Implications of AI on Creativity
British psychologist Graham Wallas listed 4 stages of creativity in his book “The Art of Thought” [1]. Below I will list all 4 stages and see if there is a modern AI that can facilitate the work in that stage.
Stage 1: Preparation.
The first stage is about accumulating information and resources about the problem at hand. It might be internal (i.e. brainstorming the ideas in your head) or external (i.e. reading books, consulting with others, surfing the internet, etc.).
AIs that may help at this stage include currently the most powerful language model, OpenAI’s GPT-3. It can convincingly complete a started text with different styles. It can generate blog posts, stories, songs, conversations, even computer code [2]. The model was trained on Wikipedia, a corpus of books, and other resources. Thus, an artist might start typing his/her initial idea and let GPT-3 generate the rest of the idea. This way, it might come up with an idea the artist has never thought of before by using the vast resources it was trained on. Even though it has flaws, GPT-3 might become a basis for AI which would help an artist develop his/her ideas.
To research, we usually use Search Engines. Nowadays, Search Engines are also powered by Artificial Intelligence. For example, Google uses Natural Language Processing (NLP) to rank the search results by relevance, Deep Neural Nets to fix spelling errors, selects certain passages as answers to the queries by using AI [3]. Thus, in this case, AI makes the Stage of Preparation faster and easier.
Stage 2: Incubation.
At this stage, after putting immense efforts into researching the project, the artist stops working on the project and instead works on something else. In the meantime, the artist’s brain gets involved in passive thinking trying to connect different resources together and creating the idea for the project. This stage is similar to the AI model’s training stage. However, I cannot think of any ways an AI can help with it to a human.
Stage 3: Illumination.
This is the stage when the idea comes suddenly to your mind after a period of incubation. I believe this “aha” moment is similar to when an AI model learns and successfully generates good data instances. However, I don’t believe AI can currently help an artist achieve this stage faster.
Stage 4: Verification.
At the final stage, the artist uses the idea he/she generated to create the final product. Verification is all about testing different approaches to realizing the idea. In my opinion, this is the stage where AI is and will be most helpful to an artist. AI can speed up the time for testing different approaches. AI, for example, can nowadays apply different styles to photos and paintings. Below, for example, I am applying the style of Edward Munich’s painting on my own without putting any effort in drawing it by using a pre-trained model “Neural Style”.
AI can even help you quickly draw your ideas. A new NVIDIA Canvas App, for example, transforms your simple scribbles into a realistic landscape picture. The objects in the final portraits are affected by each other. So for example, if there is a tree near the water, the tree will be reflected in the water.
DeepFaceDrawing is similar to the previous AI app. Instead of landscapes, however, it draws people’s faces. Even drawings of amateur painters resulted in very good faces. It is incredible how the model could learn to emulate even the most intricate details of a human face such as beards and cheeks.
Conclusion
To sum up, 2 of the 4 stages, in my opinion, are and will be well facilitated by AI apps. However, they will still serve as assistants and won’t replace the humans. The other 2 stages can be achieved by only humans.
References
- Gregoire, C. (2019, October 18). Understanding the four stages of the creative process. WeWork Ideas. https://www.wework.com/ideas/professional-development/creativity-culture/understanding-the-four-stages-of-the-creative-process
- Heaven, W. D. (2020, July 20). OpenAI’s new language generator GPT-3 is shockingly good — and completely mindless. MIT Technology Review. https://www.technologyreview.com/2020/07/20/1005454/openai-machine-learning-language-generator-gpt-3-nlp/
- Raghavan, P. (2020, October 15). How AI is powering a more helpful Google. Google The Keyword. https://blog.google/products/search/search-on/
