ControlShift + AI

We're excited about they ways that AI features can help our organizations respond more quickly to emerging issues and be more impactful in their work. When it comes to implementing AI into ControlShift, our overarching considerations are:

• ensuring that we're protecting users' personal information. We've chosen our tech to ensure platform data doesn't leave our infrastructure and isn't used to improve third-party models. We're using AWS Bedrock.
• implementing only the features that will actually be useful and impactful. We've seen plenty of 'we've added AI so that we can say we have AI' features in other tools, and we want to be more judicious in how we implement AI within ControlShift.
• protecting the magic of distributed organizing and authentic voices. Our power comes from regular people taking action in their communities and bringing their authentic selves into the campaign. We  want to keep that authenticity and not build any features that undermine the real people behind our campaigns.

The first AI feature we're releasing is available for admins only: AI assisted moderation. Moderation is of course a key task for admins in ControlShift and we're hoping that assisted moderation may make it easier and faster to complete those tasks. The current version is only providing a suggestion of how we think the petition should be moderated based on the text that's been entered.

Moderators can then provide feedback about the recommendation, which we'll use to help refine the model. Org admin moderators have the final say over the moderation status for newly-created content, the AI serves solely to assist and provide recommendations. For organizations operating in languages other than English, we think this will be a more accurate way of flagging content compared to our current, word-list-based profanity filter.

Future versions of this feature may include automatically limiting the visibility of content deemed questionable by AI while awaiting final moderation by a human. This would help keep organizations safer from spam and/or malicious petitions, while still ensuring that final moderation is performed by a staff member.