Can Artificial Intelligence be an Artist?
Tried to tackle this problem from the complete AI beginner.
First of all, I tried from making actual AI system!
These days, there are a lot of softwares and websites that enable us to experience making AI system. Runway ML and teachable machine are the typical ones. I tried from creating image classification on teachable machine and then created an image generate model on Runway ML.
For Teachable Machine….
・A Bad Rock Paper Scissors Machine
We built a team of 5 and made a image classification game called, “A Bad Rock Paper Scissors Machine.” We collected 1000pictures of each hand gestures and set them to teachable machine. Then we made a Website in software called glitch and created a game. By activating the website, the image input from the Web Camera are classified in to rock, paper, or scissors by a AI system that we built. In the game, websites always wins according to what player shows.
・Magnify your face!
Next we changed the input datas from hands to emotions, we’ve selected 6 emotions and took a photo of ourselves. We collected 1250 images for 1emotions as a data set. We also used teachable machine and glitch for this project. Websites classifies the facial emotions. Then the backgrounds and filters exaggerate emotional expression.
Hands-on fabrication of AI system enabled me to learn how the AI system works and how important the type of data is to the purpose.
Runway ML allows us to make a GAN(Generative Adversarial Networks) easily. Our team challenged to generate imaginary manga covers by Machine Learning.
1.We used actual Manga covers as a data set.(500 covers from mainly Jump Comics)
These are the samples from the data set
2. Using a Style GAN in Runway ML, we let the system learn the dataset.
→The more system learns, more precise the generated image be close to data set.
Unfortunately, because of the free model that we were using, we could take only 1500 steps.
3. Show the results!
The results were very interesting. It kind of reminds us of the title and illustrations, although it’s out of human creativity or perspective.
(please be aware, if you are easily get disgusted by images)
Generated images transformation
to find more please visit our websites.
Creativity and AI have complex implications. The reason why is that many factors can’t be expressed by words or statistics. Although, by unraveling each tangled unexplainable factor by history and other aspects, it shows us an overall picture of it. Also, it intrigues our speculative mind of being creative by the usage of AI.
First of all, Creative Adversarial Networks (CAN) is a modified version of GAN. It intentionally decreased the function of style classification and style ambiguity to express more creativity. According to Colton 2008, the criteria of the creative system are the ability to produce novel artifacts, to generate quality artifacts, and to assess its creation. The former two principles are implemented by the adjustment of GAN. The last principle leads to the function of a discriminator which evaluates the generated work as “art or not”. Besides, it also classifies whether it is a part of established styles. For those functions, it enables to generate an artwork.
When it comes to the discussion of creativity and AI, it becomes complicated because it claims to us what is creativity and what is art. According to these papers, I saw some points in common. Although it is based on artificial intelligence, the human always stands in the background. For example, the art ordinarily has to have some perspectives of empathy and ethical consideration that move “human” inner consciousness. From the CAN experiment, they chose the subject “human” to evaluate the artwork which doubts us is the “human” perspective the last subject to judge creativity.
Thorugh those understandings, I think that the usage of AI to increase creativity has more options and possibilities. As the “Can Computers Create Art?” stated, we can widen the possibilities by coming up with new ideas of using CAN or AI in new and unpredictable ways. In my opinion, human creativity should be reflected here the most in all processes. The unique creativity by AI will be established by doing so.
Changing the biased idea of AI is one of the options. The arts generated by algorithms are said to be brittle and bespoke. On the other hand, artworks created by infants are also brittle and bespoke. Although, it has some art contests for them and also there are few “human” who ironically criticize them. So what is the difference? I think it is because of the influence on the inner consciousness of humans. People tend to expect warmth and effort in paintings. On the contrary, people have a perception of cold, easy, or fast toward artificial intelligence because it sounds there is no space of human intervention. To tell the truth, there is a lot of human intervention in AI. For example, choosing what sources to let the machine learn, setting parameters, and so on. I think these efforts are equivalent to infants’ artwork. Therefore, knowing the actual system of AI would change the appearance of AI artwork to people.
Next, I suggest it would be interesting if the human learns our creativity by using CAN. The 2 papers both introduced the future of making artwork by technology. Although, developing creativity is not only limited in outgoing the work to society but learning by creating. Going back and force through analog and digital is an effective way to enhance our creativity. When I learned about UI designing, I started sketching the popular smartphone applications UI design by pencil and paper. It made me realize the composition that enables a convenient user experience. Also, there is an artist called Kei Imazu. She creates her paintings after sketching on Adobe Photoshop. She collects images from the internet and edits a collage on photoshop. Then she recreates the collage and the unique effects of photoshop on canvas by painting. From those examples, technology gives us new clues and perspectives of creativity. AI can sufficiently be one of them. I believe art pieces made by getting inspiration from AI-generated art can be very creative.
Lastly, changing the whole idea of “art” through AI would give a new era of style. When I read these papers, I wondered why they are trying to make a CAN work as close to existing artwork as they can but not imitating. If AI also learns data that are not current paintings, such as sceneries and etc, they would be more surprising and emergent. This could give new values to AI-generated art.
Resources:
-CAN: Creative Adversarial Networks, Generating “Art” by Learning About Styles and Deviating from Style Norms
Ahmed Elgammal, Bingchen Liu, Mohamed Elhoseiny, Marian Mazzone
-Can Computers Create Art?
Aaron Hertzmann
