Wonder how this could be used for communication and captchas? Could be a really expandable approach to determining "probability of bot" for a lot of internal use cases in companies that might involve requesting access to specific resources only meant for humans.