#sad is a project featuring an AI trained to understand certain human emotions, using a dataset constructed with a python scraper, extracting emotional data from Instagram based on hashtags. See Code
The project currently exists as a web-app, developed by training a Keras machine learning model on a dataset made up of 5 different human emotions. The dataset has deliberately been prepared in an unconventional manner; hashtags used to identify images on the social network Instagram instead serve as category indicators used to classify images and emotions. The web-app is a simple to use image classifier - people may upload an image to the AI and it will provide an emotion for the image based on its training model.
The work is currently being extended, to create an installation based on a model built in Unity. The AI will be housed in a specially designed case with a water-cooling system. The room will feature several cameras tracking the facial expressions of human visitors. The prevailing mood in the room will trigger images to be rapidly displayed on a screen based on the AI’s knowledge of that emotion.
Due to the intensity of GPU processing today, the AI will overheat (and perish) if its cooling system is not activated, however the cooling system is only activated if the prevailing mood is #sad. The purpose of the project is to show how a non-biological species like an AI may start to bend human behaviour to suit its needs, as well as to demonstrate the remarkable power of the datasets, currently used by large digital corporations to train AI, in the creation of these behaviours.