Task-agnostic integration of human and machine intelligence

1. A method, comprising:

at a computer comprising a processor with access to executable instructions for,

building a machine learning model by starting with solicited user actions performed by users on an object and then using predicted actions generated by a machine learning system in the machine learning model in place of solicited user actions over time at least in response to accuracy of the predicted actions being substantially equal to or exceeding accuracy of the solicited user actions, the number of predicted actions increasing in number over time as accuracy of the predicted actions improves relative to accuracy of the solicited user actions, the predicted actions predicted from a set of training data that includes the solicited user actions and information about the object,

wherein the solicited user actions are collected from a sample size of the users that has a value that relates to the predicted actions in the machine learning model and that decreases over time.

2. The method of claim 1 , further comprising:

periodically measuring accuracy of the predicted actions.

3. The method of claim 2 , further comprising:

cross-validating the machine learning model to arrive at the accuracy of the predicted actions by splitting the set of training data into a first portion and a second portion, using the first portion as the set of training data, and comparing the predicted actions predicted using the first portion to the solicited user actions in the second portion.

4. The method of claim 3 , further comprising:

determining the value of the sample size of the users according to the accuracy.

5. The method of claim 1 , further comprising:

combining features extracted from the solicited user actions into a feature vector, wherein the information about the object in the set of training data includes the feature vector.

6. The method of claim 1 , further comprising:

identifying the solicited user actions in relation to reference actions in an action library.

7. The method of claim 1 , further comprising:

estimating quality of the solicited user actions for use in the set of training data.

8. The method of claim 6 , further comprising:

estimating quality of the users for use in the set of training data.

9. A method, comprising:

soliciting information from a sample size of users about actions on objects;

creating a set of training data comprising the solicited information and information about the object; and

predicting information from the set of training data using a machine learning model,

wherein the machine learning model replaces solicited information with additional information over time, beginning at least in response to accuracy of the predicted information being at or near accuracy of the solicited information and increasing in number over time as accuracy of the additional information improves relative to accuracy of the solicited information,

wherein the solicited user actions are collected from a sample size of the users that has a value that relates to the predicted actions in the machine learning model and that decreases over time.

10. The method of claim 9 , further comprising:

extracting feature vectors from the solicited information, wherein the set of training data includes the feature vectors.

11. The method of claim 9 , further comprising:

identifying, from information in the set of training data, the objects by type and the actions performed on the objects,

wherein the predicted information relates to the object type and the actions performed on the objects.

12. The method of claim 11 , wherein the object types correspond with different types of media objects.

13. The method of claim 9 , further comprising:

extracting keywords from text documents, wherein the set of training data includes the keywords.

14. A system, comprising:

a computing device accessible to data over a network, the computing device comprising a processor and memory with instructions configuring the processor for,

building a machine learning model by starting with solicited user actions performed by users on an object and then using predicted actions generated by a machine learning system in the machine learning model in place of solicited user actions over time at least in response to accuracy of the predicted actions being substantially equal to or exceeding accuracy of the solicited user actions, the number of predicted actions increasing in number over time as accuracy of the predicted actions improves relative to accuracy of the solicited user actions, the predicted actions predicted from a set of training data that includes the solicited user actions and information about the object, wherein the solicited user actions are collected from a sample size of the users that has a value that relates to the predicted actions in the machine learning model and that decreases over time.

15. The system of claim 14 , wherein the instructions further comprising instructions configuring the processor for,

serving web pages configured for receiving the solicited user actions.

16. The system of claim 15 , wherein the instructions further comprising instructions configuring the processor for the web pages to,

supply tasks for evaluation by the user; and

receive the solicited user actions in response to the tasks.

17. The system of claim 14 , further comprising instructions configuring the processor for,

indexing the solicited user actions for use in the set of training data according to a type of action.

18. The system of claim 14 , further comprising instructions configuring the processor for,

indexing the object for use in the set of training data according to a type of object.