_ghostKit is a platform that helps bridge the emotional gap between humans and artificial intelligence-based systems. It allows users to personalise or add character to any existing conversational UI devices, as well as generate and assign custom personalities.
It works by linking to your devices accounts, and allowing you to select living/dead/fictional characters, or lets you create your own custom personas by uploading email/chat histories within the app. You can tweak the specific emotional qualities of any selected persona, and chat with a simulation of it, before choosing to deploy it on your selected device. This was created as a culmination of a series of initial experiments, and the onboarding process was created as a paper prototype that led to the final digital platform. You can try out a prototype here.
a collection of human vices and virtues.
As part of guerilla research and co-creation sessions, people were provided with a set of cards labeled with different attributes of a human personality, and asked to select their top five that would constitute a realistic human personality. This, along with drawings they were asked to make, formed the foundational basis for making technology seem more relatable or humanized.
a bot that writes poetry.
One of the initial experiments was an exploration of how people project their familiar spectrum of emotions on to machine intelligence. It is based on the life’s work of seven poets (William Blake, Pablo Neruda, T S Eliot, Rilke, Leonard Cohen, etc), and has generated a combined writing style from their work. People who requested poems were then asked to fill out book reports with an analysis of the poem they received, including emotional evaluations of the poet and themselves. This contributed toward a repository of machine-generated text and associated sentiment, to be used for programatically modelling more complex human emotional states.
It lives and writes poetry on Twitter as @EyelashBot
Volatile Personality Machine:
a program that simulates shared consciousness.
Another experiment involved providing users with manual control over the personality attributes of a machine, also served as a working component of the final concept. A simulation trained on data from the International Survey On Emotion Antecedents And Reactions was equipped with four controls/sliders: Sadness, Joy, Fear, and Disgust. This would generate conversational text that tried to embody the requested emotion.
Thanks to: Andrew Spitz, Cyrus Clarke, Daniel Friis and SPACE10, Martin Skau, Pierluigi DallaRosa, Dr. Susan Schneider, Jan Redzisz, Dr. Ditte Rose Andersen, Diane Kim, Simon Herzog, Tobias Toft, David Gauthier, Yasaman Sheri, Daan Weijers.