Feedback Loop is a project that seeks to control the Instagram Explore Page algorithm. This process explores uploading the machine-generated GAN images from the ‘Suggested Posts’ series to a separate Instagram account and using my personal Instagram account to engage with these images. This process forces the Explore Page algorithm to make assumptions about these images, therefore suggesting new results.
The Instagram Explore page consists of photos, videos, stories and products tailored for each user based on their Instagram engagement. I wanted to see how interacting with machine-generated images would affect the content that appeared on my Explore Page over time. Would this completely change the type of content tailored to me? After I used my account to interact with numerous posts containing GAN images (from the Suggested Posts series), this forced the algorithm to show more content comprising vibrant, scenic landscapes. The GAN images from my ‘Suggested Posts’ series consisted of muted pastel tones and the occasional splash of vibrant or rich colours. I then collected a dataset of posts from my newly updated Explore Page – these images were fed back through my machine learning model to see what the machine would produce. This data set consisted of 342 images from my newly updated Explore Page. After completing this process once, the GAN images began to resemble real-life objects, scenes or places(such as Utah) from a glance. After the second process, the GAN images started to look more abstract. The more I uploaded these GAN images to my Instagram account and used my account to interact with these images, the more natural-looking the content on my Explore Page became.