2021: It has been one year since many of us had to leave our offices, schools and universities to work, study and communicate solely from home. Our lives have shifted rapidly into the digital world. Simultaneously, we are getting more connected than ever, navigating websites, apps and software daily. What we mostly don’t notice is the data we leave behind.
How many times has an app asked for your location, your contacts or browsing history, when there seemed to be no need for this information? It has become normal for us to use free online services. However, the hidden cost of such services is our data.
We created Lost in Truncation as an object that makes this exchange visible. Once you enter the installation, you are promised a futuristic, AI-generated passport photo of yourself. All you need to do is to take your picture and answer a couple of questions. You might believe that your answers will influence the outcome of your AI-generated picture, but that’s not entirely true. All our AI actually needs, is your photo. With that, it will create a unique passport photo.
The answers you give to our set of questions are forwarded to a hidden second AI. This AI will collect your answers and generates an analysis of you, full of assumptions about your personality. This analysis is printed out into a glass container: Fully visible to everybody, but still locked away. You will leave the installation with a unique and futuristic passport photo, but some questions may still linger: What will happen to your data after you left? Was it worth it, giving your personal information away for a simple picture?
Truncation is the process of truncating or shortening something by removing a part of it. It is often used in statistics and studies but is also a value that is used when training neural networks. By excluding neurons in the hidden layer of multilayer neural network you can optimize your model and get better results. The name was perfect for our project because it did not only match the method with which we trained our model – It also stands as a metaphor for the shortening of the user's personality. As of right now, an AI analyzing a person based on just a few questions will not be able to fully grasp the person's whole character. But who knows how this technology could possibly evolve in the future?
For the text output we used GPT-2, an open-source artificial intelligence which is able to generate text on a level that can be compared to humans. For the image output we trained our own model based on StyleGAN. We created a dataset containing diverse sets of images: plastic objects and products, futuristic materials and textures and contemporary 3D-renderings. So we trained StyleGAN, which is a model based on human faces, to become more abstract and futuristic. The outcome is a symbiosis of the two.
Both are unique because it’s based on the input you gave like taking a photo or answering the questions