AI Artists: Training My Replacement 

Adriana Zavala

Introduction and Motivation: 

For my project I wanted to tackle the threat of AI takeover from the point of view of a digital artist. In my original proposal, I was going to create a digital work of art that represented fears and anxieties of what an AI takeover would look like, but then I changed it. I thought it would actually be better to actually initiate a sort of AI takeover than to just imagine it. 

For class we read the Prelude “The Tale of the Omega Team” for Max Tegmark’s Life 3.0, which introduced us to the fictional artificial intelligence program Prometheus who was capable of producing feature-length animation films all on its own. In the story the creators of Prometheus were impressed that the AI could create such a meaningful piece of work just by watching and learning from thousands of films and animations. This story was kept in mind when we later read Martin Rees’ On the Future. In his chapter Humanity’s Future on Earth he warned that AI will one day advance to “replace many white- collar jobs: routine legal work (such as conveyancing), accountancy, computer coding, medical diagnostics, and even surgery” (p.88), which made me wonder what other technology-based job could they replace? The most immediate jobs I thought about were my own, digital artist and graphic designer. These jobs rely on computer programs and software for production, editing, and distribution of creative assets. Could it be possible that one day AI will be able to create digital art and graphic design work all on its own if given just the right amount of criteria from an employer? And if so, should the emergence of ‘creative’ AI be seen as a threat to digital artists or inspire them to use it as a tool to take their work to a new level they’ve yet to reach?

 For further inspiration on what to do for my project to address these questions I did a little digging into existing AI programs that are already taking over the digital art world and graphic design. What I found was AICAN, an AI that creates original digital paintings based on thousands of existing artworks by many artists of varying styles. The works from this AI have sold for over $200,000 in digital galleries as they do not resemble anything recognizable from it’s inputs. Another existing program that I found was a company called Tailor Brands that utilizes AI to automatically develop a logo, color palette, and marketing assets for new companies based on what industry, product, and audience the companies are in the business for. It seems AI is already starting to creep up on digital artists and graphic designers as a competitor, but does that mean we should totally abandon our craft and leave this work solely up to this machinery? For my project, I want to answer the question of whether we should see AI as a competitor in the industry or view it as a tool to excel our business into producing work we couldn’t have before. By answering these questions, I am also hoping to translate my findings to make conclusions about a broader concern for the ultimate AI take-over of human civilization. I think the severity level of AI-takeover could be determined differently depending on the perspective we choose to take on it. 

The way I decided to tackle this issue of AI threat is to lean completely into it and jumpstart the beginnings of my own job take-over. I decided to train an AI program to be my replacement as a digital artist. After training 2 Playform AI models on a total of 88 of my artworks and doodles, they created a total of 10,240 snapshot images of its learning process and 100 final art pieces. Below I have curated 15 of my favorite images that most resemble my artstyle in a completely original way. 

Input:

Output:

FrAInkensteins

FAIces

Process: 

First, I had to find a free (or cheap) AI program or software that would produce images strictly using just my artwork as the input. In researching what was available I kept running into programs like Runway and DeepDreamGenorator that only let the users convert their art into something else using pre-trained AI models that worked from unknown inputs. With a little more digging, I found Playform, a cheap AI program that allows you to train your own model using a minimum of 30 inputs. The models could be trained to synthesize a single input group into original pieces or mesh two different input groups into morphed images. Both options only allowed the AI to be trained on input and didn’t allow the user to specify exactly what the outcome should look like, which is exactly what I was looking to do. I wanted the AI to have complete control over what patterns it finds in my work and produce something similar. 

For my first model (My FrAInkensteins) I trained the model for 2 and a half hours using two sets of inputs, 45 drawn faces and 43 complete works. The lack of specificity and variability of the inputs produced extremely abstract works. However, I think they still carry an essence of strangeness that one could find in my own work. 

In my second model (FAIces) I hoped to receive an output that was a little more recognizable and could be believed to be drawn by my own hand. I trained the second model on just the 45 drawn faces for 5 and half hours. In Playform, the longer you train your model the better the quality of the output will be. It was shocking to see that after 5 and a half hours, my model was creating original faces in my style. In a few pieces the similarity in line work for features like the nose and eyebrows really impressed me, and I only wished I could have trained it longer to see if it could improve its mouths. 

 

Conclusions:

AI as an Artist:

Even though I “pressed the button”, and trained the model on inputs that were created by me, I think it wouldn’t feel right to not give credit to the machine for recognizing patterns across my works and producing original pieces. Especially since we are not totally sure on how AI synthesizes material and draws its connections across inputs, we should give credit for its originality and authentic reaction to what it has been given. AI artworks from my model and ones like AICAN aren’t programmed to output a specific design, so I am curious as to how it chooses what to produce and what patterns to draw from. 

AI as a tool:

Programs like Playform open a door to the next chapter in digital art and graphic design. Not only are the artists and users of AI programs in full control of the input, but we are also in control of which outputs we want to present and how to use them. I see AI becoming a tool that helps inspire us to create new things that maybe we can’t fully picture in our minds, or create pieces in a style that we don’t have the skills to execute yet. Personally, I think I will continue to use AI to help with my brainstorming process and take advantage of programs like Tailor Brand to point me in the right direction of targeting a certain audience in graphic design. AI introduces us to a new genre of art and artmaking that will allow artists to push the limits of their imagination and explore the processes of image-making through model-training. 

AI as a threat:

AI’s threat to digital artists is analogous to AI’s threat to humanity. We could see AI as being our competitor, possibly our replacements, or we could see it as a tool and a collaborator to enhance our existence. What I learned from my experiment with creating my replacement was that AI needs very specific instructions and inputs to output something that we find satisfactory. I think we should be both cautious and specific with what materials we are allowing it to find patterns between, as well as be cautious with the instructions on what to do with the patterns once it does. For example, the FrAInkenstein pieces from my model were not at all expected and were a result of lack of specificity and instruction from the inputs. I see AI as a blank slate only working with what we give it. The use of super AI wouldn’t necessarily develop its own intentions to overtake the world, but in the wrong hands and with not enough training with existing data on how humans live within the world and interact with eachother, the AI could misinterpret instruction to create an undesirable output that could potentially endanger our civilization. As we move forward with advancing AI to aid us in the solving other existential threats, we should remind ourselves that AI is a machine that does not think like us and we should not assume it knows exactly what we intend to solve. We have to be specific with our instructions and the input we give it as we use it as a tool to launch us forward into a better future. 

Bibliography:

REES, M. (2018). HUMANITY’S FUTURE ON EARTH. In On the Future: Prospects for Humanity (pp. 61-119). PRINCETON; OXFORD: Princeton University Press. doi:10.2307/j.ctvc774d5.6

Tegmark, Max. Prelude. In Life 3.0: Being Human in the Age of Artificial Intelligence. , 2017.

Sources:

AICAN

https://theconversation.com/meet-aican-a-machine-that-operates-as-an-autonomous-artist-104381

Tailor Brand

https://www.tailorbrands.com/

Playform

https://www.playform.io/

Runway

https://runwayml.com/