Granted Patent B2

US 12,731,066 · App. 16/382,016 · Granted Sep 8, 2026

Inventor: Christine Meinders (Altadena, CA)

G06N20/00

Abstract

Claims (78)

1 . A computer-implemented method, comprising:

receiving input, by one or more processors of a computing device and from a communications channel at an interface on the computing device, wherein the input comprises a dataset, metadata comprising AI tags comprising machine-readable content descriptors, demographic information, input data, context descriptors, and technical data corresponding to the dataset, and an output medium;

selecting, by the one or more processors, and based on each of (i) the received dataset, (ii) the metadata corresponding to the dataset, and (iii) the output medium, at least one algorithm selected from a plurality of algorithms, including: a decision tree, a Bayesian network, an artificial neural network, a support vector machine, a convolutional neural network, and a capsule network;

training, via the one or more processors of the computing device, the at least one algorithm using the received dataset, the metadata corresponding to the dataset, and the output medium with the at least one algorithm to yield an output, wherein the output comprises a trained artificial intelligence system, the received dataset, and the output medium;

applying, via the one or more processors of the computing device, a trained machine learning algorithm to generate an inference based at least on the metadata characterizing a relationship between the received dataset and the output;

determining, via the one or more processors of the computing device, and based on the inference, whether the trained artificial intelligence system comprises at least one unwanted bias in a plurality of biases using pattern recognition, wherein the plurality of biases comprise two or more of social biases, cultural biases, gender biases, racial biases and interaction biases;

identifying, by the one or more processors, a portion of the received input corresponding to the determined unwanted bias based on determining that the artificial intelligence system comprises the at least one unwanted bias, wherein identifying the portion of the received input corresponding to the determined unwanted bias further comprises processing the metadata corresponding to the dataset using the trained machine learning algorithm, wherein the metadata comprises identification of the at least one unwanted bias in the plurality of biases;

providing the artificial intelligence system and the identified portion of the received input corresponding to the determined unwanted bias at the interface on the computing device;

removing, by the one or more processors, the identified portion from the received input to yield updated input;

retrieving, via the one or more processors of the computing device, supplementary input data in a database of artificial intelligence systems, wherein the supplementary input data corresponds to the identified portion of the received input and does not comprise the at least one unwanted bias; and

retraining, via the one or more processors of the computing device, the at least one algorithm using the supplementary input data, a second dataset, and supplementary metadata comprising AI tags and corresponding to (i) the second dataset, and (ii) the supplementary input data to yield an updated output, wherein the updated output comprises an updated artificial intelligence system that does not comprise the at least one unwanted bias.

2 . The method of claim 1 , wherein selecting at least one algorithm further comprises:

determining whether the received input corresponds to requirements associated with each algorithm in the plurality of algorithms; and

selecting algorithms of the plurality of algorithms, based on determining that the received input corresponds to requirements associated with the selected algorithms.

3 . The method of claim 1 , wherein the input further comprises any of: a format for the output, a supplementary dataset, a type of the dataset, and input consideration variables.

4 . The method of claim 1 , wherein the at least one algorithm comprises an artificial intelligence model selected from a plurality of artificial intelligence approaches, including: an artificial narrow intelligence approach, a non-symbolic artificial intelligence approach, a symbolic artificial intelligence approach, a hybrid symbolic and non-symbolic artificial intelligence approach, and a statistical artificial intelligence approach.

5 . The method of claim 1 , wherein the at least one algorithm was trained, via the computing device, on a subset of a database of artificial intelligence systems, the subset comprising artificial intelligence systems with at least one of: datasets comprising metadata corresponding to metadata of the received dataset, and the output medium.

6 . The method of claim 1 , wherein the output further comprises:

an indication of whether the at least one algorithm successfully processed the received input.

7 . The method of claim 1 , further comprising:

displaying the supplementary input data at the interface on the computing device.

8 . The method of claim 1 , further comprising:

displaying the updated output at the interface on the computing device.

