Triple Adversarial AI
Collective Artificial Wisdom
GANs have been a breakthrough in helping create realistic images of people. New technologies have come around for generation. However, what if there was another level of adversary that could pick the best of the best of AI generated content?
We are seeing now the capability of technologies like Google’s Imagen, DALL-E2, and others create such amazing imagery that it rivals many artists you can find on Fiverr. However, how do we scale the generation of usable images?
We need another adversary against the generative services that is going to pick the best among a batch and then run the generation service again. This adversarial service combined with an automated workflow will mean that whole plots, animations, comics, books, etc., can be created within a system. The service can create whole content without supervision.
At least at the start, bad content will still make it through. However, over time, the system will get better and picking what we’ll find the most appealing.
By combining multiple AI adversaries together, it’s possible different types of AI become a voter base. This disparate group of AI’s will become a voter base and there will be a collective artificial wisdom. It’s likely that many AI’s will be better at choosing something good than a single AI with potential flaws.
