Category: Artificial Intelligence

Microsoft Search – helping users without AI

Posted by on in Artificial Intelligence, Microsoft 3651 Comment

Introducing Microsoft Search in Microsoft 365

Whilst Artificial Intelligence (AI) is riding high on the peak of inflated expectations, we can often forget that there are already existing practical information management solutions to solve some of our existing problems. Ones that we can use right now, without resorting to AI. One of these is the out-of-the-box Microsoft Search configuration.

Here at Metataxis, we are sure at some point soon, Microsoft Search will be effectively (and possibly entirely) Copilot driven, but there are many things you can do to help your users find stuff.

Right now.

What is Microsoft Search?

Microsoft Search claims to “empower people to find the information they need by unlocking knowledge and expertise.”

Integrated with Microsoft 365, Microsoft Search is a secure, easy-to-manage search experience that works across all Microsoft 365 data and platforms to deliver highly relevant results. Using deep learning models and Microsoft Graph connectors, you can even index your external systems to deliver intelligent results across all Microsoft 365 apps and services, regardless of where that information resides.

How to configure Microsoft Search

Microsoft Search configuration can be set up centrally, all in one place, and is automatically applied to SharePoint, Bing, Teams and even the Microsoft 365 app to intelligently connect and consolidate information across the organisation. Having worked on dozens of Microsoft 365 implementations, we are constantly surprised that very few people are using these advanced search configuration options.

Most of them are easy to implement (accepted some are not) and can increase productivity, keep employees engaged and deliver actionable insights.

Microsoft search image

In this forthcoming series of blogs, Marc Stephenson, Director here at Metataxis, will look at these advanced search configuration features available in Microsoft 365 and present a steer on how to best use them. He’ll share with us what really works and what the documentation doesn’t explain well – or at all.

He’s gone through the pain of this, so we don’t have to! Watch this space for his next coming blog soon.

Microsoft Copilot vs ChatGPT – which one works best?

Posted by on in Artificial Intelligence

How many animals did Moses take on the ark?  

Comparing how would ChatGPT and Microsoft Copilot answer this question

You could be forgiven for saying “two of each species were taken on the Ark” which is a common response to this question. The correct answer of course is “none” because Noah built the Ark, while Moses liberated the Israelites from Egypt.

The so-called “Moses Illusion” highlights “shallow semantic processing,” which causes most of us to overlook errors, perhaps because our attention is drawn elsewhere, such as solving the problem of how many animals were on Noah’s Ark.

At Metataxis, we have been exploring the uses for AI, considering the practical information management applications as well as the information architecture implications. When I learned about the Moses Illusion, I just had to see whether common AI tools such as ChatGPT and Microsoft Copilot would fall for it. Of course, the answer is no, though each of the tools handled the query quite differently.

Let’s see how they compare:

ChatGPT

ChatGPT was not take in by the Moses Illusion and in fact, asked me which of the following two answers I would prefer:

Answer 1: “Moses is not associated with the story of the ark. The biblical account of Noah and the ark is found in the Book of Genesis, where Noah, not Moses, is the central figure. According to the Bible, God instructed Noah to build an ark to save himself, his family, and pairs of every kind of land-dwelling animal from the impending flood. The specific number of animals taken onto the ark is not specified in the Genesis account.”

ChatGPT logo

or…

Answer 2: “Moses did not take any animals on the ark. The story of Noah’s Ark is found in the Bible, specifically in the book of Genesis. According to the biblical account, Noah, not Moses, was instructed by God to build an ark to save himself, his family, and two of every kind of animal from a global flood. Moses is a different biblical figure, known for leading the Israelites out of Egypt and receiving the Ten Commandments on Mount Sinai.”

What is interesting about these responses is that they try to achieve two things:

1) Answer the question as it was intended; albeit in typical ChatGPT non-answer form: “that information is not available to me, Dave”

2) Highlight the cognitive oversight inherent in the question and provide disambiguation between Moses and Noah.

So, in terms of performance there is some impressive predictive power, but the source material available to answer the question perhaps is not as great. I also must recognise that no matter how many ways I ask the question, ChatGPT has the good sense to repeat that the question cannot be definitively answered. (“It’s ineffable.”) This is fair.

Were I interested in actual attempts to calculate the number of animals in the Ark, ChatGPT isn’t going to get into the specifics. ChatGPT isn’t going to suddenly start quoting James Ussher.

The tool does try to be even-handed, pointing out whether views are theological or scientific, but I can’t help feeling a bit like it doesn’t think much of Creationism.

