IBC Panel Discussion - Practical AI Tips for Clerks and Staff

Hi everyone who's joined so far. We'll just hold a couple of minutes to let some others join. Um, so don't panic that you can't hear us. We're we're just holding for a moment and then we'll get started. If you've just joined, uh, we're just waiting for a couple of minutes for some more people to uh, join in and then we'll get started. So don't panic that you can't hear us at the moment. Uh, probably two minutes and then we'll get going.

Okay, good afternoon everyone. There's 50 odd people on the call. We're expecting some more to join, but they can catch us up when they do. Um, welcome to today's webinar on practical tips for Clarkson Chamber staff on AI. My name is Rowan. I'm the CEO at 36 Group and I am a self-confessed AI and technology nerd. Um, I've been hearing a lot from Clarks and Chamber staff recently that there's so much being said about AI, but there isn't much practical advice out there about what we can actually use it for properly in a Chambers context. So we decided to pull together um this session to try and impart uh some practical tips um some advice around the use of AI in chambers from a staff perspective.

So before we get started, a couple of quick housekeeping things. We are recording the webinar. Um, you shouldn't be on screen. It's just the four presenters on screen. Um, so, uh, if you've got any issues, um, with that, you you you may want to leave now if you've got an issue with the webinar being recorded, but you shouldn't be shown on screen. Uh, please feel free to participate actively. You can post your questions in the Q&A tab. Um, if you can't see the Q&A tab, just post them in the chat. I will try and keep an eye on those during the session. We'll try and address as many as we can at the end of the session. Um, and if we can't, we'll um, we'll see what there is and we'll see if we can come back to those later.

Now, luckily for for you, you don't have to listen to me talk for an hour um as we've got a panel of real experts here who are going to do all of the hard work. Um, so, let me introduce you um to them. So, we'll start with Kaylee. Kaylee Jeff is the lead civil and public law clerk at 30 Parkplace. Kaylee has extensive experience in managing a thriving Clark's room and is particularly passionate about applying innovative tools to make the running of chambers more efficient. Katie heads up the IBC IT committee, is on the bar council IT panel, and is a member of the LPMA IT or team. So a bit of a mouthful that Kaylee, quite a lot on on the CV there. Um, next we've got Anna. Anna Lawrence's product manager from Instant IT. Anna's responsible for delivering private and secure AI tools to chambers and works with many chambers on their individual implementation requirements for AI in their own settings. And last but not least is Alistister. Alistister Huntingford who's the director at indui IT services um with a focus on compliance and security. Alistister provides IT and AI services to chambers um and is on the cutting edge of understanding how chambers can use AI tools effectively to create better results faster. Many of you will recognize Alistister from his presentation at the IBC conference last year. So thank you to the panelists for joining us today.

Um, to start, I'd like to just set a bit of context and perhaps set some minds at rest of people who are listening. AI has become a a broad spectrum tool that people are adopting at different speeds. Um, often depending on their stage on the AI learning curve. Some of us on the call will be early adopters and will be using AI for vast amounts of our day-to-day processes already um while others may never have opened an AI at all um or still be skeptical or cautious about it. All of those experiences are valid here. Um, so please don't panic that everybody is ahead of you. Um, and that you're not going to understand. Our goal today is to make AI feel approachable for everyone by offering practical tips and real life examples showing its potential to improve day-to-day work in a chambers.

So to get us started and to hear from the panel, I wonder if I could just ask a really simple but important question of each of you um about the role of AI in chambers. Are you positive? Are you worried? Or are you somewhere in between? Anna, could we perhaps start with you?

>> I think it's um it's a very exciting opportunity. It's a very exciting and disruptive thing that's happening in technology. Um and it's going to the opportunities are kind of endless but the impact it's having or having and will have um are going is going to be more in the sphere of reducing time spent on tasks and it will also improve time to upskill. Um it's improving the efficacy of research and enabling everyone to learn learn new skills even if it's just how to use AI itself. Um there are some some concerns with any new tech um AI induces introduces a suite of security concerns um which haven't been haven't been cause for concern before um and I think there is a reaction both in chambers and in public life against poor poor use of AI and misuse of AI that hasn't been checked um where where it can be where it has been used to generate false or inaccurate information. But overall, I think it's it is a really powerful tool. So, if you're mindful of the those considerations, um you can really harness it to improve your everyday life or day-to-day life in chambers, I think.

>> Brilliant. Thanks, Anna. Alistister, what what about you? Where do you sit on the on the feeling of AI in Chambers at the moment?

>> Well, firstly, thanks uh to the IBC and for yourself Ran for arranging this as well as to my guest contributors. Um, I've worked with Chambers obviously for a for a number of years now on an on an IT perspective and almost I don't want to say fell into AI but it was a natural consequence of what I do being around security and compliance and so I've heard many varying opinions all of which for me feel equally valid um in the sense that you know there's some concerns around things like skill atrophy um and you know compliance around data what is AI doing with our data? Does it comply or should we be concerned about this? Can we input data that, you know, is considered private or or personal inside these tools? And I think especially recently with I've had a surge of queries around um the new BSP guidance that was released as well. And I think what that is telling me is that there is a lot going on within the AI world. Um but for me personally, I feel like it's something really positive. uh being a personal user of AI um even if you put chambers contextually aside and you look at it just on a global scale I think there are many areas that it can offer offer a level of improvement with that as Anna touched on there are some concerns around for me personally misuse um I have three children and I am slightly worried sometimes about what the future looks like if people embrace AI but they don't consider the ethical ramifications of it um how to work with it what does to do, how to use it, and importantly, some of its pitfalls. But I think if those can be managed appropriately, then things are looking certainly brighter. So overall, I would say I'm positive.

