Phillip Kalantzis-Cope – Machine Learning

Review by Paul Anderson •

Two questions come to mind when looking through the 2022 photobook Machine Learning by Phillip Kalantzis-Cope. First, can ”machines” learn a specific task, and second, is this productive learning? The author, Kalantzis-Cope, presents us with ten examples of a specific kind of “machine learning.” In each example, he provides a single titled image (we will call this the reference image) accompanied by four associated untitled images.

The four untitled images have presumably been generated by a “machine” (a computer and its related software) based on the machine’s past learnings and a look at a reference image. There is no text in the book to confirm this presumption, but the book’s title leads us to assume this is the case.

Machine learning^[1] is an application of artificial intelligence (AI). In these examples, the machine has been trained to create four new images that are similar to, yet distinctly different from, the reference image. The 1975 Random House College Dictionary defines the verb “to learn” as “to acquire knowledge of or skill in by study, instruction, or experience.” The Oxford online dictionary defines machine learning as “the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data.”

Thus, we are concerned with computer systems that have acquired knowledge or skill without following explicit instructions. In the cases presented here, the acquired skill is the ability to generate new images that are similar to a reference image – that is, the machine produces images that have similar objects, physical relationships, textures, and colors to the reference image.

Assuming that a computer system did generate the four images associated with each reference image, then the answer to the first question posed above is yes – the machine has learned to generate images similar to a reference image.

Now let us consider whether or not these generated images have improved upon the aesthetic value of the original image or are of practical use.

From an aesthetic point of view, it is fair to say that the generated images are on a par with the reference image. The machine seems to take its aesthetic queue from the reference image, neither degrading it nor improving it, with the possible exception of the quirkiness of generated text-like elements. Although it is clear that this type of image generation is quite a technical accomplishment, it is perhaps an empty one if used as an end product. That is, what is the point of merely generating images that are variations of a reference? One possible use might be to suggest compositional variations upon which an artist can further improvise.

From a practical point of view, the generated images do not seem to add further insight into, for example, railroad crossings or the aesthetics of wind turbines (see the attached images). One might be able to use this process to generate alternate images for, say, advertising purposes, but that is fairly limited. The reference images could serve just as well. The reference images seem sufficient by themselves to convey a sense of place.

Current AI tools and applications do have a productive place in the arts. For example, AI tools can very expertly sharpen photographic images, or increase the resolution of a digital image while respecting the original lower resolution image or generate images for backgrounds or for insertion into other digital images, or the generation of newly-imagined scenes. Also, manipulation or inclusion of machine generated images like these into other works can greatly increase or personalize an artwork’s aesthetic value.

The value of this book is that it brings these issues to our attention, and by doing so asks us to consider how these tools are used, and how they could be made more productive.

This is a small, informal soft cover book. It will be of interest to those who work with AI tools in the field of the arts.

____

Phillip Kalantzis-Cope has been previously featured on PhotoBook Journal: Middlescapes.

^[1] https://azure.microsoft.com/en-us/resources/cloud-computing-dictionary/artificial-intelligence-vs-machine-learning#:~:text=Machine learning is an application, its own, based on experience.

Paul Anderson is a photographer/digital artist, working in Hermosa Beach, CA

______

Machine Learning, Phillip Kalantzis-Cope

Photographer: Phillip Kalantzis-Cope; born Athens, Greece; currently resides in Champaign, Illinois.

Text: English

Softcover, saddle stitch, 5.8 x 8.3 inches, 48 pages, limited edition of 100, Printed in Champaign Illinois, USA, ISBN: 978-1-7355008-6-7

Book designer: Phillip Kalantzis-Cope

____

Articles and photographs published on PhotoBook Journal may not be reproduced without the permission of the PhotoBook Journal staff and the photographer(s).