Shadow of AI
For individuals, confronting the massive network of interests underlying AI in personal privacy, such as cookies on websites, can be particularly hard. I propose the development of a projection-based interactive experience that replicates the user's data shadows with real-life shadows and presents them in a novel and approachable way. Audience will be inspired to investigate the story behind their unintentionally submitted data online by interaction with the virtual shadow. By observing the fake shadows, they could compare their current shadow with those fake shadows. With the interactive installation, new ideas should be generated by those who attend this project: the future of the Internet will be taken over by the false shadow of humanity.
Speculative Design
The design of the future, especially in relation to technologies that may have a significant impact on it, is expressed through a design approach which is highly interactive, experiential, and inspiring, leading the viewer to understand the many possibilities of the future and form their own unique opinions. When everyone is involved in thinking, discussing, and generating subjective ideas, society is more likely to avoid the inevitable fetters of a future outlined by the profit motive of business, and to develop in a desirable direction.
Data
The data explored in this project were obtained from my daily image of shadows. As a designer, I was strongly guided by keeping the dataset interesting but, at the same time, maintaining its integrity and scientific validity. From 19 to 26 July 2021, I recorded the movement of the shadow with a go pro attached to my backpack only on sunny days ensuring that the shadows will be sharp and clear.
Design Process
To train a GAN to create fake shadow pictures, I used go-pro to capture my shadow at different day times. I then hand-selected 92 photographs from over 1000 frames of images and used align_images to align and crop most of them around the shadow's centre. The rest of the images were manually cropped in Adobe Photoshop. Training was done using a StyleGAN2 on Google Colaboratory. Paperspace is another popular choice, however the recommended Colab Pro is currently only accessible in the United States. The initial model was based on the published StyleGAN2 model using transfer learning. Take note that, while the finished photographs retain the basic aesthetic of the training set, they are unique in that they do not contain the shadows of actual people from the training set.
Implementation
Throughout the process, one of the primary techniques I utilised was to invite participants to co-create speculative fiction by fusing and interacting with the false shadows generated using StyleGAN. The story revolves around the real and fake shadows of the audience and illustrates the challenges of inadvertently leaving a data shadow in the online world. The aim, along with speculative design, was to offer a longer-term and more critical approach to technology development.
I rented a windowless room in the ECA to achieve total darkness so that the projected reality would be clearer. After the projector was turned on, I tried to adjust the angle, direction and distance so that the audience is always with the best shape and size of shadow in front of the projection. Afterwards, the first audience came to experience the project. The background of the audience was the students at the University of Edinburgh, who are very fond of surfing on the internet and are arguably the most frequent internet users today. They entered the room one by one after listening to my brief introduction.
I recorded the audience's experience in the whole project, starting from entering the door, to stopping to look at the fake shadow, then reaching out and interact with the shadow. Finally, leaving for the interview. One finding is that students who are not from informatics background feel that the project is rather obscure, and they don't know what to do with their own shadow in relation to the virtual shadow. In addition, StyleGAN was also a difficult concept for them to understand without prior explanation. On the contrary, students from an informatics background will get more insight from it.