>> Brilliant. Thank you. Um Katie, how do you feel at the moment about AI and chambers?

>> So obviously I'm looking at it from a different perspective to you guys, but I'm more from the compliance side of things. Um, but I'm positive about it, but I'm just really cautious. Um, personally I love it. I use it, but I know that I've got the safeguards in place. There's definitely lots of benefits for the efficiency and speed and things it can be used for in the classroom, but I'm also terrified of the risks because I think there's a serious risk of overtrust. >> Um, the confidentiality side of it worries me. Um, the accuracy worries me. So, I think it is great, but it has to be used with the correct boundaries in place.

And Kylie, I think that brings us on really nicely to talk about some of those safety and compliance aspects. Um, we we need to consider governance. We need to consider training, safety protocols. Um, it's vital to get those pieces of set before we kind of dive into it too much. Could you give us your kind of perspective on why we need to do that and and how we should be thinking about going about that?

>> Yeah. So, with the training side of things, it terrifies me because any training you give is relatively out of date by the time you finish it. Um, because AI is changing so frequently. Um, there's a huge risk in any uncontrolled use. The BSB guidance is quite clear that you need to make sure you've got an AI policy, um, acceptable use rules and data handling procedures in place. Um, from my perspective, we'd made sure that we had security work completed before anyone had any um, AI put in place within our server. Um, I tend to try and make it our team think of it as outsourcing. So, if the security work's not been done, would you give the information to a stranger? If the security work has been done, could you trust a colleague with it and would they be able to identify it? Um, you have to keep in mind, we aren't regulated in the same way as barristers obviously. Um, but our actions directly impact on barristers. So if you mess up then it's their insurance, it's them that are affected, it's chambers reputation that are affected. Um, so it's transferring that risk whether you like it or not. Um, I definitely think you need to be looking at what you're putting in to AI. Are you putting in case details because that creates a huge risk. Are you putting in emails, client names, any anything like that needs to be anonymized. Um, I will only accept any paid for us like versions within chambers because at least we know that's much more secure than non-paid for versions. You know that you've then got to look at the um data sharing agreements because you need to factor into where's your data going in if you get a subject access request. How are you building that into your policy? Um, I know that most people don't seem to keep in mind that it doesn't understand the law. AI doesn't understand, but it does remember what you put into it. Um, I like to use it for rewriting anonymized data. So, if you're doing an email, that's fine. You can use the body of the email and get that rewritten, but don't put any client data in it or any of your colleagues data if you can avoid it. Um, we've had a surge in subject access requests come in that are AI generated. And personally, I will use AI to help draft the response to the subject access request, but only with anonymized data. Um, there's a lot of things that can be used for in that way. You have to remember that it looks very professional, but it can still be completely wrong. We did um just as an example of how wrong it can be. We did a news item and it was to do with a murder in Swansea and we put in the defendant's name. We put in the details. It was all fine. We put in the links to BBC News. So it should have generated a beautiful news item for me that said the barrista exactly what happened. No, it pulled three different murders from Swansea and merged them all together. So, had we not proof read that and just chucked it on our website, it would have looked absolutely appalling. So, you have to remember that even though it looks great, it can still be completely wrong. Um, I'm just trying to think the um oh, the BSB guidance is very very clear that you must review and verify any outputs that come out of um AI just to be on the safe side. You should do it anyway, but you have to, you know, keep in mind the BSB guidelines. Um, the other thing always ask yourself before you use AI, if you were asked where it came from, whatever the output is, and you've used it, can you justify the use of AI? If not, why are you using it? Um, you need to keep an audit log and a prompt history. The BSB guidance is quite clear that you need to have records of how you've been using it, why you've been using it, and exactly what you've put into it, which then feeds into, have you actually included it in your privacy notice? Have you got it in, if your barristers using it, they need to make sure it's in their notice, but actually in your chambers's notice that you've probably got published on your website, have you included that you're using AI? Because if the clerics are using it, you have to. Have you built it into your subject access request process? We have a clear subject access request policy and process and I've had to build it into that because you still need to be requesting any data from the AI companies that you're using when you have a subject access request. So for me it's about handling the client info but making sure it's with accuracy and controlling the risk in any way that you possibly can building into your policies.

