Triad Principal Consultant Chris Blake explains how to get the best from Microsoft’s three easy-to-access AI services: Bing Chat Enterprise, Bing M365 and Microsoft Copilot.
Generative AI is taking the information industries by storm. Advances in computing power, AI approaches, and data access have catapulted this rapidly developing concept into a useful mainstream suite of products. Microsoft has three easy-to-access AI services: Bing Chat Enterprise, Bing M365 and Microsoft Copilot. Here are my tips on making the most of them.
Tip 1 – Understand the tools. Celebrate the differences
If asked the same questions, the three Microsoft tools can be inconsistent in their answers and presentation. And this is why:
- Bing Enterprise Chat is part of a search engine and, unsurprisingly, presents results in a search engine style, referencing websites and their content that might be useful to us.
- Bing M365 Chat is similar to Bing Enterprise Chat, but is more focused on searching content within your organisation. For many, this will likely be a document or file-heavy content, necessitating a slightly different approach to Bing Enterprise Chat.
- Microsoft Copilot is the MS Office suite’s generative assistant and is built into Edge, Word, Outlook, Teams, Excel, and PowerPoint. Each product has a different use, so the focus of Copilot in each is different, ranging from content generation, summarises, formula help, etc. But in all cases, it is designed to simplify your life as an information worker.
The above being the case, I suggest resisting the temptation to focus on one instead of using all three, playing to their strengths as you do.
Tip 2 – Don’t assume that the tools understand your question
When Copilot responds, it can be easy to focus on the given response. Also, check the question it has decided to answer. You may find that this is different from the question that you asked. That’s because Copilot is designed to accommodate far more challenging questions than your typical smart device. It is easy for your smart device to recognise “Set a timer for 20 minutes”. This simple statement can be dissected into nouns and verbs, and based on a set of supported verbs, the smart device can look for a match of the noun or noun phrase and provide a set response.
Typical tasks undertaken by Copilot and Bing are more complex, often requiring a vast dataset. Copilot’s default is to simplify the question. It may also use other similar questions to help understand yours.
So, check the question it is answering. If you are not happy, try rephrasing until you feel it has been understood. Oh, and resist the temptation to discard previous attempts. You might still find them useful.
Tip 3 – Check the source
Copilot’s results are only as good as its information sources. Although Bing tools tell you its sources, Copilot won’t unless asked. And when you ask, Copilot may not include all the references it has used to construct the response. And it may not have interpreted its reference material as you would. This is compounded by Copilot not always providing the same response to the same question, even if you ask it just seconds later. For example, I asked Copilot, “What are the contradictions in the UK government’s offshore petroleum policy?”. It gave me an answer, and when I asked the same question again, it did so this time by mentioning two new pieces of material (the Paris Agreement and UN Framework Convention on Climate Change), which it hadn’t explicitly referenced in the first response.
This lack of consistency is fundamental to Copilot, other generative AI, and AI in general. Al is different from typical computer programs in that, by design, they do not produce results by following a strict sequence of steps. The central algorithms that make results are not algorithms. Instead, they are connections of nodes that, through training, recognise patterns. For this reason, users can’t find out how the interconnections and their weightings have been used to create certain results. Check the source and remember that the sources it chooses can skew the response.
Tip 4 – Be suspicious
You probably saw this tip coming! At its current stage of development, I recommend treating generative AI as a useful adjunct to your normal way of working.
It can provide spectacularly useful results and insights. And it can mislead. Remember, it doesn’t understand what you have asked, your intentions, or the results it provides. Nor are its results inherently subjective or objective. For example, as with the previous example, Copilot initially highlighted inconsistencies in government policy. However, the second response went a step further, stating that the policy had favoured the petroleum industry and “failed” on its climate goals. Whether you agree or disagree with the latter, it is an interpretation and was not factually proven beyond doubt in its text, so in this case, the way the conclusion was presented was misleading. Your challenge is to take what is produced and modify it as needed to ensure the end product meets your needs. To do this, I would suggest asking the same question to as many AIs as possible multiple times and using their collective responses to help form your judgment.
Microsoft is quite clearly going “all in” on AI. Their $10B investment in OpenAI (aka ChatGPT) is well known, but they, like their competitors, are hedging their bets with several other AI start-up investments.
Microsoft have also announced their new Azure Maia AI processors alongside their new RISC CPU series to ensure they have a foothold in the lucrative AI market.
In my view, Copilot is a starting point for Microsoft, not an end goal. In the short term, I expect to see an improvement in the accuracy and completeness of Copilot’s responses. I also expect Copilot’s interface to improve while becoming more proactive rather than reactive as it becomes part of the products and the product itself. After all, in an AI world, why do I need to worry about which application to use?
In future iterations, I expect to ask Copilot to do something and give it the freedom to choose how best to action my request without me having to be an expert in the applications it uses. And in this, there is a broader implication. Generative AI, let alone Artificial General Intelligence, will likely be an extremely disruptive technology. Just as the Industrial Revolution brought tremendous societal and work changes, so will Generative AI. The workplace in 10 years will likely be very different from today’s as AI skills reduce the need for skilled people.
Please note that the opinions stated in this blog are those of the author. We hope that you have found this blog useful. If you are interested in AI Microsoft tools or have a question for the Triad team, please get in touch.