Fundamentals of Generative AI and Copilot

0:01My name is Simona, that's me here. And this is the first skill in my course designed to accompany the Microsoft AB730 exam for the AI Business Professional Certification. Now, the focus for this course will be on using Microsoft's generative AI-powered productivity tools, that's Microsoft 365 Copilot, to improve your day-to-day work. You know, to speed up creating documents and analysing business content and whatnot. More on all of that lot later. But wait, if we describe Copilot as a generative AI-powered tool, well, that's the first thing I'd like us to tackle.

0:37What do we mean by the term generative AI? What is generative AI? And you can see just a single sentence here to put it into a nutshell. So, we can describe generative AI as a subdomain or a branch of artificial intelligence, which is designed to produce or to generate brand new content, such as text or images or even video and audio, from just a simple description. So, no programming required.

1:08Well, by that I mean, there's no programming required by us, the end user. But of course, there's an awful lot going on behind the scenes. Because in fact, under the hood, actually, we are programming when we use these tools. It's just that we're programming in plain language, you know, natural language, as if you were talking to a human. And so, we'll pick up on that idea again another time when we learn how and why we need to put some thought into the simple description that we do type into these tools rather than just typing a random thought and hoping for the best.

1:38Oh, and I should say that this simple description would be known as a prompt. But as I said, more on all of that another time. Now, for our purposes for this course, this definition really is all I need you to worry about when it comes to understanding what generative AI is. And while it falls woefully short of painting anything like the complete picture of generative AI, I'd like you to be confident that yes, generative AI generates new content, marvellous, but also that it's just a small part of the wider field of AI.

2:13So, it's just a branch or a subdomain of the overall field of artificial intelligence AI. And in fact, let me show you this slide here, which I stole from my esteemed colleague, Jonathan Barrios, who has created a lot of AI training here at CBT Nuggets. And he is quite possibly one of the most interesting people you could ever hope to meet. But here you can see that he says, if we think of AI as a box of goodies, then yes, generative AI is in the box, but it is not the whole box. So AI, artificial intelligence, is the broad field, including, for example, machine learning and deep learning.

2:48And yet here we go, generative AI. And I'm guessing that you have already experienced, perhaps to some degree, using a generative AI powered tool. For example, for many of us, well, certainly for me, it was the chat GPT chat bot that was my first experience of seeing generative AI in action. And of course, since then, we've seen others pop up with Gemini being Google's generative AI tool and Copilot, of course, being Microsoft's. And there are others too. Now, I'm not going to step on Jonathan's toes by digging any deeper into the technical side, but there is just a term you might hear in the context of AI and generative AI, and that's LLM, which stands for large language model.

3:30And LLMs are key players in generative AI because they provide the foundation for your generative AI tool, whichever one you're using, to be able to understand your natural language requests rather than programming. But anyway, under the video in this skill, I have given you a link to one of Jonathan's AI courses, the first skill of which does a really fantastic job of exploring AI and its different types and therefore how generative AI fits into the bigger picture. And I urge you to take a look if you would like to.

4:01It's not super technical. It's designed for end users and can provide some really interesting insight. But that is optional extra homework for you. You do not need that background for the AB730 exam. So I'll leave that up to you. But if you really do want to get more technical, then Jonathan has got loads of other AI courses as well. And I urge you to take a look if that is the direction that you want to go in. And finally, in this short introductory video, in order to set the scene for us, as business users making use of generative AI tools, well, why should we care what's in it for us?

4:35I mean, if you're as old as me, you might be thinking, well, I've managed this long without it. Why do I need to bother? Or maybe there's a feeling of kind of not wanting to tempt fate and start using it for fear that it'll end up doing your job better than you can and you find yourself out of a job. And while I do understand that concern, I think we need to appreciate that it can help you to do your job better. It can make life easier for you. And very quickly, without really knowing very much about AI at all, you'll start finding it fantastically helpful for, for example, summarizing long or complex or badly written content.

5:10I mean, who's got the time to read, I don't know, a 96 page legal brief? Just give me the highlights, please. Then I can decide if I want to spend the time to read all 96 pages. So summarizing content is a fantastic application of generative AI, which anybody can use right off the bat. And even if you prefer to listen to a summary rather than read a summary, you can get it to do that as well. And it's not just summaries for long and complicated documents, but also, as you can see here, for badly written content.

