Discriminator — Grant Report #1

Project Update

Discriminator is well into production, with ongoing work in research, animation, music and web development.

The initial phase of the project was in the creation of a set of prototypes to explore the possibilities of using facial recognition within a web documentary. This prototype explored the possibilities of the "wav2lip" project, which modifies images and video files by synchronizing their mouth movements based on a submitted audio file. The goal was to present a "deepfaked" video by seamlessly integrating it with static video files. The results can be seen here:
https://three-videos.jaeperris.com/
(Google Chrome recommended)

In parallel, an extensive research phase was undertaken to understand how the author's flickr photos were used in facial recognition databases. In partnership with researcher Adam Harvey, we explored how the Yahoo Flickr Creative Commons 100 million database (YFCC100M) became the basis of several large facial recognition databases including MegaFace. This research modified and focused the direction of the project - rather than exploring disinformation and synthetic media, the project now more specifically focuses on the problematic use of CC licensed images in facial recognition databases. This is explored more full on Adam Harvey's website, https://exposing.ai/

Building on these foundations, we are now moving forward on animation with Darren Pasemko, music with composer David Drury, AI and web development with Jae Perris, and design and creative technology with Hang Do Thi Duc

Progress on objectives

Goal 1 - "incite curiosity and educate users to the social concerns surrounding facial recognition, AI and disinformation"

This is our central goal, and the documentary is designed to educate while being entertaining. As research has progressed, however, it became clear that the primary dilemma was the unanticipated uses of creative commons photos. Like many others, I (Brett Gaylor) uploaded many photos to Flickr under a Creative Commons license only to discover many years later that they had been included in large facial recognition databases such as Megaface as a result of licensing choices. The goal of the documentary, therefore, is to widen understanding and conversation about the unintended consequences of openly sharing data.

Goal 2 - create a popular and professionally produced documentary that other creators can study and emulate - this is going well, and we are releasing all of our code on Github

Goal 3 - experiment with new ways that we could finance our work and establish more direct relationships with our audiences

As the focus of the film has tightened, I have become more aware of advocacy groups in this space. For this reason, I have decided to focus the web monetization component of the project on raising funds for groups pushing back against surveillance systems. I am in talks with the Surveillance Technology Oversight Project.

Key activities

Our key activities are research, web development, composing and animation.

You can see progress in all of these areas in this prototype
https://three-videos.jaeperris.com/
(Google Chrome recommended)

Communications and marketing

Our project was recently featured in The New York Times
https://www.nytimes.com/2021/01/31/technology/facial-recognition-photo-tool.html

What’s next?

We are gunning for the finish line! We aim to have music composed by the end of february, animation completed in early March, and web development in April.

What community support would benefit your project?

What we will likely need the most is word of mouth promotion.