Artificial

Spent a few minutes this afternoon experimenting with Ask AI, an AI-powered chat-bot assistant that includes functionality to generate images from a text prompt using artificial intelligence.

Here’s the text that I provided the image generator:

A morning in a cobblestone alley behind a mid-century church. The morning sun casts a shadow of a utility pole and power lines across the back of the church. The two-story wall has two stain glass windows on the second floor protected by plexiglass, and two smaller windows on the basement floor protected by steel grates. One of the lower windows has an old window air conditioner unit installed. Beige paint is starting to peel away from the cinderblocks. Someone has left two mattresses leaned against the wall. Weeds and some bare dirt make a strip of green and brown along the church wall. Overall the view shows a slightly run-down scene.

I wasn’t happy with the first image, which was too gothic, so I asked the generator to try again with more detail. For some unknown reason it decided the translate my initial prompt into French:

Une matinée dans une ruelle pavée derrière une église du milieu du siècle. Le soleil du matin jette une ombre d'un poteau utilitaire et de lignes électriques sur l'arrière de l'église. Le mur à deux étages comporte deux fenêtres en verre teinté au deuxième étage protégées par du plexiglas, et deux fenêtres plus petites au sous-sol protégées par des grilles en acier. Une des fenêtres du bas a une unité de climatisation installée. La peinture beige commence à s'écailler des parpaings. Quelqu'un a laissé deux matelas appuyés contre le mur. Des mauvaises herbes et un peu de terre nue forment une bande verte et brune le long du mur de l'église. Dans l'ensemble, la vue montre une scène légèrement délabrée avec un mélange de textures et d'éléments qui évoquent un sentiment d'abandon et de négligence.

Here is the second, revised, result:

Scene generated by Ask AI

Unsatisfied, I tried Midjourney, another popular AI generator option.

Scene generated by Midjourney

Compare & Contrast

For the experiment I decided to use a text description of an image that I took yesterday morning. while not expecting a result that would mirror the original inspiration—I was a bit surprised at the result. The generator ignored several aspects of the description, and the resulting scene has been visualized as much older, and more romantic, than I anticipated. Perhaps it would have been more-helpful to specify an American city, or to specify in the style of Walker Evans or Lee Friedlander. I

My digital photograph

What is this AI-thing?

As with many images created using artificial intelligence I am left with two overall impressions:

• Hyper-realistic presentation

• A sense of Déjà vu

This type of AI is still relatively new, and we can expect it to rapidly improve, but for now I am left with the feeling that it is something of a confidence game. Like the mirror of Narcissus AI shows us what we want to see, based on something familiar. And like a good con artist, or fortune teller, AI uses those fragments to construct a reality that impresses us because it is both realistic and familiar.