9 . A method, comprising:

receiving input, by one or more processors of a computing device and from a communications channel at an interface on the computing device, wherein the input comprises an artificial intelligence system comprising at least one algorithm selected from a plurality of algorithms, including: a decision tree, a Bayesian network, an artificial neural network, a support vector machine, a convolutional neural network, and a capsule network;

determining, via the one or more processors of the computing device, and based on an inference characterizing the received artificial intelligence system generated by an application of a trained machine learning algorithm, whether metadata comprising AI tags comprising machine-readable content descriptors, demographic information, input data, context descriptors, and technical data associated with the received artificial intelligence system comprises at least one unwanted bias in a plurality of biases using pattern recognition, wherein the plurality of biases comprise two or more of social biases, cultural biases, gender biases, racial biases and interaction biases;

identifying, by the one or more processors, a portion of the received artificial intelligence system corresponding to the at least one unwanted bias, wherein identifying the portion of the received input corresponding to the determined unwanted bias further comprises processing the metadata using the trained machine learning algorithm, wherein the metadata comprises identification of the at least one unwanted bias in the plurality of biases;

displaying, at the interface on the computing device, the identified portion and the at least one unwanted bias;

retrieving, via the one or more processors of the computing device, supplementary input data from a database of artificial intelligence systems, wherein the supplementary input data corresponds to the identified portion of the received input and does not comprise the at least unwanted one bias;

displaying the supplementary input data at the interface on the computing device;

receiving a request, via the interface for the computing device, to process a second selection of input data comprising the received input with the supplementary input data in place of the identified portion; and

training, via the one or more processors of the computing device, the artificial intelligence system using the second selection of input data to yield an output, wherein the output is an updated artificial intelligence system that does not comprise the at least one unwanted bias.

10 . The method of claim 9 , further comprising:

displaying the output at the interface on the computing device.

11 . A non-transitory computer-readable medium in which is embedded computer-readable code that when loaded on a computing device causes the computing device to perform the steps of:

receiving input, by one or more processors of the computing device and from a communications channel at an interface on the computing device, wherein the input comprises a dataset, metadata comprising AI tags comprising machine-readable content descriptors, demographic information, input data, context descriptors, and technical data corresponding to the dataset, and an output medium;

selecting, by the one or more processors, and based on each of (i) the received dataset, (ii) the metadata corresponding to the dataset, and (iii) the output medium, at least one algorithm selected from a plurality of algorithms, including: a decision tree, a Bayesian network, an artificial neural network, a support vector machine, a convolutional neural network, and a capsule network;

training, via the one or more processors of the computing device, the at least one algorithm using the received dataset, the metadata corresponding to the dataset, and the output medium with the at least one algorithm to yield an output, wherein the output comprises a trained artificial intelligence system, the received dataset, and the output medium;

applying, via the one or more processors of the computing device, a trained machine learning algorithm to generate an inference based at least on the metadata characterizing a relationship between the received dataset and the output;

determining, via the one or more processors of the computing device, and based on the inference, whether the artificial intelligence system comprises at least one unwanted bias in a plurality of biases using pattern recognition, wherein the plurality of biases comprise two or more of social biases, cultural biases, gender biases, racial biases and interaction biases;

identifying, by the one or more processors, a portion of the received input corresponding to the determined unwanted bias based on determining that the artificial intelligence system comprises the at least one unwanted bias, wherein identifying the portion of the received input corresponding to the determined unwanted bias further comprises processing the metadata corresponding to the dataset using the trained machine learning algorithm, wherein the metadata comprises identification of unwanted biases in the plurality of biases corresponding to each of the received input;

providing the artificial intelligence system and the identified portion of the received input corresponding to the determined unwanted bias at the interface on the computing device;

removing, by the one or more processors, the identified portion from the received input to yield updated input;

retrieving, via the one or more processors of the computing device, supplementary input data in a database of artificial intelligence systems, wherein the supplementary input data corresponds to the identified portion of the received input and does not comprise the at least one unwanted bias; and

retraining, via the one or more processors of the computing device, the at least one algorithm using the supplementary input data, a second dataset, and supplementary metadata comprising AI tags and corresponding to (i) the second dataset, and (ii) the supplementary input data to yield an updated output, wherein the updated output comprises an updated artificial intelligence system that does not comprise the at least one unwanted bias.

12 . The non-transitory computer-readable medium of claim 11 , further comprising:

displaying the supplementary input data at the interface on the computing device.

13 . The non-transitory computer-readable medium of claim 12 , further comprising:

displaying the updated output at the interface on the computing device.