When you ask “Has anyone tried to calculate how many animals were on the ark?“, you receive the more nuanced but equally dismissive statement which starts with: “The question of how many animals were on Noah’s Ark has been a topic of discussion and debate for centuries. Various scholars, theologians, and researchers have attempted to calculate the number of species or animals involved in the biblical account of the flood. However, it’s important to note that the Bible provides limited details on the exact number or species of animals taken onto the ark.”

ChatGPT still requires a series of prompts from me to get a fuller answer, say if I were interested more in the attempts people have made to answer the question of the animals in the Ark, or about the Moses Illusion itself. This means that the onus is still on the user to ask the right question and most significantly, not to give up after asking the first question. I had to formulate five questions to get the full answer I needed.

It’s not enough that it knows what I meant to ask. I still need to be more inquisitive to get my best answer. Through its often phlegmatic responses, ChatGPT has prompted me to provide more specific and thoughtful questions. Is it teaching me or am I teaching it?

Microsoft Copilot

Having completed this experiment in ChatGPT, I was now keen to test how Copilot for the web dealt with exactly the same query.

Not only was Microsoft Copilot not taken in by the Moses Illusion, it also saw me coming:

Microsoft Copilot for 365 Logo

Copilot replied: “The question of how many animals Moses took on the ark is an intriguing one, but it actually involves a delightful cognitive illusion known as the Moses Illusion. Allow me to explain: When asked, “How many animals of each kind did Moses take on the Ark?” approximately 50% of people confidently respond with “two” instead of the correct answer, which is “none.” Despite knowing that it was Noah, not Moses, who took animals onto the Ark, the illusion persists. The similarity between the names Noah and Moses plays a significant role in this cognitive mix-up. Their semantic and phonological resemblance—both being two-syllable names with similar stress patterns—leads to this intriguing error.”

To add insult to injury, this explanation of the Moses Illusion is footnoted with a link to a reputable source: Oxford Reference. (Moses illusion – Oxford Reference) Immediately I can see how Copilot outperforms ChatGPT by being up front with its receipts, giving me the opportunity to assess the quality of the sources for its responses.

And that’s not all. Of course, Copilot answers the question correctly, again citing its sources: “The correct answer: In reality, it was Noah who followed God’s instructions and took two of every kind of animal onto the Ark. This monumental task ensured the survival of various species during the great flood. So, Moses didn’t take any animals on the Ark; it was all Noah’s doing”

Unlike ChatGPT, Copilot for the web is willing to draw on theological sources to answer questions. These sources are cited so that it is left up to the individual to use judgement to evaluate answers. What is noticeable is that Copilot seems to accept the story of Noah’s Ark uncritically without any of the caveats about worldviews given to us by ChatGPT.

The icing on the cake though, is that Copilot goes on, unbidden to attempt to answer the actual question: “Fun Fact: If we were to estimate the number of animals based on the Moses illusion, we’d have quite a menagerie! But let’s stick to the facts: Noah’s Ark carried a diverse collection of animals, both clean and unclean. The total number of individuals could vary from 254 (127 pairs) to 889 (127 pairs multiplied by 7), depending on the species.”

The source for this calculation you might ask? I don’t know (user error), and here’s where things fall down.

Having carried out my Copilot query, copied and pasted the text of the answer, I then got interrupted, as commonly happens in our current work environments. On returning to my screen a few hours later, my query and the Copilot response history was gone. I still had the text of the answer because I thought to copy it, but apart from the bare facts of the Moses Illusion and the story of Noah and the Ark, I was unable to replicate the response I got to my first question.

Copilot for the web only operates in real time and does not store your queries. While I suppose this is better for security, this isn’t great for referring back. To my chagrin, I also lost the linked sources when I pasted my answer into Notepad, but I can confidently say that you can paste results into Word and carry over references. This ability to check back through sources is especially important when you are asking questions about contested world views, so I am particularly sore about losing the results of my first query.

The lack of consistency of answers is clearly due to the system pulling from different sources on the web on each occasion. What I did find was that on trying to replicate my first result, I consistently got answers that pulled from various Creationist websites, which shouldn’t be surprising because it’s not really a question the scientific community is worried about.

Which AI chatbot works best

Overall, both AI systems handled the Moses Illusion even better than I expected.

Comparing the tools has been useful to understand how they each deal with sources used to answer questions, but also seems to highlight some inherent bias. Perhaps ChatGPT is more equivocal because its sources are less transparent, but it does seem to gently be telling us the Bible is not to be taken literally.

