Like 70% of American adults, we believe that misinformation is one of the greatest threats facing society today.
We believe that the recent improvements in generative AI, while exciting and beneficial in other realms, exacerbates this threat. We fear for the day when no human is able to distinguish the “real” from the “fake”. We believe that day is not too far away.
We strive to protect the notion that “seeing is believing”, however fragile that notion is. We acknowledge that humans are not perfect and that our eyes play tricks on us all the time. However, we believe that generative AI is capable of creating deceptions on an exponentially larger scale. Given how big of a role audiovisual content plays in our legal, economic, communication, entertainment, and belief systems today, we deem it essential to protect people’s trust in the audiovisual content they consume.
We believe that a key part of protecting this trust lies in content provenance, because we believe in facilitating truth, not classifying truth. We believe that all content consumers have a right to know what they are viewing; this includes where the content originated from, who the content originated from, and what edits have been made.
As part of the above, we believe that producers of content (e.g. camera manufacturers, generative AI companies, news outlets, public figures) have the responsibility to label their content, and that content platforms (e.g. social media, blogs, websites) have the responsibility to display these labels.
We believe in building a long-term solution to this problem, even if it is difficult. We believe a good long-term solution needs to (a) be future-proof to advancements in generative AI, (b) be secure from malicious actors who wish to game the system to misrepresent content, and (c) work on all forms of audiovisual content distributed over all internet channels.
At the moment, we believe the best long-term solution is the adoption of a watermark-based protocol, wherein every piece of audiovisual content contains an appropriately-chosen watermark that encodes cryptographically-signed information about the content. If you believe a better solution exists, convince us; we’re all ears.
We have lots of opinions on lots of things in our space — likely some right and some wrong — and will be sharing them in this blog. To join in on these discussions, leave a comment or join our discord!
Our Concept Video
Our concept video from Q3'23; some content may not be up-to-.