5:41Have you ever received a long, rambling email from someone? Or even worse, imagine that that long, rambling email has had several long, rambling replies and you find yourself looking at it and thinking, huh, is there any action for me here? What is it I should be taking away from this email thread? Do I really have to untangle this great long collection of emails in order to figure out the key takeaways? And this next one, suggesting innovative ideas. Well, if you need to create a presentation or write a report, it can sometimes be really hard to get started. It's quite daunting, isn't it? Sometimes looking at a blank piece of paper or staring at a blank screen, I suppose I should say.

6:16So to get your generative AI-powered tool to draft something for you can really help get you started. And actually, I can remember my very first experience of using ChatGPT when I was creating one of my project management certification courses, because when we put together certification courses here at CBT Nuggets, obviously we scour every detail on the exam content outline provided by the vendor. And I can remember looking at what was it? Yeah, this bullet point here. I'm thinking, really? How is a real time survey or a poll? How is that a meeting tool?

6:51I mean, my mind just went blank. I was exhausted and drained from having to interpret every single bullet point on this very long document. And yeah, you guessed it. I asked ChatGPT how real time surveys or polls could be useful meeting tools. And it gave me some really interesting ideas. So it got me over that block. So, yeah, at the simplest level, without knowing much about how AI works, we can use these tools to do our jobs better by ultimately streamlining the boring stuff like drafting emails, generating reports, automating mundane tasks. And that's got to be a win.

7:27Right. So to wrap up this introductory video then with a quick look at what we'll cover in this course. So the rest of this skill will be picking up on a bit more background information about the Copilot product itself, using AI responsibly and how security and privacy is managed, that kind of thing. And then in the following skills, we are going to be actually learning how to use Copilot in your day to day Microsoft Office experience. So you can see, yeah, best practices for managing prompts, drafting and refining business content, analysing and visualising data, obviously in Excel, using Copilot in Teams and Outlook and even creating Copilot agents.

8:06And the ultimate goal of this course is twofold. So, yes, it will help you prepare for the AB730 AI Business Professional exam, and it will also help you as an end user see how to use Copilot in your Microsoft Office applications to help you with your daily work. Does that sound good to you? Let's get started.

0:01If somebody were to say to you, oh yeah, I'm using Copilot, well that could mean lots of different things because as per the title of this section, there is quite a variety of different Copilot experiences. And so I think it's best to think of the term Copilot as the generic name for Microsoft's AI-powered functionality but exactly how and where you see that functionality and what exactly it's able to do for you, well that will vary. If you use Windows and Microsoft's Edge browser, then it is kind of hard to avoid Copilot.

0:34You really can't escape it because it's built into pretty much everything that Microsoft does. For example, here in Edge, you can see the Copilot logo just here at the end of the search box. You can also see the Copilot logo next to this chat button up here at the top right-hand corner. And look, you can also see that I apparently have the preview for something called Copilot Discover switched on. And when I researched this, because this was new for me the other day, basically it appears that Copilot Discover controls what is shown just here with a personalised experience driven by, you guessed it, AI.

1:07But anyway, let's just do a couple of standard web searches so we can see how anybody can start using Copilot without even really realising. So I'm just going to start off with a straightforward web search. I don't want to use AI just yet. And the example that I'm going to do, and to save me typing, I will just pick it from the history here. So, can I retire at 50 with 500,000? So I do that. And sure enough, as you would expect from a standard web search, no matter what search engine you use, as I scroll down, I'm getting various links here which I could then use to learn about that topic and discover what I need to discover.

1:42But you'll also see I do appear to have an AI-generated summary up here. So I can expand this and it's giving me quite an interesting summary here that seems to have links to different sources, which is quite nice. But for me, at first glance, I'm thinking, oh, well, hang on a minute. It's talking about dollars here. So I'm in the UK and we don't use dollars. We use British pounds. So I'm already thinking, you know, how relevant is that? So I think you do get better AI-driven summaries if, look at the top here, you choose the Copilot search button just here.

2:14You can see there's an outline of the same Copilot logo. So if I choose that, then, you know, it's put in that same search term there, but this time kind of as part of a chat. And I'm getting, again, a summary. But, oh, look at this. It's now showing me UK pounds and it's giving me much more specific information about retiring in the UK. So it's telling me, for example, there we go, that in the UK, private pensions can't be accessed until you're 55. And it's going to tell me at some point when I get the state pension here in the UK. So that actually gives me a really good summary of whether or not one could retire at the age of 50 with 500,000 pounds.

2:50And now I'm engaged in this Copilot chat experience. Look down the bottom. It's inviting me to ask a follow up. So I might type in something like, would it be better if I carried on working for another five years? So I just press enter and I'm carrying on the conversation. And you can see that it's going to continue that same chat on that same topic. And it says to determine whether continuing to work for five more years is better than retiring at 50 with 500K, then blah, blah, blah.