Does Copilot have a fundamentalist Christian bias? I’d like to give Copilot the benefit of the doubt and say that it is egalitarian in its presentation of answers. But the fact that the tool immediately picked up on the Moses Illusion and explained it to me, but didn’t provide commentary of theological and scientific views about the Ark, makes me hesitant to grant this easily.

While both tools do the same thing, I’m learning that they do them just differently enough to not assume they are interchangeable.

Our team has been testing Copilot for Microsoft 365 and learning more about the power of generative AI.

Take a moment to check out their initial observations and intriguing findings. 

Microsoft 365 Copilot logo

Copilot for Microsoft 365: Game changer or over hyped?

Posted by on in Artificial Intelligence, Microsoft 3651 Comment

Metataxis tests Copilot and the power of generative AI

The world is awash with Microsoft Copilot stories – social posts, webinars, announcements, reports and more – all shouting and sharing about the power of generative AI and the exciting potential this innovative AI technology can apparently offer you and your organisation.

Did you know, there are several versions available in the Copilot product line-up?

There is Copilot and Copilot Pro, both for individuals, as well as Copilot for Microsoft 365 for organisations.

Microsoft Copilot products

Here at Metataxis, we have been taking a keen interest in the development and release of Copilot for Microsoft 365 and the reflections of firms who have been testing it.

Just last month, Microsoft announced that they would be removing the 300-seat purchase minimum for Copilot for Microsoft 365 commercial plans. Perfect timing – as this now presents the ideal opportunity for us to test Copilot for Microsoft 365 for ourselves. We want to see first-hand how this generative AI technology works, understand the opportunities and challenges it presents, so ultimately, we can better support our clients become AI empowered and gain the most from this new era of generative AI.

What is Copilot for Microsoft 365?

Copilot for Microsoft 365 includes Copilot in Word, Excel, PowerPoint, Outlook, OneNote, and Microsoft Teams – combined with your universe of data in the Microsoft Graph. It has enterprise-grade data protection which means it inherits your existing Microsoft 365 security, privacy, identity, and compliance policies.

copilot family

It also includes Copilot Studio, allowing organisations to customise Copilot for Microsoft 365 and build their own custom copilots and plugins – as well as manage and secure their customisations and standalone copilots with the right access, data, user controls, and analytics.

How to purchase Copilot for Microsoft 365

Of course, before we can begin to test the value of Copilot for Microsoft 365 – we need to buy it! There is lots of information widely available on how to purchase the product and the process was relatively simple. We just went to the M365 Admin Marketplace and selected our licencing options.

Here, we noted that there were two subscription options available, for either 1 or 3 years. Typically, Microsoft subscriptions are charged on a monthly basis as opposed to annually. We assume the introduction of an annual subscription specifically for Copilot may be due to:

1) it takes time for users to understand and leverage its potential
2) it takes time for Copilot to build the connections and effectively, the “intelligence,” to yield useful results.

We therefore assume it will get better the more you use it. Watch this space …

How to install Copilot for Microsoft 365

We found the installation process to be relatively straightforward, although not necessarily obvious without doing a little digging and initial research. However, there is lots of information available online if you search for it.

New Outlook logo

We quickly realised that you do need to have installed the “New Outlook” in order to use Copilot for Microsoft 365 for emails. So, if you are not comfortable with the new Outlook or ready to switch just yet, this could present a bit of a challenge.

Testing Copilot – Game changer or over hyped?

We are looking forward to understanding first-hand all the opportunities presented by Copilot, as well as identifying any risks associated with the implementation of Copilot into an organisation.

It will be invaluable to learn more about how it actually works, in particular how it works with metadata in SharePoint. Microsoft is shouting about how much Copilot will drive productivity for businesses, so it will be interesting to ascertain if this is actually the case.

Microsoft claim “Copilot for Microsoft 365 can help reduce this daily grind, giving business owners valuable time back to focus on what matters most: growing their business.”

As part of the testing process, we’re looking forward to discover if Copilot for Microsoft 365 will make M365 search much better AND if it can actually remove the need to perform these so-called “boring routine actions” that Microsoft claim it will.

Observations to date

We will share our observations as we test over the next few weeks. Initial testing undertaken to date has however, revealed some intriguing results. 

