According to Microsoft, their AI integration Copilot is “A whole new way to work” and will “Transform the way we work”. That’s true, but not as quickly or as well as their overstated hype suggests. Here’s why I’m cautious about the Copilot promises.
Microsoft is rushing both publicity and release to keep ahead of Google, at least in terms of public perception and media coverage. That’s been a successful strategy with most of the ‘ink’ focusing on Microsoft’s AI promises.
Usually, Microsoft announces a new product alongside some public beta release, like they did with Designer (which is continuing with its long public preview). Copilot was announced with much fanfare but nothing for anyone outside Redmond to try themselves.
As we’ve already seen with Bing, there are a lot of traps in ChatGPT spawned AI when it’s used by the general public. The Chat feature seems too easy to lure into strange and troubling rabbit holes.
I suspect that Copilot won’t work as well for complex tasks with documents from the real world as it does with Microsoft’s demos. The demo of merging things like meeting notes and documents into a Powerpoint presentation seems like a ‘bridge too far’ to me. I suspect that’ll be another Microsoft feature that doesn’t work well at first (beyond controlled demos) but gradually improves over time until it finally matches the initial promises.
Rewriting text with Bing and Copilot
Smaller, more niche tasks, are another matter. ChatGPT and Bing can do a decent job rewriting text or answering questions in selected styles. A similar feature in a Word side-pane should work well.
Here’s what Microsoft’s Bing can already do, choose a style (Professional, Casual etc), format (paragraphs, blog or ideas) and length. Copilot is based on the same technology.
ChatGPT (the base for Copilot) lets you insert text and ask for a rewrite into a different style or format like this. Expect to see Copilot do similar things for Word, Outlook and PowerPoint.
Excel’s existing Natural Language Analysis
Excel already has some ’natural language’ analysis tools, What if and Forecasting — it seems Copilot will greatly enhance those existing features by showing ‘insights’ from a table or list and also answering questions about that data.
Will Copilot mean bland and uninteresting?
The risk is that people will just accept Copilot’s work without checking that it’s accurate or even appropriate. AI rewritten text might be better than the original but also have a bland and uninteresting sameness.
Think of how many Word docs and PowerPoint decks look boring because they reuse the same templates and color schemes – now imagine that spreading to AI generated text that’s blindly accepted with no human editing.
It’ll be great to see Copilot for real and it will change how we work. But not as quickly as Microsoft is promising and there’s sure to be a downside, big changes like this always do.
Naturally, Office Watch will report on Copilot as it develops based on real world testing not Microsoft press releases.