3:23So, yeah, that is a great generative AI experience without really knowing anything about how Copilot works or how generative AI is being used to create this very comprehensive looking answer to my question. But I have to say my preferred way of doing this type of task in the Edge browser. Let me just give myself another tab here would be to use the Copilot chat. So there's the button at the top right hand corner. And if you click that or use the keyboard shortcut, let me just zoom in so you can see that there if you're interested. So control shift and then a comma.

3:54Well, if you click that, it opens up this chat panel at the right hand side, which as we will see when we move on to the next video in this skill. This becomes quite a consistent thing throughout all of the Copilot experiences. But anyway, the point is, it's opened up this chat panel at the right hand side. So I've still got the main part of the screen available. So down here. Well, let's just put in that same search again. So can I retire at 50 with 500K? So just the same search term again.

4:24Let's just hit enter. And what you will see is that, yes, I'm going to get some very similar results, but this time they are confined to the panel at the right hand side. So as I say, all very, very similar. So as I scroll down and what I particularly like about doing it this way. So it's still it's still generating is look when you see these sources, which we did also see when we did the same sort of search in the main body of the screen. But what I love about this is if you if you're reading this, you think, oh, that's an interesting source. You can click the source and it will display that in the main body of the screen.

4:56So I love this. It means you can browse through what it's generated for you. And then when you need to click the source to actually go and visit that particular website or whatever it is that it was using to create this content. So that does work really nicely. So, as I said, you will see the Copilot icon popping up everywhere. And indeed, on a Windows computer, you might also have Copilot down here on the taskbar. It'll add itself down there by default. You may or may not have that in your business environment, but certainly on a home device, you almost certainly have it there.

5:26So if you click that, then again, it's going to be the same sort of thing, but embedded into a Windows app. I suppose it gives you a more immersive experience. So let me just maximize this. And well, I suppose we could type in the same sort of thing again. So can I retire at 50 with 500k? Press enter. Again, I'm going to get the same sort of results. And what you will notice this time, I mean, the actual end result is going to be very much the same in terms of the content it's going to present for me. But I get a few interesting features. For example, look at the top right.

5:57I've got an invite button, which would mean I could invite other people to join in this chat, which could be quite nice. And also the next button along from that invite button, if you just click that, open Quick View, that will give you this floating panel, which I do quite like when I'm perhaps using Word or Excel alongside a copilot chat like this. So I think that's quite a tidy way of working. But let's just pop it out again and maximize it. And let's just do another example just to prove another point.

6:28So give me an image of a tricolour Shetland sheepdog wearing sunglasses. Now, throughout this course, you might hear my dog in the background and she is a tricolour Shetland sheepdog. So what you'll see here with copilot, either in the app like I'm doing here or in the Edge browser, you can get it to generate pictures for you. So we'll see what it's going to come up with for this particular example. It might take a moment or two to produce it, so I'll probably speed up the recording here when I edit the video.

7:00But yeah, that really does look very much like my dog wearing my sunglasses. Fantastic. Now, throughout the rest of this course, we're not going to be saying too much more about the functionality we have just tried in this video, in the Windows app and in the Edge browser, because I just wanted to clarify that, yeah, copilot in some form or other will be present wherever you find yourself in the Microsoft world. But in the next video, we're going to start to look at how the copilot experience differs if you're a business user.

7:30And that's going to be of more interest for us as we prepare for the AI business professional exam, which, of course, is the focus for this course.

0:00Let's move on now to how CoPilot might look in a business environment. And to set the scene for that, here in the Edge browser, I'm going to click the CoPilot icon that you can see just here in the search box. So if I click that, after a second or two, there we go, it says, oh, hang on a minute, which CoPilot experience are you looking for? Because, and this is quite a key point I would like you to retain, your CoPilot experience will be different depending on whether you are logged in with a work account or a personal account. So what we looked at in the last video, without any real thought as to what I'm logged in as, that's the type of functionality that you can easily get without needing to worry about having to pay any extra for a work account.

0:41So in this video, as I said, we'll concentrate more on some of the things to look out for if you're logged in with a work account. And just notice this, it says here, a secure and compliant CoPilot integrated with your enterprise account. So I'm going to be testing you on that point in the next video. So stand by for this idea of secure and compliant. And you can also see that on my computer, it has picked up that my CBT Nuggets, Microsoft 365 account might be the way to go. But throughout this course, I am not going to be using my live CBT Nuggets email address, because I have set up an alternative demo environment called Green Beans UK.