Copilot in action

We asked Copilot to produce a status update of our activity in relation to a specific entity: The output resulting from a single prompt was a well written document in terms of tone, language and layout. The professional presentation and expeditiousness with which Copilot generated the content was highly impressive. However, a significant number of facts were incorrect. The sheer exhilaration of having the content readily available in a matter of seconds could, for some, be enough to overshadow any inaccuracies within the content itself. One to watch.

A further limitation we encountered while testing Copilot was, in this specific case, was its inability to disclose the sources of its research. As a result, the only way to verify the accuracy of the information presented was to cross-check it against our pre-existing knowledge of the subject matter. This underscored the crucial role of information literacy, emphasising the imperative for employees to engage in critical thinking when evaluating the outputs of AI.

We will continue to familiarise ourselves with Copilot’s capabilities and put Copilot through its paces to establish whether it really is ready – and if we are really ready – to create an AI-powered world.

Microsoft Copilot – why Information Governance is more essential than ever

Posted by on in Artificial Intelligence, Information Governance, Microsoft 3651 Comment

Embracing innovative technologies like Microsoft 365 Copilot and Generative AI marks a significant leap in how we interact with data.  These AI-driven advancements hold immense potential, reshaping workflows and enhancing productivity. However, amid this rapid technological evolution, the imperative for stringent information governance becomes apparent.

What is Microsoft 365 Copilot?

Copilot is one of Microsoft’s AI assistant tools that combines Large Language Models (LLMs) with organisations’ content in the Microsoft Graph and the Microsoft 365 apps to increase users’ productivity. 

Consistent themes are emerging from organisations who are testing Copilot as well as software development houses, who are designing solutions to enhance its potential. 

This blog aims to explore why information governance holds crucial significance in today’s technological landscape, especially in light of recent advancements like Microsoft Copilot and Generative AI. We will also highlight some of the potential information risks.

Are we ready for Copilot?

We recently attended an event hosted by the Information and Records Management Society (IRMS) which focused on Copilot’s readiness for meeting organisations’ requirements, including privacy and security. Vivek Bhatt, CTO at Infotechtion, shared insights from his team’s own experience of testing Copilot in M365, as well as collaborating with other companies.  At a separate event, we heard from the Product Manager and Engineering Team Lead at Syskit, a software company for Microsoft 365 management and governance, who discussed ways to secure and govern Copilot for Microsoft 365.

Challenges and opportunities

The adoption and utilisation of Copilot can amplify prevailing information and governance challenges, emphasising the importance for organisations to proactively manage their governance risks.

Beyond the actors that traditionally play a part in information governance (e.g., Information Management, IT, Security, Legal), embracing products such as Microsoft Copilot presents an opportunity to engage users in security and governance, where historically these protocols have often been perceived as barriers. With effective change management, users are more likely to buy-in to the notion of managing information so that they can leverage the benefits of Generative AI.

Key information risks

The majority of information risks relating to the adoption of Microsoft Copilot fall broadly into two categories: security risk and data quality.

User requests can be generated from a number of different applications, e.g., Word, Excel, SharePoint, and Teams, then Copilot will use data from the organisations’ Microsoft tenant (set of services assigned to your organisation) to enrich the request before it goes to an LLM, and then via Microsoft Graph before returning a response. LLMs reside beyond the tenant boundary so do not have access to data within the boundary but do receive some data. This means that without sufficient controls in place there is a risk of sensitive information leaving your tenant.

Continuing the theme of security, Copilot can access sharing links, so where users have chosen to share a link with everyone in an organisation, as opposed to restricting access to an individual or group, Copilot could surface material which was not previously visible to someone.  Lack of data protection controls, including sensitivity labels, can exacerbate this issue. Similarly, public Teams sites should also be taken into consideration as Copilot can access those too.

Low quality input yields low quality output, so organisations need to ensure their data is current, accurate, relevant, complete, and void of ROT (Redundant, Outdated, or Trivial content.) Implementing lifecycle management, including sensitivity labels and retention rules, can help mitigate risks around data quality and ensure that Copilot returns meaningful results.

Generative AI is helping us create more content, but without sufficient controls and a governance framework in place, it will become increasingly challenging to manage.

Echoing these sentiments, Lisa Heneghan, KPMG’s Global Chief Digital Officer, noted that it “isn’t just about using technology, it is also about the policies you put in place to encourage ways of working.” Lisa also highlights the importance of having a clear and curated information architecture to ensure the integrity of the data, and setting appropriate permissions which is imperative for creating a secure environment. KPMG was among the limited number of global organisations granted early access to Microsoft 365 Copilot, during which time 300 professionals from selected functions across the organisation tested the technology.

