Model-based attribution for content generated by an artificial intelligence (AI)
In some aspects, a server trains an artificial intelligence (AI) to create a trained AI. The training includes: selecting a content creator from multiple content creators to create a selected content creator, selecting a plurality of content items associated with the content creator to create selected content items, training the AI using the selected content items, determining a creator influence of the selected content creator on the trained AI based on an aggregate influence of the selected content items on the AI during the training, and including the creator influence in a static attribution vector. After the training is completed, the trained AI receives an input and generates an output. The server creates an attribution determination based at least in part on the static attribution vector and initiates providing compensation to one or more of the multiple content creators based at least in part on the attribution determination.
1. A method comprising:
training, by one or more processors, an artificial intelligence to create a trained artificial intelligence to generate a particular type of derivative content, the training comprising:
selecting a content creator from multiple content creators to create a selected content creator;
selecting a plurality of content items associated with the content creator to create selected content items;
training the artificial intelligence using the selected content items;
determining a content item influence of individual content items of the plurality of content items on a plurality of training techniques used to train the artificial intelligence based at least in part on determining a difference between:
a first set of parameters associated with the artificial intelligence, and
a second set of parameters associated with the trained artificial intelligence, wherein the second set of parameters comprises a plurality of weights, a plurality of biases, or any combination thereof;
aggregating the content item influence of individual content items of the plurality of content items on the plurality of training techniques used to train the artificial intelligence;
determining a creator influence of the selected content creator on the trained artificial intelligence based at least in part on the aggregating; and
including the creator influence in a static attribution vector;
after determining that the training of the artificial intelligence has been completed, determining, by the one or more processors, that the trained artificial intelligence has received an input;
generating, by the artificial intelligence, an output based on the input;
creating, by the one or more processors, an attribution determination based at least in part on the static attribution vector; and
initiating, by the one or more processors, providing compensation to one or more of the multiple content creators based at least in part on the attribution determination.
2. The method of claim 1 , further comprising:
determining a transformativeness factor of the trained artificial intelligence relative to the artificial intelligence; and
determining the static attribution vector based at least in part on the transformative factor.
3. The method of claim 1 , wherein the training further comprises:
determining a content item influence of individual content items of the plurality of content items associated with the content creator using a converging function.
4. The method of claim 1 , the method further comprising:
while generating the output:
determining a plurality of activations of different weights in the trained artificial intelligence;
aggregating, for individual content creators of the multiple content creators, the plurality of activations of different weights in the trained artificial intelligence; and
determining a dynamic attribution vector based at least in part on the aggregating.
5. The method of claim 4 , the method further comprising:
further creating the attribution determination based at least in part on
the dynamic attribution vector.
6. The method of claim 1 , further comprising:
determining the content item influence of individual content items of the plurality of content items based at least in part on the plurality of training techniques used to train the artificial intelligence.
7. The method of claim 1 , wherein the output comprises:
a digital image;
a digital media item that includes text;
a digital audio item;
a digital video item; or
any combination thereof.
8. A server comprising:
one or more processors; and
a non-transitory memory device to store instructions executable by the one or more processors to perform operations comprising:
training an artificial intelligence to create a trained artificial intelligence to generate a particular type of derivative content, the training comprising:
selecting a content creator from multiple content creators to create a selected content creator;
selecting a plurality of content items associated with the content creator to create selected content items;
training the artificial intelligence using the selected content items;
determining a content item influence of individual content items of the plurality of content items on a plurality of training techniques used to train the artificial intelligence based at least in part on determining a difference between:
a first set of parameters associated with the artificial intelligence, and
a second set of parameters associated with the trained artificial intelligence, wherein the second set of parameters comprises a plurality of weights, a plurality of biases, or any combination thereof;
aggregating the content item influence of individual content items of the plurality of content items on the plurality of training techniques used to train the artificial intelligence;
determining a creator influence of the selected content creator on the trained artificial intelligence based at least in part on the aggregating; and
including the creator influence in a static attribution vector;
after determining that the training of the artificial intelligence has been completed, determining that the trained artificial intelligence has received an input;
generating, by the artificial intelligence, an output based on the input;
creating an attribution determination based at least in part on the static attribution vector; and
initiating providing compensation to one or more of the multiple content creators based at least in part on the attribution determination.