1:20And I've set up two users in the Green Beans UK environment, which I'll be using instead. So let me just get rid of this. And what we're going to be seeing is that in my Chrome browser here, in dark mode, because I'm going to be using dark mode for the first user we want to talk about, and that's Nora Nugget. So you can see here, because we're in dark mode, I'm playing the role of Nora Nugget in my demo environment. So as Nora, if she goes to her work, Microsoft 365 homepage, so she's just got a link here, then what you'll notice when that loads for Nora is at the heart of the Microsoft 365 homepage, well, at the heart of that is a co-pilot chat.

2:02And that's quite a change compared to the old days when this homepage here would have had icons leading to all of the different apps, you know, the apps that we now see down here, you know, whether it's Outlook or SharePoint or whatever, but no, now we have chat at the heart of the business Microsoft 365 experience. But the apps are still there. And in fact, if I do just quickly show you Outlook, because you'll need to know this for the example I'm going to do in a second. So we're logged in as Nora here. So in Nora's calendar, we can see that Nora has got quite a few appointments this week.

2:33And also, we can also see if I go into SharePoint, we can see that Nora has got access to the marketing SharePoint, on which there are several useful marketing documents, including this brand guidelines document. Okay, so hold that thought for the examples that we're about to do. So let's imagine as Nora, she hasn't really had any training, but she thinks, well, you know, I can just experiment here. So she thinks to herself, well, actually, yes, maybe I could use co-pilot to help me with the task I've got in hand today, which is to produce 500 words on recent trends in online learning.

3:08Now, I'm cheating by picking this from the history rather than typing it, because I did test this before. But you know, she would just type in some sort of prompt there. Notice, however, there is a dictation button. So you could absolutely use that in your own environment if you wanted to, but I'm not going to because it does interfere with my recording software. But yeah, let's as Nora, just pop in that prompt and see what it comes up with. So as you're starting to expect co-pilot as a generative AI powered tool, it's going to generate some stuff for me. You can see there are some sources there, like we saw when we were just trying this in the Edge browser as well.

3:39So, you know, that's working pretty well. As Nora, though, look at the top right hand corner, and there are a couple of things to point out. So can you see this green shield here? That means that this conversation is secure and compliant. And that relates to what I pointed out to a second ago when I showed you this. There we go, a secure and compliant co-pilot in your work environment. So that's what we're seeing in the Nora experience. That green shield is what is telling us that I'm using a work account and it is compliant and secured.

4:13So we'll talk a little bit more about what that means in the next video. But anyway, as Nora, let's just click this button to start a new chat. Now, you know, we're going to be talking more about writing good prompts and how to manage chats and conversations and whatnot later on in this course. But just to get us started as a quick intro in this particular skill, but let's imagine Nora thinks to herself, well, this could be very useful. Perhaps co-pilot can help me plan my week. So she's wondering, have I got a free day this week? Now, we have already seen Nora's calendar. She does have a few appointments in her calendar, and she's too lazy to go and look at it.

4:46She wants co-pilot to do it for her. But look at this. It's saying, I don't have access to your calendar directly. So that didn't really work too well. But hold that thought, because that's something that I want to show you an alternative when we're logged in as me in a second. But anyway, as Nora, she thinks, OK, not to worry. So I'm just going to start another new chat as Nora. And she's going to perhaps be producing some sort of documentation today. And she's wondering, does magenta fit in with our brand guidelines? Fit in with our brand guidelines.

5:19Because we saw that there was a brand guidelines document on the marketing SharePoint. And she's wondering if co-pilot can help her determine whether magenta is the color that she's allowed to use here at Green Beans UK. So again, she's going to ask that question to co-pilot. But again, you can see that co-pilot hasn't really answered her question. It's just given her some fairly useful information about generically what sort of things we might expect to see in a brand guidelines document. And it is suggesting that if we had a copy of our brand guidelines to hand, then perhaps co-pilot could help her further.

5:52But it hasn't actually answered her question. So that's Nora's experience. So Nora, as I said, is logged into Green Beans UK with a business account. And throughout this course, as I said, Nora will be shown in dark mode like you see here. But the other user in my demo environment is me. So you can see here, I'm logged in bottom left hand corner here. I'm logged into Green Beans UK as Simona Millam. And you can see my silly profile picture there. And I'm not using dark mode.

