Microsoft recently announced its low-code tool Microsoft Copilot Studio at Ignite 2023. Copilot Studio users can build standalone Copilots or customize Microsoft Copilot for Microsoft 365. This enables AI-driven conversational capabilities for ad-hoc enterprise use cases.
Copilot Studio is an end-to-end conversational AI platform that enables IT professionals and manufacturers to create and customize Copilot using natural language and graphical interfaces. Copilot Studio users can design, test, and publish Copilots that can later be used within a Microsoft 365 context or for custom enterprise purposes.
A standalone copilot is an application that addresses users' natural language queries through a conversational interface. In addition to handling user interactions, the co-pilot may be required to retrieve information from authorized databases or perform actions on the user's behalf on external systems. Copilot Studio uses the same authoring canvas as Microsoft Power Virtual Agent.
Manufacturers implement conversational dialogs as a tree of nodes, where each node represents an action, such as displaying information to the user, prompting the user with a question, calling an API, or running a Power Automate flow. Some (or all) of the dialog tree can be configured as follows: topic. Topics often include the following set: trigger phrase– Phrases, keywords, and questions that end users are likely to use to express the needs addressed by the topic.
Copilot Studio AI analyzes end users' natural language input and assigns a confidence score to each configured topic. The topic confidence score reflects how close the user input is to the topic's trigger phrase.
If the confidence scores of some topics exceed a confidence threshold (e.g., 85%), the end user may be asked to select the applicable topic (disambiguation mechanism). If only one topic clears the confidence threshold, that topic's dialog runs immediately. Microsoft Copilot Studio can also delegate natural language understanding to the Azure AI Language Studio suite of tools.
Authors have access to large-scale language model generation capabilities within the topic dialog. Gary Pretty, his manager of principal products at Microsoft, demonstrated how a prospective customer from Holland America Line could query his standalone bot for information about cruises (e.g., “Do I need a passport for a cruise?”). did. Makers can create their bots in a few clicks by simply referencing the code below. www.hollandamerica.com as an important source of information. The bot passes the end user's input to the generative model and uses the referenced content to answer the query (e.g. “Yes, I need a passport for a cruise”) […]”). Bots continue conversations while tracking context and conversation history, allowing users to implicitly or explicitly reference past information.
This use case corresponds to what has been widely seen in generative models like ChatGPT. However, this time the bot's answer is based on the referenced content. Such grounding can help reduce false responses from bots.
CoPilot can also provide natural language interfaces to application programming interfaces. We detailed the “Get Excursions” topic where the bot asks the user if they have an existing reservation. The user then enters the reservation number. The bot then calls the relevant API (via Power Automate) and displays the results. However, goal-oriented applications may require a large amount of domain-specific handcrafting that correlates to the complexity of the goal (number of steps, conditions and branches, management of errors and edge cases, etc.).
At Ignite 2023, David Conger, principal product manager at Microsoft, provided an example of complex orchestration of APIs to achieve user goals. Microsoft 365 Copilot can create a PowerPoint presentation from a text document and then modify that document in response to your commands. In order to pinpoint the steps to take, safely perform the identified actions, and recover from errors, Conger says that Microsoft has adapted his Domain Specific Language for Office (ODSL) LLM to suit his LLM. I explained that I had asked for it. Microsoft 365 Copilot dynamically constructs prompts with relevant information within token limits so that LLM can generate the correct his ODSL program. The ODSL program is then parsed and validated using automatic code modification, transpiled to the native Office API, and executed.
Arguably, many enterprise use cases are much simpler and fit a no-code approach. The combination of generative AI and no-code authoring tools creates engaging demos for the simplest use cases. However, technology purchasers may want to tie the licensing, configuration, and no-code development costs that come with the technology to specific, valuable use cases for their own specific companies.
CoPilot can be distributed through a variety of channels, including Microsoft Teams, your website, and Skype. Microsoft Copilot for Microsoft 365 can further leverage Copilots created with Copilot Studio.
Manufacturers can also use multilingual copilots. This allows you to communicate with your customers in different languages ​​while keeping all your content on her one copilot. In many cases, such a co-pilot will automatically detect the user's desired language based on her web browser settings and respond in the same language.
The functionality of Microsoft Power Virtual Agents (also known as Power VA) is fully included in Microsoft Copilot Studio. Copilot Studio integrates with Microsoft Azure OpenAI Studio, Azure Cognitive Services, Azure Bot Service, and other Microsoft conversational AI technologies. The Copilot Studio and Copilot for Microsoft 365 integration is now available in public preview.
