AI, what is means for the society and creativity

I always thought AI were a ‘tool’ to automate tasks, with input and outputs that discovers relationships. Having a prior knowledge in machine learning and deep learning, I believed these models was invented for the sole purpose of scientific research, with no creativity at all. However, enrolling into this class have shaped my perspectives on AI and creativity in general. I would like to share my knowledge and perspectives gained from this class, and my view on AI and creativity, society.

Teachable Machine

One of the projects was using “Teachable machines” to create an AI interactive system. It was surprising to see how easily a neural network model could be trained online, with no prior knowledge about neural networks. I realized how deep learning technologies are becoming easily accessible in this day, allowing any enthusiast to experiment with it, at no cost.

Image of Teachable Machine’s interface

For the project, I have decided to create a ‘hands free emoji feedback system’. The motivation for this project was due to transition of physical class to online class for my university. Ever since it became online, I felt the feedback and communication between the students and teachers decreased significantly. Therefore, by creating an hands free emoji feedback system, people can understand what others are thinking or experiencing. With Zoom requiring the user to manually click the emojis to provide feedback, I thought automatically detecting the face of the user and outputting it will make it easier and increase the usage of emoji feedback.

Images of the Zoom Emoji detection system

Although the proposed system were sometimes inaccurate and unsteady(due to lack of training data), I realized the potential of how AI can be used to improve our lives/society. Since the system was technically ‘hands free’(due to audio and image classification), AI can used to increase the accessibility of potential systems. AI’s ability in learning features of inputs data can be used effectively to automate systems, that can create a better or seamless experience for the users. AI also has the potential to create opportunities to connect people, society together. In the future, we should automate repetitive and redundant tasks to AI to free up time for creative and innovative tasks. As innovation and creativity is the key of creating and interesting and fruitful world.

The accuracy and effectiveness of AI depends on the training datasets. Thus, if the datasets does not generalize well or have enough diversity, the model will not work properly in some occasion. AI model will not work well if the model gets an unfamiliar data or unknown data. This potentially increases the likelihood of AI system to have discrimination and bias. For example, the model for my emoji feedback system used only Japanese male faces, therefore it would not work well for other nationality and gender, which may lead to segregation(if the system was published). When an AI model does not have enough diversity in the data, there is the potential risk in the future of discrimination and bias. Therefore, when creating an AI system, it is important to have large diverse datasets to prevent such risks from happening.

RunawayML (GAN)

Another project was using RunAwayML, to create images using Generative Adversarial Network(GAN). Using GAN model have made me realize the potential of AI and creativity. GAN model has the ability to capture the features of the input images and create similar looking images. For my project, I used images of Sakura Trees and clouds/sky pictures to create ‘Japanese’ looking clouds.

Sample images used for GAN
Image produced from GAN: Sakura Trees

However, the model gave me unexpected and surprising looking images. One of the images had a cloud looking pictures, but had a cellar looking sakura tree instead. There were also other images that did not capture the main features of the images I wanted. I realized GAN model can extract features that are not known or visible to the human eyes and give us surprising, creative outputs. The GAN model was indeed creative in creating the images.

The ability of the GAN model to create creative images may risk the decline in the overall creativity of humans, where we might a see a future where we depend on AI for creativity. I believe we should strive to be innovative and creative, using AI as one of the options to gain creative insights. AI should also be recognized as a ‘challenger’ or ‘rival’ to us, instead of thinking them as a tool. By doing so, we can continue to strive to be creative to compete with them and also get inspiration and ideas from them. Accepting AI into our society is one of the first step in getting creative insight.

Using AI for Music Generation(Recurrent Neural Network)

AI is not only capable of generating images, but music as well. There are recurrent neural network(RNN) that is network that uses sequential data, such as sentences and music sheets. This network is named ‘recurrent’ because of its structure, where each layer has a ‘hidden state’ that stores the previous output(calculations) and uses as the input for the next layer. This structure allows for the network to remember the past information. These type of models allows to generate music, in which we can say the AI is the composer. Although the music generated by these AI are not perfect, but it opens up for new creativity. AI can generate brand new sequence/pattern of music that has never been heard before. For example, one of the famous classical composer Franz Schuber had a symphony called the “unfinished symphony”(also known as Symphony №8).

Music sheet of Franz Schuber “unfinished symphony” (Wikipedia) [1]

The symphony is exactly what the name implies, the symphony is unfinished. A network was trained to complete the unfinished symphony[2], which supports AI indeed has creativity. I believe not all music enthusiast are accepting or recognizing AI as an artists yet. There are still a barrier between AI and music, that prevents it from coexisting together. For our future, we should begin to accept these new technologies and style, so that we can continue on being creative, and inspiring others are the same time. I hope for the future where AI generated music and AI artists may become the new norm.

Conclusion

I like to think AI as a double-edged sword. It has the potential of making the world more creative and innovative, while at the same time having risks of discrimination and uninspiring environment. For the future, it is important to accept and recognize AI as a both a ‘rival’ and a ‘creative tool’. By recognizing them as a rival, we will continue to strive to be creative and innovative, while getting inspirations and ideas. We should also rely AI to handle repetitive tasks and automate the process, so we can focus on other creative tasks. I am excited to see a future where AI will coexist with humans to bring a colorful life.

Reference

[1] “Symphony №8 (Schubert).” Wikipedia, Wikimedia Foundation, 4 Aug. 2021, en.wikipedia.org/wiki/Symphony_No._8_(Schubert).

[2] Davis, Elizabeth, and Lizzie Davis. “Schubert’s ‘Unfinished’ Symphony Completed by Artificial Intelligence.” Classic FM, Classic FM, 6 Feb. 2019, www.classicfm.com/composers/schubert/unfinished-symphony-completed-by-ai/.

--

--