IBM - Artificial Intelligence Reference Architecture (Conversation Diagram)

IBM - Artificial Intelligence Reference Architecture (Conversation Diagram)

Computing with AI is quickly emerging as a transformative technology that enables organizations to gain a business advantage. AI technology augments human expertise to unlock new intelligence from vast quantities of structured and unstructured data, enabling the development of deep, predictive insights.

This diagram represents the Conversation architecture. It is a runtime architecture showcasing the components involved in a trained and deployed conversation system. The Discovery architecture diagram uses the IBM Watson® Discovery service to discover trends and patterns in diverse structured and unstructured data sets.

The Flow on the Conversation Architecture Diagram

The case on the diagram is this: a home improvement store enhances its customer service with a cognitive decision assistant for do-it-yourselfers, professionals, and store associates.

Runtime flow - Step 1 - A customer accesses the mobile application by using voice to ask what kind of connector is needed for a dishwasher. (The manufacturer does not provide the connector.)

Runtime flow - Step 2 Application logic determines that the request is a voice request and invokes the extended speech-to-text component.

Runtime flow - Step 3 The extended speech-to-text component converts the voice request to text.

Runtime flow - Step 4 Application logic uses the text to check with the conversation component for a trained response. Conversation is not trained with a specific response to connectors for a dishwasher.

Runtime flow - Step 5 Application logic uses the API to determine whether the information is available in the core ERP system. The system has the manual and information about the type of connector. Still, it lacks detailed information on how to connect the dishwasher to the garbage disposal using the connector.

Runtime flow - Step 6 The application logic checks whether the discovery component is trained to retrieve specific information. Discovery is trained and gives the information to the application logic.

Runtime flow - Step 7 Application logic uses the conversation component to get the correct response.

Runtime flow - Step 8 Application logic uses the extended text-to-speech component to translate the response from text to voice.

Runtime flow - Step 9 Application logic sends the voice response to the customer's mobile application.

Design time flow - Step A The ingestion application crawls customer feedback and comments on social media.

Design time flow - Step B The ingestion application uses the discovery APIs to add the social media content to the collection.

Design time flow - Step C The ingestion application crawls product information, product catalogs, descriptions, and product manuals stored in the customer data center.

Design time flow - Step D The ingestion application uses the discovery APIs to add the data center content to the collection.

Design time flow - Step E Subject matter experts train the conversation.