Document AI: Human-in-the-loop deprecation

[joostvd]

We've been using HITL for our operations by manually triggering the `humanReviewConfig` endpoint when certain fields were not found / have a low confidence score. Our internal labelling team then reviews these documents in DataCompute.

Furthermore, we set up a subscriber to the `ReviewDocument` event through EventArc and PubSub. In this way, we could handle the reviewed document as if they were automatically extracted.

This setup has served us well. After receiving the deprecation message of HITL (https://cloud.google.com/document-ai/docs/deprecation), we were wondering if this setup would stop working entirely, or parts of it would be still possible? 

[nceniza]

Yes, it will be available for a time depends on the ToS but will shut down nonetheless. 

After a service, feature, or product is officially deprecated, it continues to be available for at least the period of time defined in the Terms of Service. After this period of time, the service is scheduled for shutdown.

Also its recommended working with a Google Cloud certified partner like Devoteam, Searce, or Quantiphi.

[GS_23]

Situation: HITL appears to be deprecated.

Question: Is there a GCP that will allow our internal user (that has our GCP credentials) to verify and correct data extracted by our Document AI processors, and manually correct extracted data, BEFORE it's used in critical business applications?

Reason: HITL aims to improve the accuracy of machine learning models by incorporating human feedback into the training process. I'm worried that we don't have this anymore.