot appeared to carry on “intelligent” conversations on several topics of interest. However, chatbot can’t tell me where it gets its information and ducks some areas of discussion with a set response:

As a machine learning model, my knowledge is limited to the text that I have been trained on, and I do not have the ability to browse the internet or access additional information. The information that I have provided is based on a general understanding of the topic, and it is not intended to be a definitive or comprehensive overview…

Occasionally, it will then follow on by offering its “opinion” anyway.

When one begins conversing with a chatbot, it is difficult to avoid anthropomorphizing the transaction. ChatGPT’s tendency to make up facts or present information that, in one way or another, does not correspond to reality is called hallucination by researchers and laypeople alike. Others prefer fabrication—something made—or confabulation — a psychological term meaning “generation of a false memory without the intention of deceit”—which adds an air of innocence to the bot’s occasional and sometimes egregious errors.

Several weeks before the announcement of ChatGPT’s public release, I had stumbled onto Midjourney, a similar form of artificial intelligence. Instead of generating a verbal response to a written query, Midjourney produces visual responses to verbal prompts. As a photographer steeped in the history of art, I was and still am intrigued.

For me, creating images with Midjourney is like probing the subconscious of a group mind; the hallucination metaphor may be appropriate. Over time I may learn some of the code and prompt phrasing get more control over the outcome. Below, are some of my early prompting experiments and their results. I initially called this series Mid-conjour-neying Dragons. In the oriental horoscope, like anyone born in 1940, I am a metal dragon. The dragon is a mysterious mythological symbol of great power. One of my first Midjourney prompts was (typo included):

Imagine: a five-toed metal dragon emerging from Chaos preceded by a sinning orb of pure, near ultraviolet energy:

Midjourney returns a four-image grid. I ask it to enlarge and to produce variations of the images. None of these images is what I was looking for, so I revised the prompt, correcting the typo and adding the word “Oceanic”:

Imagine: a five-toed metal dragon emerges from an Oceanic Chaos preceded by a spinning sphere of pure, near ultraviolet energy:

That was not what I thought I wanted either, but it is intriguing. If you look closely there are really no dragons, but the bot has captured the essence of a dragon-like creature evolving inside the orb and beginning to peak out into the world. It’s these images that brought conjuring to mind. Finally, in my quest for dragons, I tried yet another prompt, calling on the Buddhist bodhisattva of great compassion:

Imagine: Quan-Yin and a five-toed metal dragon emerge from the Ocean of Chaos:

Quan-Yin never shows up, although here is a suggestion of her silhouette in the upper left. After asking for a set of variations and up-scaling the one on the lower right, I decided to call it a night. I had conjured a pair of creatures that from a distance can pass, but if you look closely what I had conjured were not what they seemed.

 
We welcome your comments below. Thanks.