6:22So if you see it in sort of the light mode like this, you know I'm logged in as me. And we will start to see some differences because I have a different type of license. Well, again, like Nora, we can see the green shield there at the top right indicating that because this is a work account, this conversation is secure and private and compliant. But look at this. I have got work mode and web mode. So I've got this switch here, which Nora did not have. So if I just flip back to being Nora, can you see that?

6:55Well, there's nothing up there. If I create a new chat for Nora, there's nothing there. That work and web switch just isn't there for Nora because Nora has got a different sort of license, which I will show you in a second. But anyway, as me, I get to put myself in work mode. And I am going to ask co-pilot, have I got a free day this week? So we already saw that with Nora, it didn't really help her too much with what she actually had going on this week because it said, sorry, I can't access your calendar. But for me, my account has got superpowers.

7:27I'm in work mode, which means that the source for co-pilot's responses will come from my work environment. And we call this term grounded in. So the co-pilot responses are grounded in the work environment as opposed to being grounded in the web, which would be the default for Nora. So great. Summarizing what meetings I've got on my calendar this week. So do I have a free day? So yes, it looks like Friday is completely free. Whatever. Fantastic. And similarly, if I just create another brand new chat, again, staying in work mode there.

8:00If I ask that same question that we tried with Nora, does Magenta fit in with our brand guidelines? Let's just let's just correct the spelling there because I'm a perfectionist like that. But again, let's just send that into co-pilot. And I think you know what's going to happen here. Co-pilot is going to be able to access that brand guidelines document on the SharePoint site because I'm in work mode, which is telling it to use my work data as the source. And sure enough, it's saying, you know, based on the current green beans, organic wines, brand guidelines documents that's been shared internally, Magenta is not part of the approved color palette.

8:38And it's listing the approved colors that are in that document. So it has read that document for me and answered my question. That's fantastic. So this is a co-pilot response grounded in my work environment. It's using my work data. And that is pretty amazing. So you might be wondering, well, what is the difference then between the two users in this green beans demo environment? You know, me and Nora. So I am the super user. I've got the superpowers, as you could see. Now, I do also happen to be an administrator in this demo environment.

9:12So I'm just going to quickly pop into the admin console. Now, an ordinary user would not need to know or care about this. So you, as an end user, you probably won't be going into the admin center like we see here. But I'm just going to pop in so that I can show you in the users part here, in the active users, you can see in my demo environment, those are the two users. But look at the licenses that these two users have got. So Nora has just got a regular Microsoft 365 Enterprise license and the particular flavor is E3.

9:46That's just one of the license types. Whereas me, well, I also have an E3 license plus the Microsoft 365 Copilot license. So that's an add-on license that's been added on to my enterprise account to give me extra superpowers. The most obvious extra superpower of which is that it can actually ground its responses in my work data rather than only using web data like Nora. So Nora does still get a Copilot experience. We saw it, but it's only grounded in the web.

10:18And it does work a little bit differently if you want it to use particular documents. Whereas with the extra Copilot license, you can ground the responses in your work data. So that is fantastic. And this is mirrored in the desktop applications themselves as well. So if I go into Microsoft Word, for example, and just create a brand new blank document and I'm hoping that it's going to remember. So here I am logged in as me in the Green Beans account. So this is me with my superpowers. And you'll see that I've got a Copilot button just here.

10:49That will start the Copilot chat at the right-hand side, which is quite similar, isn't it, to in the Edge browser when I clicked the Copilot button in about this same sort of position. That also started the Copilot chat. But look, because I'm logged in with an enterprise account, you know, business account, I've got the green shield that tells me that this is secure and private. And I've also, because I'm logged in as me being the superpower, I can switch between work mode and web mode. And if I just switch to, I've got some screenshots here just to show you this.

11:20So if I switch to just a few screenshots in a PowerPoint presentation here. So right. So there we go. So that's a screenshot of pretty much what we've just seen. So this is where if you're logged in with an enterprise license with the additional Copilot add-on license, then yes, we've got that work and web mode toggle. And we also know that you've got the green shield because it's an enterprise account. Whereas this next screenshot, this is what it would look like from Nora's point of view. So again, dark mode is your visual clue that yes, we're logged in as Nora.

11:55Nora would still get the Copilot button, which would give her the chat window. She still gets the security and privacy of the enterprise account, but she does not get that toggle to switch between work and web modes there. So that's a key difference. And then just for completion, this is what it would look like if you are logged in with a paid for subscription, but in the personal world. So not in the business world, but this, for example, is a screenshot of my mum. And I think she's got as her profile picture, a picture of her garden strangely.

