MODEL-BASED ATTRIBUTION FOR CONTENT GENERATED BY AN ARTIFICIAL INTELLIGENCE (AI)

Smart overview of the Invention

A server determines when an artificial intelligence (AI) has completed training to become a trained AI. It then identifies an attribution vector created during this training process. Upon receiving input, the trained AI generates output, and the server assesses individual content creators' contributions based on the attribution vector. Creators who meet a certain contribution threshold receive attribution, while those below this threshold do not.

Importance of Content Attribution

Generative AI can create derivative content that resembles original works used during its training phase. However, current systems lack methods to gauge how much of the original content influenced the generated output. This gap hinders the ability to provide proper attribution and compensation to original content creators whose works contributed to the training of the AI.

Mechanism of Attribution Determination

The server's process involves selecting a content creator and associated content items for training. It measures the proximity between the trained AI and its pre-training state to determine the creator's influence. This influence is incorporated into the attribution vector, which ultimately informs compensation decisions for creators based on their contributions during the training phase.

Provenance in Generative AI

Provenance refers to tracking and authenticating digital content created by generative AI, linking it back to its original creators. This includes storing attribution information on a blockchain for decentralized tracking. By establishing clear provenance, creators can receive recognition and potential financial rewards for their contributions to derivative works generated by AI.

Applications Across Generative Models

The described systems and techniques apply to various generative AI models, such as diffusion models and GANs. For instance, a latent diffusion model utilizes an auto-encoder for efficient training, allowing it to generate data similar to its training inputs. This adaptability ensures that proper attribution can be maintained across different types of generative AI technologies, enhancing fairness and transparency in content creation.