A new frontier in data-driven innovation

The advent of Microsoft 365 Copilot and Generative AI presents a new frontier in data-driven innovation. However, the true key to unlocking their potential lies not just in their capabilities but in the implementation of robust information governance practice. As echoed by industry experts and affirmed through recent experiences, embracing information governance in not merely a choice but a necessity in harnessing the true power of these technological advancements.  It’s in this strategic alignment between cutting-edge technology and meticulous governance that organisations will thrive in the AI-driven era.

Find out more

Metataxis helps organisations address their information governance challenges, particularly in M365, which will prepare our clients well to embrace Copilot. We have extensive experience in developing information governance frameworks, access models and devising strategies how M365 functionality can be best used. To learn more, simply get in touch!

CollabDays – A brave new world with Microsoft 365

Posted by on in Artificial Intelligence, Microsoft 365

Fantastic community day for Microsoft 365

Siobhan King, Senior Consultant here at Metataxis, shares her experience of attending the most recent UK CollabDays event, that took place at the end of September at the National Museum of Computing. Read on to discover what she learnt at this exclusive event:

The National Museum of Computing in Bletchley Park is perhaps one of the strangest and yet apt place to hold a community day for Microsoft 365. The irony of Enterprise Architect Julian Charles demonstrating the mechanisms behind having a conversation in the Tunny room was not lost on those in attendance. The figure of Alan Turing loomed large as Charles took us through the sematic garden of forking paths necessary to discuss with a machine the care of his plants back in Canada. Chatbots with well-designed semantic models could almost have you thinking you were chatting to a real person.

This was just one of the presentations on the promise of the brave new world of AI which was a favourite topic of the day. There were dedicated sessions to the use of tools like CoPilot, and every session seemed to at least mention the potential for using AI in conjunction with common Microsoft 365 tools and applications. This is why it was heartening to hear that so many speakers made very good points about the importance of strategy (Alex Pearce), governance and compliance (Nikki Chapple and Simon Hudson) and having a Common Data Model (Julian Charles).

Much of this may sound a little speculative but it was all very worthwhile food for thought and there was also much good practical advice to take away from the day. Some morsels I picked up were:

Compliance and governance

Simon Hudson and Nikki Chapple talked about the practicalities of getting your organisation to a basis level of governance compliance. Both took a very realistic and realistic approach to considering what an organisation can actually achieve when you consider that some packages like Purview require literally thousands of actions to meet full compliance. They also spoke sensibly about the Microsoft 365 Governance Risk and Compliance Maturity Model again with a clear focus on what is necessary and achievable. 

Viva

At another session on at the same time was presented by Anouck Fierens who provided some very practical guidance on using Viva Topics to build a SharePoint knowledge base. I was unable to attend but many of the presentations have been shared and I did have a good read through the slide deck :”Build your SharePoint knowledge base more efficiently with Viva Topics” during lunch!

Lifecycle management with Purview and Power Automate

Peter Baddeley managed to give a comprehensive view of lifecycle management, a challenge considering how much there is to say on the topic. 

For me, it was really interesting to see how tools like Power Automate can be used together with Purview to manage a record during the whole of its lifecycle. An ongoing frustration of mine is that there are many gaps where Purview doesn’t quite address the requirement. Baddeley very aptly showed how Power Automate can be used to carry Purview over the line so to speak. With recent developments of Retention Labels triggering Power Automate flows, I’m beginning to think this is the next skill that records managers will need to get their heads around in Microsoft 365.

Rehabilitating Microsoft 365

I really enjoyed the frank presentation given by Edyta Gorzon on adoption, particularly, what to do when adoption fails.

We all know that it happens, and we’ve all been there. Gorzon took us through the steps involved in rehabilitating Microsoft 365 with users where adoption has failed. 

Collabdays

She provided a very pragmatic approach to re-adopting in an environment where users dissatisfied or downright hate the product! This was one of the best group discussions with many in the audience also chipping in their approached to re-engaging with unhappy users..

A jam packed agenda

Collabdays put together an extremely packed agenda covering a list of wide ranging topics. If I had any niggles about the day, it was that there were too many sessions to choose from and I couldn’t get to more of them. Did it stretch me? Absolutely. Some of the topics could be really technical but every single presenter made their topic accessible despite this.

If you’re thinking about going to a Collabdays event, I would thoroughly recommend going.