Back to results
US 12,586,366 B1Granted
Artificial intelligence based content detection
AssigneeAmazon Technologies, Inc.
InventorsDaniel Neil MacTiernan, Katherine Goodrich, Aaron Sahagun
Life of this patent
Estimating the term…
Abstract
The AI detection system processes content. The AI detection system extracts text with optical character recognition technology. A machine learning model is applied to the text to determine a likelihood of whether the text was generated by AI. A machine learning model is applied to the text to determine a likelihood whether the text includes a flaggable claim. A machine learning model can be applied to the image to determine a likelihood of whether the image was generated by AI. The AI detection system causes an indicator, such a warning, to also be presented to the user computing device that received the content.
Claims20
A computer-implemented method comprising:
- receiving (1) first content for presentation comprising image data and (2) metadata associated with the first content, wherein the metadata comprises a category for the first content;
- performing an optical character recognition operation on the image data that results in text data;
- applying a first machine learning model to first input based at least in part on the text data, wherein the first machine learning model outputs a first likelihood that the text data was generated by artificial intelligence (AI);
- applying a second machine learning model to second input based at least in part on the text data, wherein the second machine learning model outputs a second likelihood that the text data includes a false, misleading, or opinion claim;
- applying a third machine learning model to third input based at least in part on the text data, wherein the third machine learning model outputs a derived category for the text data;
- determining that the first likelihood satisfies a first threshold;
- determining that the second likelihood satisfies a second threshold;
- determining that the category for the first content does not match the derived category;
- generating structured data, wherein the structured data comprises (i) a first indicator that the first content was possibly generated by AI, (ii) a second indicator that the first content possibly includes the false, misleading, or opinion claim, and (iii) a third indicator that the category for the first content does not match the derived category;
- presenting, via a graphical user interface, the first content;
- and
- presenting, via the graphical user interface, the first indicator, the second indicator, and the third indicator.
9 more elements
A system comprising:
- a non-transitory computer-readable medium to store computer-executable instructions;
- and
- a computer hardware processor in communication with the non-transitory computer-readable medium, wherein the computer hardware processor executes the computer-executable instructions to at least:
- receive first content for presentation comprising image data;
- perform an optical character recognition operation on the image data that results in text data;
- apply a first machine learning model to first input based at least in part on the image data, wherein the first machine learning model outputs a first likelihood that the image data was generated by artificial intelligence (AI);
- apply a second machine learning model to second input based at least in part on the text data, wherein the second machine learning model outputs a second likelihood that the text data includes a flaggable claim;
- determine that the first likelihood satisfies a first threshold;
- determine that the second likelihood satisfies a second threshold;
- generate structured data, wherein the structured data comprises (i) a first indicator that the first content was possibly generated by AI, and (ii) a second indicator that the first content possibly includes the flaggable claim;
- present, via a graphical user interface, the first content;
- and
- present, via the graphical user interface, the first indicator and the second indicator.
10 more elements
A non-transitory computer-readable storage medium storing computer-executable instructions that when executed by one or more computer hardware processors perform operations comprising:
- receiving first content for presentation comprising image data;
- performing an optical character recognition operation on the image data that results in text data;
- applying a first machine learning model to first input based at least in part on the image data, wherein the first machine learning model outputs a first likelihood that the image data was generated by artificial intelligence (AI);
- applying a second machine learning model to second input based at least in part on the text data, wherein the second machine learning model outputs a second likelihood that the text data includes a flaggable claim;
- determining that the first likelihood satisfies a first threshold;
- determining that the second likelihood satisfies a second threshold;
- generating structured data, wherein the structured data comprises (i) a first indicator that the first content was possibly generated by AI, and (ii) a second indicator that the first content possibly includes the flaggable claim;
- presenting, via a graphical user interface, the first content;
- and
- presenting, via the graphical user interface, the first indicator and the second indicator.
7 more elements
Dependencies are inferred from the claim text.