AI and Creativity: The Controversial Mix
Exploring the idea of AI and Creativity using the GAN model.
As the digital world progresses rapidly, AI, merely thought of as a science-fictional idea of the 1900s, has become significantly more realistic in the modern era. Though not as imagined in the 1900s, AI has broad implementations, from simple tasks such as adjusting the temperature in the room to complex tasks such as gaining consciousness. Even though there is a general implementation of AI, one of the most controversial discussions on using AI is within the aspect of creativity.
To explore the idea of AI and creativity, the GAN(Generative Adversarial Networks) models were produced using a web-hosted machine learning platform, RunwayML. The process of generating these GAN models will be used to determine if AI can be creative or influence creativity.
What is creativity?
Before we dive into the process of generating the models for observation, we must first reach a consensus on what creativity is. Since the definition of creativity differs from person to person, Boden's (1998) definition of creativity will be used to come to a consensus on what creativity is. According to Boden, there are three main types of creativity:
- Transformational Creativity
- Exploratory Creativity
- Combinational Creativity
If AI were considered "creative," out of the three types of creativity presented, it would most likely be categorized as combinational creativity as AI requires inputs to generate "art."
Generating GAN models
RunwayML has made it simple to generate GAN models by only requiring a dataset. For this case, 800+ images from the book Dune by Frank Herbert and its 2021 movie adaptation were web scraped and used for the dataset for the image training.
Dune was chosen because of its diverse and strange imagery(which I like) in its vast complex world. Within Dune, a range of images can affect the training of the GAN models, from the infamous sandworms to the deformed humanoid figure. However, there is a need for similarity to train the models. This similarity is the background of these unique images, the dunes.
How does the training work?
The training is the step in which the AI will make a hypothesis of the new data using the patterns of its old data by setting its parameters to make predictions about the model or completely generating a new model mimicking the original trained model. To train the 800+ Dune images, there is an option for RunwayML to choose the number of "steps" that the model takes. These "steps" indicate the number of times the images go through the training process. To produce the GAN models for this case, 3,000 steps were taken, which took approximately four hours to generate.
Is AI creative within the parameters set?
Observing the process of the GAN model training and the use of Boden's definition of creativity, it can be implied that the process that the AI is indeed creative. Taking familiar ideas(images inputted) and creating something new(the models) is the very definition of combinational creativity. Not only combinational creativity, but if AI continues to produce models, the odds of creating an artwork that is transformative creativity significantly increase.
Is AI creative without the parameters?
If the definition of creativity was not one of Boden's definitions, is AI creative? Many would argue that it is not creative because AI uses a set pattern and does not innovate outside those parameters. However, they may argue that AI has the potential to make an impact on creativity.
Final Thoughts
The idea of if AI is creative may be the discussion of the decades. However, the question of whether AI can impact creativity can be answered. If AI were thought of not as a sentient being but rather as a device to enhance our innovations, it would significantly accelerate the development of humankind. In the future, say 20 years from now, AI will most likely be used as an idea or inspiration generator. AI can be used to replace the brainstorming process and generate millions of ideas to inspire humans to construct something that can result in something transformative. However, one aspect that should be taken is data biases. In the case of generating a model using GAN, the data bias of dunes made it easier to generate similar thematic images. Though it is suitable for thematic art, AI may result in something more harmful than good if AI were to be used for something other than art. Another aspect that many worry about is AI taking over current jobs; more jobs can be created by enhancing humanity's creativity. Thus "Is AI creative?" maybe not, but "Can AI impact creativity?" definitely.
