Painting with Artificial Intelligence

Painting with Artificial Intelligence

by Andreas Kohli

Can a machine be creative? Can it create images autonomously? If we understand artificial intelligence (AI) as a multi-faceted mirror of ourselves, this opens up new perspectives on the concepts of consciousness, intelligence and, creativity.

Numerous tools offer creative interactions with AI software, thereby providing insights into the mechanisms and values underlying currently available AI applications. These devices allow hand-drawn sketches to be transformed into photorealistic images with one mouse click. Such programs generate images of faces or landscapes, for example. They are easy to use yet still allow some control over the results and variations.

Technically, a neural network uses hand drawings as image cues and generates photorealistic images from the sketches. These are based on extensive collections of sample images.

These simple possibilities of creative cooperation with an AI open up room for experimentation, and at the same time, for reflecting notions of authorship and creativity. Besides, it quickly becomes apparent how existing cultural and social expectations shape algorithmic systems and which topics and pictorial worlds are included or excluded by the AI (algorithmic bias).

  • Introduction:
    • The topic can be introduced by briefly discussing a work of art created by AI. A good example is the “Portrait of Edmond de Belamy” by OBVIOUS Collective. In 2018, the “painting” was sold for $432,500 at a Christie’s auction. The ancient-looking “painting” auction prompted questions about authorship and copyright.
    • Duration: 10–15 minutes
  • Part 1: Testing
    • In small groups, the students work with a tool that generates photographic images based on freehand sketches (e.g., DeepFaceDrawing for portraits, GauGAN or NVIDIA Canvas for landscapes).
    • They either draw directly online or upload previously made sketches to test the tool’s image-generating capabilities.
    • Goals can be, for example
      • to generate “error-free” images that are as photorealistic as possible;
      • to explore the extremes, the possibilities of the tool, and to provoke errors
      • to substantiate their image ideas, for example: by producing a self-portrait
        Screenshots are used to record the sketches and results.
    • Duration: 20–30 minutes
  • Part 2: Evaluation of the tests
    • Based on their tests, the students discuss the questions below within their group, take notes and support their thoughts with screenshots.
    • Questions may cover the following:
      • Functionality and usability, possibilities and limitations of the tool
      • How can I trick/hack/use the tool differently?
      • Factors inherent to the system (how do the interface design and the given functions influence my expectations, actions, the work process, the result, …).
      • From which visual worlds does the AI generate the image? What values underlie these visual worlds? What is included and what is excluded?
      • Is the tool valuable for my work? What would need to be changed to make it usable?
    • Duration: 20 minutes
  • Part 3: Discussion
    • The answers are compiled and discussed in a plenary.
    • The following topics can be favored during the discussion:
      • Who is the author of the image?
      • What are this AI’s underlying values (algorithmic bias), and how is an AI trained in general?
      • Can and/or should AI produce art?
      • How can I use these tools? Does this way of “painting” make sense? Are the results independent works of art, or do they merely serve as suggestions?
    • Duration: 20–40 minutes
  • Interdisciplinary cooperation with the subjects German (on chatbots) and music (on computer-generated music) makes sense.

Additional literature on authorship:

Additional literature on Algorithmic Bias:

Additional links regarding the topic of AI training:

The picture is divided into two parts. On the left, there is a drawing of a man wearing glasses. To the right, there is an AI-generated image of what a passport photo of the person drawn might look like.
Portrait 1

Portrait 1 © 2022 von Andreas Kohli und DeepFaceDrawing] unter der Lizenz CC BY-SA 4.0, Zürcher Hochschule der Künste, Schweiz.

The picture is divided into two parts. On the left, there is a drawing of the outer shape of a face and two eyes. To the right, there is an AI-generated image of what a passport photo of the person drawn might look like.
Portrait 2

Portrait 2 © 2022 von Andreas Kohli und DeepFaceDrawing] unter der Lizenz CC BY-SA 4.0, Zürcher Hochschule der Künste, Schweiz.

Andreas Kohli is a lecturer in the Department of Cultural Analysis at the ZHdK – Zurich University of the Arts. He also works as a designer in his studio Belleville.ch.

At ZHdK, he teaches in the Bachelor of Arts Education and transdisciplinary modules. His areas of interest include urban spaces and their media representation, art and artificial intelligence, ecology, interculturality.
Other topics include the development of teaching materials and e-learning content.

He has initiated and led numerous cultural projects in Europe, Africa, and Asia and is a member of the Shared Campus / Critical Ecologies 生態思 group
(link: https://shared-campus.com/themes/#critical-ecologies).

The course has been run several times with students at ZHdK since 2020.

Prior Knowledge and Preparation
Prior testing of the current availability of the AI programs by the teacher; possibly researching alternative tools.

Accessibility:
Assistance for Learners
An introductory sequence that discusses a work of art created utilizing AI raises awareness of the topic and the context.
Afterward, the students test a tool along given topics and discuss their findings in a plenary.

Additional Tools
Browser-based applications (recommendation: Chrome):