AI and Creativity
Final Assignment of Interaction Design 2020
1. What is Creativity?
Can Artificial Intelligence be creative? — This question has been argued many times until now, but we have not found a clear answer yet. This is because we do not standardize the definition of “creativity” when we discuss it.
Margaret A. Boden, a researcher of computational creativity, defined that creativity is the ability to come up with ideas or artifacts that are new, surprising, and valuable. And he mentioned that we should focus on how creative the idea is and in just which way(s), not focus on whether it is creative or not.
Furthermore, he suggested that creativity can happen in three ways.
- Combination: To produce novel combinations of familiar ideas. (e.g.: Making a collage of some images.)
- Exploratory: To explore the potential of conceptual spaces. To produce something in the way of thinking that is familiar to a certain social group. (e.g.: Painting, Making music, Choreography, etc.)
- Transformation: To make transformations that enable the generation of previously impossible ideas.(e.g.: Digital cash, Virtual Currency, etc.)
I have attempted to make some pages and models using machine learning in group assignments and my individual project. Based on these projects, I will consider the difference of creativity between AI and human while referring to the definition of creativity and the three ways to be creative.
2. Can AI Truly Be Creative?
2.1. Mask Classifiers (Group Assignment)
First of all, our team invented “Mask Classifiers” that calls on those who do not wear a mask to wear it. We took short videos of each type of mask (Pitta, Cloth, Abenomask, Surgical, N95) to use as training data for our teachable machine AI. Teachable machine then uses the training data to detect what is being shown through the webcam and whether it matches the images in the training data.
Output of this system was just judging whether people wear a mask or not and which masks people wear. In other words, AI was just an optimization tool in this case. Through my experience of inventing this system, I learned that AI can hardly be creative if we make AI do a simple task like a classifier system.
2.2. thisbanksydoesntexist (Group Assignment)
Secondly, our team attempted to make “thisbanksydoesntexist”. This is a GAN model which enables to create a new style by combining Banksy and Impressionism.
We extracted 7,448 images from Google Images and searched the keywords “graffitti” “street art” “banksy” and “impressionism”. We removed images that did not fit the criteria from the dataset and put everything into one directory. We used Runway ML to train it over 3,000 steps. We used one for the default Style Transfer GANs available in Runway.
This model achieved to combine familiar artworks and to create new and surprising ideas. And also output could inspire us in the artistic aspect and perhaps form new culture.
2.3. thiskeithharingmayexist (Individual Project)
Thirdly, I individually tried to make a GAN model using Keith Haring’s works by applying the previous project, because I guessed that his works are so simple and characteristic that AI can easily recognize their feature.
I gathered 613 images from Google Images and Pinterest. I searched the keywords “Keith Haring” and imported images. Then I removed images that did not fit the criteria from the dataset and put everything into one directory. I used Runway ML to train it 2,000 steps. I used the default GAN model available in Runway which can generate new illustrations based on the dataset.
As a result, I could see the thick lines and vivid colors like Keith’s works from output, but shapes of human body that can be almost always seen in his work could not be seen. In order to make AI to detect the shapes clearly, I should have strictly picked up the images for dataset.
Although this output was so far from ideal in this way, I thought that it was of great value because it can stir our imagination.
3. Conclusion
Through the practices of this class, I learned that AI can create ideas that are new, surprising and inspirational especially when we make AI do complicated tasks. Nevertheless, the products created by AI may not always be valuable even if AI creates something novel. To give an example, this is an artwork painted by AI. A lot of eyes of an animal can be seen in a human’s body. It will probably makes us feel new but scary or not beautiful. Is AI really valuable and creative that could also create something unpleasant or strange to us?
As opposed to humans, AI cannot instantly understand what humans feel and our cultural contexts. It will be a future problem when we make AI be creative. This problem will cause bias problems and other ethical issues. In order to achieve singularity, AI should be trained huge amount data and understand our emotion and cultural contexts accurately.
We humans cannot make big impacts like AI in creating, but we can combine different ideas and come up with new ones while understanding our feelings and having knowledge of culture and ethics. AI and humans have to make up for each other’s weak points and live together.
