Microsoft 365 Copilot: ChatGPT-powered generative AI assistant explained, release date - and how does it work?

As the AI boom continues, Microsoft has added more artificial intelligence-based technology to its word processing, spreadsheet and slide presentation programmes with a new feature called Copilot.

The use of AI has grown significantly in recent years, with systems like the chatbot ChatGPT quickly integrating into daily life.

And in recent days, technology experts, including Elon Musk, have urged scientists to halt development of the technology in order to ensure that artificial intelligence does not endanger humanity.

But what can Copilot do, and how will it be changing the way users interact with Microsoft’s Office suite of products? Here is everything you need to know about it.

What is Copilot?

Gone are the days of Clippy, the often annoying paperclip mascot of programmes like Microsoft Word, that would spring up from time to time to offer you advice and assistance with certain tasks.

In 2023, Clippy has been replaced by Copilot, an AI-powered assistant designed to help users create documents, emails, presentations, and much more within Microsoft’s 365 apps and services.

“It works alongside you, embedded in the apps millions of people use everyday: Word, Excel, PowerPoint, Outlook, Teams, and more,” said Microsoft 365 head Jared Spataro. “Copilot is a whole new way of working.”

Powered by OpenAI’s GPT-4 technology, Copilot appears in the sidebar as a chatbot that Office users can call upon to generate text for documents, produce PowerPoint presentations based on Word documents, or even assist with formatting or analysing Excel data.

Available across all Microsoft Office programmes, Copilot can be used in Word to create documents based on other files from other applications, with the text produced by the AI open to be freely modified and changed by users.

Copilot’s potential uses are particularly intriguing in Microsoft Teams, where it can transcribe meetings, remind you of information you might have missed if you arrived late, or even summarise action items as they are discussed.

How does it work?

Microsoft have pointed out that Copilot won’t always be right, and during Microsoft’s AI used to unveil the feature, Spataro said: “Sometimes Copilot will get it right, other times it will be usefully wrong, giving you an idea that’s not perfect but still gives you a head start.”

Copilot uses a technique called ‘grounding’ to enhance the prompts you give it. If you ask Word to generate a document using your data, Copilot will first send the prompt to the Microsoft Graph - a powerful set of APIs (Application Programming Interfaces) that allows developers to access a wide range of data and resources across Microsoft 365 - to get the context and data before editing it and sending it to the GPT-4 large language model (LLM).

Before returning the response and commands to your Microsoft 365 app of choice, the results are forwarded to the Microsoft Graph for further grounding, security and compliance checks.

‘Grounding’ refers to the process of ensuring that the conversation between a chatbot or AI interface and the user is clear and mutually understood. This is important because it helps to avoid misunderstandings and ensures that the conversation is productive and useful.

In more traditional chatbots like ChatGPT, Grounding can be achieved through various techniques such as asking clarifying questions, providing examples or explanations, using simpler language, and providing feedback to the user.

When can I use Copilot?

A select group of businesses and individuals currently have access to the Copilot and 365 integration for testing purposes.

This testing period will be watched especially carefully by tech experts. Questions regarding ethical obligations surrounding the testing of such products have arisen as a result of worries over the speed of Microsoft’s recent releases of AI-powered products and investments.

Microsoft has said it is testing Copilot with “a small group of customers to get feedback and improve our models as we scale, and we will expand to more soon.”