Microsoft’s Copilot in Viva Engage and the Trouble With AI Overreliance
Microsoft announced the launch of Microsoft 365 Copilot on March 16. Riding the tsunami of interest in ChatGPT and large language AI models, Microsoft upped its investment in OpenAI, the creator of ChatGPT, from an initial $1 billion in 2019 with an additional $10 billion in late January; integrated ChatGPT inside its Bing search engine; and now comes this latest announcement.
Copilot's promise is it will be there wherever you are working — be it composing a Word document, drafting a PowerPoint presentations, struggling with an Excel spreadsheet or writing a post in Viva Engage — to build first drafts, suggest improvements, identify potential errors, suggest topics to post on and more.
In spite of the surface resemblance, this is not a re-launch of Microsoft’s infamous Clippy, the digital paperclip that annoyed Office '97 users. Copilot has a visible heritage path through Microsoft Graph and now ChatGPT, which adds some substance to the announcement. What is difficult to entertain is the speed with which these announcements are hitting the marketplace. ChatGPT itself is a pilot, yet its smooth talking UI has launched a rash of me too announcements from Google and Meta.
One can’t help but think that some of the AI researchers and engineers that work for these organizations must be shaking their heads at what the respective marketing engines are now promising. The market though, has been somewhat conditioned to grandiose announcements. Microsoft Launched Microsoft Graph in 2015, but some products have yet to be integrated with it. To be fair, announcements like Microsoft Graph and now Microsoft Copilot set a strategic intent. So while we may not see Copilot operating effectively for many years, it does create the expectation.
But how realistic are these expectations? I have previously written about concerns around the generative aspects of ChatGPT. More recently, I have been exploring the use of ChatGPT-generated code for the popular R programming language for data analytics. As an infrequent and novice R programmer, I was impressed with how ChatGPT confidently presented fully explained ‘model’ code snippets which precisely met my stated requirements. The code however only worked about 20% of the time, often having fatal errors. I suspect its training set might have included code from novices like myself — as I've written before, it needed a warning label.
Copilot for Viva Engage: The Solution to Common Internal Comms Problems?
It would be impossible to cover the breadth of the Microsoft announcement in the scope of this article, so I'll just focus on Copilot for Viva Engage (formerly Yammer). I have been benchmarking Viva Engage sites for eight years now, involving over 160 organizations, thousands of communities and millions of end-users. The two most enduring patterns I have seen where organizations are underperforming with Viva Engage are:
- Below average participation rates (i.e. not enough active participation).
- Below average levels of discussion and people to people engagement.
For participation issues we invariably recommend executive leaders actively participate, particularly by asking open questions and responding to staff at all levels. Our research has shown that a senior leader response can have a 146% increase in connection activity by the receivers.
While organizations regularly use Viva Engage as an internal social media platform, we observed that added value can accrue through relationships formed and competencies built within their communities of practice. We see best practice Viva Engage users show strong people-to-people interactions.
Jason Mayans wrote the following in his Copilot for Viva Engage announcement:
“Copilot suggests a set of topics to post about, pulled from trending content, organizational campaigns, and data-driven best practices. Copilot offers a summary of sentiment on that topic, including prompts to help jumpstart communications by providing a first draft to edit. Using these prompts, you can drill into what other leaders or employees are talking about, add specific details to complete your posts, and even choose the tone of your message before personalizing with your own voice”
Learning Opportunities
On the surface, this looks like it addresses the two main shortcomings in Viva Engage usage mentioned above. Making it easier for senior executives to post and reply helps address the participation problem. And helping staff write more engaging and informed posts and replies, should in theory improve the breadth and depth of conversations on Viva Engage.
We don’t think this is too much of a stretch. We've already identified trending posts and Topics with sentiment on Viva Engage. We've also seen ChatGPT generate competent first drafts. It brings to mind Viva Engage's existing “post on behalf of” feature, targeted at senior executive corporate communication support staff, who often aim to write in the style of their bosses. One of our clients said the downside of this feature was "staff can smell a corporate communications post a mile away" and this lack of authenticity has a negative effect.
So the questions remains: will AI be able to imitate the rushed post of a busy senior executive, typos and all?
Related Article: How Copilot in Microsoft 365 Will Impact Productivity
The Dangers of Overreliance on AI
The potential lack of authenticity speaks to the core of human relationships and the trust they embody. Will staff be willing to carefully review their AI suggested posts and make the sometimes subtle adjustment that represents their unique voice? We have already seen the consequences of an overreliance on automated co-pilots on aircraft: pilots lose their skills to manually fly planes and are reluctant to override the automation in risky situations. One could conceive a situation where a majority of Viva Engage posts are written by AI, posts which are then recycled into the training sets for the AI engines. The result would be a colorless suite of perfectly spelt and narrated conversations, akin to pitching two automated chess players against each other. Who wants that?
Perhaps we can learn from the airline industry, with the potential de-skilling effect AI can have when leaned on too heavily. Perhaps we can’t mandate how much of a post, or how many posts a human must write. But we can at the least highlight the auto-generated parts. Again, who would want to be associated with 100% auto-generated posts?
The generation question remains the most problematic. Mayans also wrote that, "Copilot also automatically aggregates information about each knowledge area into a curated view complete with copy, FAQs, and resources so people can dive deeper into the topics they are curious about." That’s something that we could all do with.
Learn how you can join our contributor community.
About the Author
Laurence Lock Lee is the co-founder and chief scientist at Swoop Analytics, a firm specializing in online social networking analytics. He previously held senior positions in research, management and technology consulting at BHP Billiton, Computer Sciences Corporation and Optimice.
Connect with Laurence Lock Lee:
Featured Research
Buyer's Guide
Read Now
Research Report
Read Now
eBook
Read Now
Research Report
Read Now
White Paper
Read Now
Research Report
Read Now
Guide
Read Now
Guide
Read Now
Related Stories
Entries for the 2023 Reworked IMPACT Awards are now Open