14 . The non-transitory computer-readable medium of claim 11 , wherein identifying the portion of the received input corresponding to the determined unwanted bias further comprises:

processing metadata associated with each of the received input, wherein the metadata comprises identification of unwanted biases in the plurality of biases corresponding to each of the received input.

15 . A non-transitory computer-readable medium in which is embedded computer-readable code that when loaded on a computing device causes the computing device to perform the steps of:

receiving input, by one or more processors of the computing device and from a communications channel at an interface on the computing device, wherein the input comprises an artificial intelligence system comprising at least one algorithm selected from a plurality of algorithms, including: a decision tree, a Bayesian network, an artificial neural network, a support vector machine, a convolutional neural network, and a capsule network;

determining, via the one or more processors of the computing device, and based on an inference characterizing the received artificial intelligence system generated by an application of a trained machine learning algorithm, whether metadata comprising AI tags comprising machine-readable content descriptors, demographic information, input data, context descriptors, and technical data associated with the received artificial intelligence system comprises at least one unwanted bias in a plurality of biases using pattern recognition, wherein the plurality of biases comprise two or more of social biases, cultural biases, gender biases, racial biases and interaction biases;

identifying, by the one or more processors, a portion of the received artificial intelligence system corresponding to the at least one unwanted bias, wherein identifying the portion of the received input corresponding to the determined unwanted bias further comprises processing the metadata using the trained machine learning algorithm, wherein the metadata comprises identification of the at least one unwanted bias in the plurality of biases;

displaying, at the interface on the computing device, the identified portion and the at least one unwanted bias;

retrieving, via the one or more processors of the computing device, supplementary input data from a database of artificial intelligence systems, wherein the supplementary input data corresponds to the identified portion of the received input and does not comprise the at least one unwanted bias;

displaying the supplementary input data at the interface on the computing device;

receiving a request, via the interface for the computing device, to process a second selection of input data comprising the received input with the supplementary input data in place of the identified portion; and

training, via the one or more processors of the computing device, the artificial intelligence system using the second selection of input data to yield an output, wherein the output is an updated artificial intelligence system that does not comprise the at least one unwanted bias.

16 . The method of claim 1 , wherein the plurality of biases comprise one or more of social biases, cultural biases, gender biases, racial biases, and interaction biases created through usage over time.

17 . The method of claim 9 , wherein the plurality of biases comprise one or more of social biases, cultural biases, gender biases, racial biases, and interaction biases created through usage over time.

18 . The non-transitory computer-readable medium of claim 11 , wherein the plurality of biases comprise one or more of social biases, cultural biases, gender biases, racial biases, and interaction biases created through usage over time.

19 . The non-transitory computer-readable medium of claim 15 , wherein the plurality of biases comprise one or more of social biases, cultural biases, gender biases, racial biases, and interaction biases created through usage over time.

20 . The method of claim 1 , wherein the trained machine learning algorithm is a trained neural network.

21 . The method of claim 9 , wherein the trained machine learning algorithm is a trained neural network.

22 . The non-transitory computer-readable medium of claim 11 , wherein the trained machine learning algorithm is a trained neural network.

23 . The non-transitory computer-readable medium of claim 15 , wherein the trained machine learning algorithm is a trained neural network.

24 . The method of claim 1 , wherein the AI tags comprise content descriptors of one or more of the input dataset, the second dataset, and the supplementary input data.

25 . The non-transitory computer-readable medium of claim 11 , wherein the AI tags comprise content descriptors of one or more of the input dataset, the second dataset, and the supplementary input data.

26 . The method of claim 1 , wherein the dataset comprises real time data.

27 . The non-transitory computer-readable medium of claim 11 , wherein the dataset comprises real time data.

28 . The method of claim 1 , wherein the input comprises one or more selected from the group comprising a software application, a hardware device, a product utilizing artificial intelligence.

29 . The non-transitory computer-readable medium of claim 11 , wherein the input comprises one or more selected from the group comprising a software application, a hardware device, a product utilizing artificial intelligence.

30 . The method of claim 1 , wherein the input comprises a physical form.

31 . The non-transitory computer-readable medium of claim 11 , wherein the input comprises a physical form.

32 . The method of claim 9 , wherein the AI tags comprise content descriptors of the received artificial intelligence system.

33 . The non-transitory computer-readable medium of claim 15 , wherein the AI tags comprise content descriptors of the received artificial intelligence system.

Continuity (2)

References Cited (34)