12:27But this is my mum. And you can see that, yeah, she's got a paid for account. So she too has a Copilot button and she can start a chat, but she does not have that green shield thing that we saw before that was all about security and privacy, because that's particular just to the work accounts. So she doesn't have that. And she certainly doesn't have the work mode and web mode thing that she can switch between. And then finally, if you were logged in with just a free account with, you know, you hadn't paid anything for any Microsoft subscription. So you can see here, this is what it would look like if I'm logged in just with a regular Hotmail account with no paid subscription.

13:03Then there's no Copilot button there at all. But I can't have the chat panel down the side. I mean, maybe I would launch the Copilot app in that case and, you know, have it floating to the side, like I showed you briefly in the last video. But the point I'm making here is the experience you get is different depending on what Microsoft services you have paid for. And I realised there was a lot to take in there. So here's your summary. So yes, a free version of Copilot is available to anybody in Windows or the Edge browser, or even you can go to copilot.microsoft.com to enhance your web searches and to generate content.

13:38So we saw that in action, didn't we, in the previous video. And then, yes, we also saw that in the Microsoft 365 world, you know, the Copilot chat is leading the way on that homepage. And this is where we saw for Nora, the Nora Nugget account in my example. Yes, Nora Nugget does get a Copilot experience, but it is grounded in the web only. Whereas for the other account in my demo environment, which is me, I am the super user.

14:09I've got the add-on licence, the extra Copilot licence, and that will give me work and web modes. And as we will discover, as we progress, it'll give me some other things as well, such as further in-app functionality and also these extra agents. But, oh, my goodness, more about that another time.

0:00AI is undoubtedly an amazing advance with some truly transformative benefits in all kinds of areas of the world that we live in today. But along with those benefits, it brings with it some new challenges and ethical considerations that we should be aware of. Now, there is an awful lot we could say about this topic, but let me just pick up on a couple of examples for you. The first one of which is bias. Now, the problem we've got here is that AI systems often learn from data that may contain inherent biases. So, for example, in some healthcare systems, lower accuracy has been shown for some minority groups due to under-representation in the underlying data for those minority groups. So, if, for example, I don't know, in an AI system screening scan results for abnormalities and the data it's using largely comes from white men, well, you don't need me to tell you that that's going to be a problem because white men are only one type of person. So, anybody who isn't a white man who might present differently for the condition that the system is scanning for, well, they're not going to get diagnosed as reliably as a white man would. And the trouble is AI can perpetuate and even amplify those biases. So, the AI system needs to be really sure to use good data,

1:17good processing and appropriate algorithms in order for any underlying bias to not, you know, dominate. So, it's all about making sure the AI system is appropriately designed in order to be aware of any potential bias. But next, privacy, the point being here that the use of large data sets potentially containing personal information will raise concerns about data privacy and security. Again, probably no surprise that that's going to be a concern given that AI can only do what it does by using large data sets. And also, in these kinds of discussions about challenges with AI, you'll sometimes see the term explainability. Now, is that a made-up word?

1:57But the point here with explainability is that when AI creates stuff or makes decisions or whatever, it's somewhat disconcerting for humans to understand how did the AI system come to that decision? You know, where did it get that idea from? You know, is its decision accurate? You know, is what it's produced indeed true? So, there is a growing demand for explainable AI so that we can see the thought process used by the system, which helps us to monitor therefore and understand the AI output and have more trust in the AI tool that we're using.

2:32And so, for our purposes in learning about using Copilot for our day-to-day work, it is useful to be aware of the commitment that Microsoft has made to designing AI technologies responsibly, bearing in mind these challenges and ethical considerations. And they have distilled their commitment to responsible AI into these six principles shown here. So, you can see they say, responsible AI is built on foundational principles that guide the ethical development and deployment of AI systems. Now, I have given you a link in the skill to Microsoft's responsible AI site. So, by all means, take a look at that if you're interested. You do not need the detail.

3:10I'm just going to give you the highlights now. So, I think fairness and inclusiveness probably go together. With fairness, meaning that when designing AI systems, Microsoft will take steps to ensure that the AI systems don't discriminate and, I quote, they'll ensure that they do provide equitable outcomes for all users. And inclusiveness is meaning that the AI systems must be designed to avoid perpetuating the bias that we talked about on the last slide. So, addressing these challenges is going to require the way that Copilot works to be using, you know, good data, good processing and, you know, good algorithm design. Accountability and transparency, I think, also go together.