9. The server of claim 8 , the operations further comprising:
determining a transformativeness factor of the trained artificial intelligence relative to the artificial intelligence; and
determining the static attribution vector based at least in part on the transformative factor.
10. The server of claim 8 , wherein the training further comprises:
determining a content item influence of individual content items of the plurality of content items associated with the content creator using a converging function.
11. The server of claim 10 , wherein the converging function comprises one of:
batch gradient descent;
stochastic gradient descent; or
any combination thereof.
12. The server of claim 8 , the operations further comprising:
while generating the output:
determining a plurality of activations of different weights in the trained artificial intelligence;
aggregating, for individual content creators of the multiple content creators, the plurality of activations of different weights in the trained artificial intelligence; and
determining a dynamic attribution vector based at least in part on the aggregating.
13. The server of claim 12 , the operations further comprising:
further creating the attribution determination based at least in part on
the dynamic attribution vector.
14. The server of claim 8 , wherein the trained artificial intelligence comprises:
a latent diffusion model;
a generative adversarial network;
a generative pre-trained transformer;
a variational autoencoder;
a multimodal model; or
any combination thereof.
15. A non-transitory computer-readable memory device to store instructions executable by one or more processors to perform operations comprising:
training an artificial intelligence to create a trained artificial intelligence to generate a particular type of derivative content, the training comprising:
selecting a content creator from multiple content creators to create a selected content creator;
selecting a plurality of content items associated with the content creator to create selected content items;
training the artificial intelligence using the selected content items;
determining a content item influence of individual content items of the plurality of content items on a plurality of training techniques used to train the artificial intelligence based at least in part on determining a difference between:
a first set of parameters associated with the artificial intelligence, and
a second set of parameters associated with the trained artificial intelligence, wherein the second set of parameters comprises a plurality of weights, a plurality of biases, or any combination thereof;
aggregating the content item influence of individual content items of the plurality of content items on the plurality of training techniques used to train the artificial intelligence;
determining a creator influence of the selected content creator on the trained artificial intelligence based at least in part on the aggregating; and
including the creator influence in a static attribution vector;
after determining that the training of the artificial intelligence has been completed, determining that the trained artificial intelligence has received an input;
generating, by the artificial intelligence, an output based on the input;
creating an attribution determination based at least in part on the static attribution vector; and
initiating providing compensation to one or more of the multiple content creators based at least in part on the attribution determination.
16. The non-transitory computer-readable memory device of claim 15 , the operations further comprising:
determining a transformativeness factor of the trained artificial intelligence relative to the artificial intelligence; and
determining the static attribution vector based at least in part on the transformative factor.
17. The non-transitory computer-readable memory device of claim 15 , wherein the training further comprises:
determining a difference between a first set of parameters of the artificial intelligence and a second set of parameters of the trained artificial intelligence; and
determining the creator influence of the selected content creator on the trained artificial intelligence based on the difference between the first set of parameters of the artificial intelligence and the second set of parameters of the trained artificial intelligence.
18. The non-transitory computer-readable memory device of claim 15 , wherein the trained artificial intelligence comprises:
a latent diffusion model;
a generative adversarial network;
a generative pre-trained transformer;
a variational autoencoder;
a multimodal model; or
any combination thereof.
19. The non-transitory computer-readable memory device of claim 15 , the operations further comprising:
determining a plurality of activations of different weights in the trained artificial intelligence;
aggregating, for individual content creators of the multiple content creators, the plurality of activations of different weights in the trained artificial intelligence; and
determining a dynamic attribution vector based at least in part on the aggregating.
20. The non-transitory computer-readable memory device of claim 19 , the operations further comprising:
further creating the attribution determination based at least in part on
the dynamic attribution vector.