Breakup Simulator

class project @ MIT Media Lab

Keywords: Deepfakes, Machine Learning, Difficult Conversations, Ghosting

To create a tool that eliminates fears of the unknown and eases anxieties about the difficult conversation of a breakup.

Erica Luzzi, Noah Deutsch, Maggie Chen

︎ Content/ Conversation Design
︎ Illustration

The rise of digital communication platforms has made it easier than ever for people to connect with others and meet new people in novel ways. However, as a result of this increased access, often digital interactions lack the intimacy and authenticity of in-person interactions. Recently, we have seen an epidemic of ghosting: the phenomena where a person cuts off all communication with someone with zero warning beforehand.

Our solution is the Breakup Simulator, a web-based platform that uses deepfakes to create an interactive breakup conversation between the user and their partner. The goal of the tool is to eliminate fear of the unknown and ease anxieties about the difficult conversation of a breakup.


Content Development

After researching countless breakup stories on Reddit, Tumblr, magazine articles, and random blogs, we imagined 5 scenarios like “I love you, but you hurt me”, “Different paths”, and “Long distance is tough”. They are representative of specific reasons to why people would want to break up but are still generic enough to be relatable.  

Each narrative is broken down into scripted lines, which are assigned to the user and their simulated partner, and they are acted out in pre-filmed driving videos with facial expressions and corresponding audio of the lines. 


On Ghosting and Deepfakes

In our team’s attempt to explore positive uses of deepfake technologies, we created this breakup simulator, with the hope that practicing breakup conversations in a simulated face-to-face context, powered by AI, can help people overcome the fear and anxiety involved in anticipating a difficult conversation like breaking up with someone.

Ultimately, this tool is meant to be a training wheel for the actual conversation-to-be-had in real life. Instead of simply staring at AI versions of themselves and their partners practicing this imagined breakup on screen (which sounds like a story out of a Black Mirror episode), users are encouraged to have the actual face-to-face conversation offline after going through this practice; otherwise, it defies the purpose of this project.

While whether this will actually happen can be up for debate, but in this experiment, we hope to reveal new ways of applying ML technologies like deepfakes to help bridge the gap between people brought forth by technologies.