An AI Ouroboros
Author: COMAP
Background
The problem of Artificial Intelligence (AI) model collapse has been documented and studied under ideal situations[1]. AI generated content is becoming more prevalent, and it is more likely that AI content engines will use previously generate content as the input used to train newer models. The result is a potential degradation of the quality of the new content that will be generated by new AI engines.
As AI models become more prevalent, different people will use different engines. People also tend to take the output of an AI model and edit and adapt the output to suit their own tastes. Additionally, people will continue to generate new content on their own. All of these different source materials will be publicly available and will likely be used to train new AI models. The question you are asked to examine is what happens when multiple AI models interact in the presence of new content that is generated by people as well as content generated by other AI models? How will the new content generated be impacted? What long term patterns will emerge?
Some additional questions to address include the following: Will the quality of AI model results dramatically decline, slowly decline, or not be impacted? What is the role of multiple AI models interacting? Will the different outputs generated by different models mitigate the impact of AI model collapse?
Problem Download
You must have a Mathmodels Membership to download Student Papers and Commentary.
If you're already a member, login here.