Copilot creation using conversational builder

  • 12 minutes

In this unit, you'll create a copilot by using Microsoft Copilot Studio and natural language.

Natural Language Understanding (NLU)

A fundamental component of Conversational AI is Natural Language Understanding (NLU). As human beings, when we talk to another person, we ask questions in our own language using words we know. A key component of conversational experiences for a user is the ability to ask questions naturally, while using vocabulary they understand, not some special language or syntax that has to be learned.

When a question is asked by a person, the person who is asked the question understands the intent from the question and acts accordingly. This discovered or assumed intent might be following up with a sentence, comment or even another question. What can make these interactions complex is matching the intent to the correct intention of the original user asking the question. This is understood using NLU technology, Natural Language Understanding, so a Conversational AI application can understand the natural language human beings use, recognize intents from questions, and then acts accordingly.

In traditional Conversational AI Development, developers have had to create, train, test and deploy a separate NLU layer with their logic and UI layers together as well as additional layers and integration around specific follow-up behavior. In Microsoft Copilot Studio, our aim is to democratize Conversational AI and make it easier for organizations to get started, manage, deploy and iterate on their copilots by providing both low-code and pro-code authoring experiences and capabilities. Copilot Studio comes straight out of the box with both Generative AI features, such as the Generative Answers, which you learned about in the previous task, and also its own pretrained Natural Language Understanding (NLU) capabilities. This means a maker doesn't need to use or provision their own NLU service and can get started straight away building authored topics and train their copilot simply by entering Trigger Phrases within a topic. Additionally, standard and custom entities can be detected on intent recognition or within an ongoing conversation by extracting the subject of user's sentence and being stored and used as a variable.

Additionally, when creating conversational experiences and more than one authored topic in Copilot Studio, makers need to ensure they limit their topic overlap to avoid disambiguation scenarios that can occur when an NLU has two or more high ranking intents, which have been matched from the user's question. This means the NLU has to confirm with the user who asked the question what the intent was between those topics that has been matched (Did You Mean X, or Y?). Microsoft Copilot Studio has features including Topic Overlap Detection to help makers along the journey of copilot building and the ability to turn off topics from showing up in the Did You Mean experience.

Generative AI features

Generative AI is an artificial intelligence technology that uses large language models (LLMs) to generate original content and provide natural language understanding and responses. In Copilot Studio, you can use the following generative AI features to retrieve and create content, either individually or all together.

  • Create an instantly useful copilot. With no manual authoring of topics required, an "empty" copilot can generate answers based on knowledge sources you specify such as websites and files.

  • Harness AI general knowledge. When this option is enabled, the copilot can answer general questions unrelated to your specific knowledge sources or topics.

  • Author topics using natural language. Describe what you want your topic to do, and Copilot Studio creates it for you. Your copilot includes conversational responses and multiple types of nodes. Use the suggested default topic or as a starting point for further development.

  • Enable generative actions. Let the copilot select the most appropriate topics and actions at runtime.

Using generative AI in Copilot Studio transforms how you build copilots, significantly reducing manual work and configuration.

Create a copilot with natural language

You can create a copilot by describing its purpose is and how it should act using natural language.

  1. When creating a copilot you can enter a description and instructions.

    Screenshot of the dialog to create copilot with natural language.

  2. In the Description field, you describe the purpose of this copilot and how it can help.

  3. In the Instructions field, you tell your copilot what its primary purpose is and how it should act.

  4. For example, enter Demonstrate the use of natural language understanding in copilots in the description text box and Create a copilot for topics relating to natural language understanding (NLU) for copilots creating with Microsoft Copilot Studio. The copilot answers in 20 words or less. Use bullets wherever possible. in the instructions text box and create the copilot.

    Screenshot of the details to create copilot with natural language.

    Tip

    To learn about how to write custom instructions, see Prompt engineering techniques and Use prompt modification to provide custom instructions to your copilot.

  5. The test pane displays something similar to the screenshot below.

    Screenshot of the returned message for natural language copilot.

  6. Enter In which ways can copilot use natural language? in the test pane.

  7. You should see a response similar to the screenshot below.

    Screenshot of the prompt return from natural language copilot.

    Note

    You do not need to supply knowledge to use this feature but adding knowledge will increase the relevance of the responses.