3:53So, accountability means holding the developers and the organizations responsible for the impacts of the AI systems they create. And transparency means making the AI processes understandable and accessible to users. So, that's addressing the need for explainability that I mentioned on the previous slide. You know, users want to know, well, how did AI come up with that thing that it's just come up with? Reliability and safety is where the AI system design should mitigate risks such as harmful content generation or, you know, exploits to your system that you're working in.

4:26And privacy and security concerns the safeguarding of user data and ensuring compliance with privacy regulations. Now, to get an idea of how Microsoft has implemented these principles in the design of Copilot, let's take a brief look at what is described as the Copilot system. So, Microsoft describes the Copilot system as a sophisticated processing and orchestration engine. And the idea is that the Copilot system harnesses the power of three foundational technologies. So, firstly, your Microsoft 365 apps, obviously. So, that's where you're going to be driving your Copilot experience. Also, the Microsoft Graph. Now, we haven't mentioned this before, but Microsoft Graph is kind of a gateway that provides connections to various Microsoft services. So, it's going to provide access to the data stored across your Microsoft 365 environment, you know, your work data. And then, of course, the other thing that it's using is a large language model. Now, we have mentioned this briefly before, you know, that's the underlying engine providing the capability to produce human readable text accessible through natural language. So, the Copilot system all starts with a prompt from you. So, within your app, you're going to write some sort of prompt that goes into

5:46Copilot. And Copilot is going to pre-process that and uses a term called grounding so that you get relevant results. And we talked a little bit about grounding when we saw the work and web modes in the demo. And one of the most important parts of grounding is when it comes to using your work data rather than just the web data. So, Copilot makes a call to the MS Graph in order to use your work data and then Copilot uses that to send a modified prompt back to the large language model. And then, Copilot takes the response from the large language model and post-processes it with responsible AI checks and security and compliance and privacy reviews and whatnot and then sends the response back to the user there. So, the point of mentioning all of this in this video is that Copilot's pre-processing of your prompt and the post-processing of the response from the LLM is part of Copilot's design which follows Microsoft's principles of responsible AI. So, if we now see how all of this responsible AI design actually manifests itself in the Copilot experience, well, first on the list we've got grounded responses. So, Copilot will base its outputs on reliable sources providing

7:05citations for the transparency that we need to provide. So, you might remember, if I just show you this slide here, that we did have these citations shown here that you can click to actually see where did it get that idea from. Oh, well, you'd click the citation there. I think they're a little bit harder to see in dark mode, but this was Nora's experience and again you can see the citations shown there. So, yeah, that's a way that we can add transparency and build trust in the system. And then content moderation. Well, this is where we're seeing the, you know, the processing that Copilot does of the response including, you know, filters and classifiers to prevent harmful or inappropriate content coming back in the response that the user gets. And then privacy by design. So, user data is protected and is not used for model training.

7:54And when it comes to encryption and data isolation, did you hear my chair creaking just then? So, sorry about that. My chair needs, I think I need to oil it. But anyway, when it comes to encryption and data isolation, Microsoft does commit that data is protected at rest and in transit. So, again, it tackles that concern about, you know, user data privacy and whatnot. Regional compliance is all about ensuring that the data processing is aligned with geographical or regional differences in regulations. For example, GDPR for EU data.

8:27And again, that's going to all be aligned with general Microsoft 365 data processing policies and whatnot. And also, you can see here access controls. This is all about ensuring that any permissions or sensitivity labels that the organization has set up are respected and not overridden, as it were, when Copilot is doing its stuff. But do you remember that green shield that we saw in the demo? Well, that is the indicator that these things here are all working.

8:57So, if I just remind you, that green shield is this bit up there. That's how we know that the enterprise data protection is switched on. And if you want to know more detail about what exactly that means, then there is a link there to the Microsoft documentation. And I provided a link for you in the skill as well. You do not need the detail, though. You just need to know that, yes, it's switched on when you're using Copilot through a work account. But this does lead me on to, well, how does the organization set these access controls for permissions and sensitivity labels and whatnot? And it is something that's worth knowing that these are set by the use of another Microsoft technology called Microsoft Purview.

9:42Now, for you as a business professional, you absolutely do not need to know how to set up permissions and sensitivity labels and whatnot. But it is useful to know that that's the technology that would do that for you. And I've taken this screenshot from one of the admin consoles because I think it summarizes perfectly what Purview can do to help you in your Microsoft Copilot environment. You know, as an IT administrator, as I say, this isn't something that the end user would do. So, for example, yes, sensitivity labels. This is where the organization would be able to label data and therefore control how it's handled and how it's processed. And yes, look at this.

10:20You would go to Microsoft Purview to manage such sensitivity labels. What about retention policies? If you wanted to control or your IT administrators wanted to control the data life cycle by deciding how long those Copilot interactions should be kept and, you know, when they should be deleted and whatnot. Well, you guessed it. Microsoft Purview would enable you to set up those Copilot retention policies. And then you can see here communications compliance. So any other regulatory issues and whatnot. Again, we go to Purview to manage communications compliance. So I don't need you to know the detail. I just need you to know the name of the Microsoft technology that will enable all of this splendid stuff to happen. And also just be aware that this isn't included in all licenses. You need one of the more expensive licenses, for example, the E5 subscription.

11:12And then finally, you might be thinking, well, how can I be sure to use AI safely and ethically? Because this is quite, you know, it's a new thing. And people are, as we saw in the first of the demos in this skill, people are able to use it without even realizing or thinking that hard about it. So it can be quite easy to slip into the world of this magical box of generative AI. So there are some points on the slide here that are worth bearing in mind. Of course, we should avoid over-reliance on these tools. We need to think of these tools as helping us to do our job better, you know, freeing up our time for more value-add opportunities by letting AI tackle the mundane stuff. And being aware of security risks such as prompt injections. I suppose that should be plural, prompt injections there. And the idea with a prompt injection, if that's a term you've not heard of before, it's where malicious prompts might contain instructions to try and trick copilot into ignoring the intended rules designed to keep our conversation safe. Or, you know, these malicious instructions might attempt to trick copilot into performing unintended actions or to accessing private data and whatnot.

12:22So we need to be aware of the possibility that, you know, there is, like any other technology, the potential of a security risk there. And the overall message is always review the output. Don't just blindly let AI create content for you. Yes, you get it to create a draft. Yes, you get it to create something for you to work on. But you must always review the output that AI produces and use feedback mechanisms. So, for example, and again, I think I took a screenshot here. Yeah. For example, you can see here, this is one of the demos we did earlier on in this skill.

12:56At the bottom, you'll see there's the opportunity to give it a thumbs up or a thumbs down. That's how you would send feedback to Microsoft about how good that AI response was. And it's really important that we do use these feedback mechanisms so that Microsoft can improve the copilot experience and always, always, always be on the lookout for incorrect content. It's up to us to be vigilant because a generative AI, just by the very nature of it being generative, it's very good at making up stuff. So you will hear the terms fabrication or hallucination, which basically means that it can create completely fictional content that looks very convincing, very convincing indeed. So we have to be alert to this. We always have to review the output with, you know, with our brains switched on to determine whether what's been produced is useful or not.

0:00This challenge has been cunningly designed to recap on the key points that I really want you to take away from this particular skill. So the first question required you to examine these four screenshots of four different co-pilot experiences within, well it happened to be Microsoft Word, so within an Office app and to identify exactly what licenses the user has. So screenshot A, well you can see the green shield so you know that it's going to be one of the enterprise licenses and because screenshot A has also got the web and work modes that means it's an enterprise license in my case E3 plus the add-on for co-pilot. Whereas screenshot B here it just had the shield with no web and work mode so that was just an enterprise license in my case E3 but it could be E5 so let me just add E5 to that as well just as examples and then screenshot C here well this is the example where we were logged in as my mum and my mum is a home user she does have a paid for subscription so yes she has the co-pilot icon but there's no green shield so she doesn't have that enterprise data protection because she's just a home user and then screenshot D here was just a free hotmail account and there's no co-pilot icon there at all, good.

1:13Okay quiz question number two then required us to match up the two halves of the sentences so Microsoft Graph was the part of the co-pilot system that allows the responses to be grounded in your work data so Microsoft Graph allows co-pilot to access data across your Microsoft 365 services. Now bias that's something that might be present in the underlying data that the AI system is using. A prompt injection there we go is a security risk where malicious instructions try and trick co-pilot and then fabrication describes fictional yet convincing content that's been generated and also here you might see the term hallucination also to mean a similar sort of thing, good. Next question then in order to get the full Microsoft 365 co-pilot experience what does a user need in addition to an enterprise license such as E3 or E5 and I'm hoping that you went straight to this option here it's the add-on license for Microsoft 365 co-pilot and then finally which is one way in which co-pilot fulfills the transparency principle of responsible AI so transparency meaning that the AI tool kind of shows its workings so this is one way by using

2:35reputable sources and then providing those citations so that the user can click those links to those sources and see the original content that the AI tool was using.