So, it's it you you make one um brilliant point early on there, Katie, about the fact that training is always out of date five minutes after it's been delivered. And and I I I completely forgot to put that in my housekeeping notices at the beginning of this talk that that um you know, don't take anything too seriously if you're watching this in a week's time because um you know, who knows what would have happened with AI um by then. Uh >> but a lot of a lot of what you're talking about there, Katie, is about prep work in advance, isn't it? Is getting your policies set up, getting your privacy notice, s process, all of those sorts of things in advance of actually um entering anything into AI, making sure that things are set up properly um and integrated properly so that we can get the best out of them. Um, and then making sure you use it properly once those are all all set up. Um, in terms of let's assume for a moment that some of those kind of prep things have been addressed, that chambers have implemented the policy, have put together a privacy notice that includes AI, etc., etc. We've we've we've done all of that. Um, and now we want to get on and start using um something. Anna, how how do we think about identifying the right tools to use? Because it it strikes me there's thousands of tools out there um that all say they're amazing at kind of managing our lives for for AI and can all kind of do wonderful things, but how do we make sure they're they're the right tool for us and make sure they're integrated properly and in our systems?

>> A really good question. So, I work with teams not just in chambers, but also other types of organizations as well. And I will say that chambers tend to have the best um the best approach to procuring secure tools. I think there's a generally a lot more of an awareness around GDPR, data subject access um and privacy notices um and what what that all means. So there I think chambers in general are on the front foot whether or not you think you know anything about AI you know I think you know a lot more about procuring secure tools um than say than say other types of organizations. Um, one point of in in choosing tools, I think, um, one piece of advice would be to ensure that you're very careful about AI tools that need admin access from your mic to your Microsoft accounts. Um, so things like you can even if you're using a purchase version of AI tools, sometimes they want uh that you'll want a notice to like give it access to your SharePoint and one drive files in those instances. Um, for example, Claude called Co-work, it can have access to your um all of your internal documentation. So you really want to think about what's in what folder and if you want to give can't launch access to uh read, write, delete um data from those folders and who has access to those. Um, so I think once you're once you're happy with the tools that you the tools that you've chosen um or once you're happy with your requirements rather um I we always like to recommend starting with a AI chat tool um similar to the interface that's really popular with the chat that chat GBT made popular rather um so we offer our clients a secure instance a private AI tool um which can be used for summarizing planning and scoping scoping tasks um if you give it the context of I need to achieve this task here are the limitations here's my target and here's other things that like should be considered what's the best way to approach it is a really that that first interaction is a really gentle way to um a gentle way and a to approach AI AI and ensures that you're getting maximum value um from your first interaction there. Um because it often comes up with like a slightly different plan to an approach that you may have considered before. Um but it allows you to um really see what it can do for you and obviously take that with a pinch of salt as well. Um our private AI tool um means it is private because it exists in your secure Microsoft 365 tenency already. So there's none of the additional risk of inputting folders or files into um other AI tools. Um one thing we've come to learn is that even if you think your organization or if your chambers isn't using AI, someone somewhere is using a free chat GBT version. Uh we call it shadow AI usage and it's it is prevalent. Um, so just having a secure version that comes with those rules and comes with the governance um is going to help get you ahead of any of those issues that we we're aware of. Um, I'd also recommend Copilot. I think any most organizations are on Microsoft and there are versions that you can there's a version for free with within your Microsoft plan. Um, so most chambers will be on business a b business or enterprise Microsoft payment tier. Um, and the version that you get for free with those is it is secure and it does um it does adhere to the data location rules um that you have set within your tenency as well. So that ensures that your your files and your and the date and the prompts that are being run, the models that are being run rather that all is secure and stays within whichever data location you've set. So usually the EA or the UK. Um, so that's a a bit of added security from Microsoft or a bit of reassurance from Microsoft. Um, Copilot's useful cuz it integrates with the programs that you're and the apps that you're most most likely using already. It just acts as your assistant to help you get a little bit more out of them. Um, and it can respond to your plain English prompts as well. Um, which is particularly useful, I think, in Excel. Um, if you're like novice or junior junior entry level with Excel, um, and you want something to happen within your spreadsheet and you're not quite sure how to get there, Copilot's really, really good for I want this to happen. Please help me. Um, please help me get there. And it'll help with maybe functions you haven't heard before and that kind of thing, but they do work. Um, so yeah, Copilot and particularly Excel is um is particularly beneficial, I think.

>> Brilliant. Thanks, thanks, Anna. I think there's, as I say, there's so many tools out there, aren't there? And sometimes just a a guide or chatting to other people and saying, "What are you using at the moment?" And um we'll go on to in a minute some some real practical uses in chambers. um some things that we're we're using AI for or that we know other chambers are using AI for. Um, before we get there though, one of my favorite um uh sentences is rubbish in rubbish out. And I think I heard that as a junior cler um when talking about um lex that we were using at the time, our diary software um that if we put the wrong information in um we're going to get bad information out at the end and my senior cler drilled it into me. Rubbish in, rubbish out. And the same applies for for AI. If we if we prompt it badly, if we ask a bad question, we're going to get a bad response. Um, Alistar, I know that you're you're a real um kind of connoisseur of prompt engineering. Are you able to take us through that briefly in terms of how how we should approach that?

>> Yeah, absolutely. And I'll try to avoid going any sort of going onto any sort of wild tangents here. Um, so to start, I think if you understand why prompting is important, I think it will allow you to gain a stronger insight into generative AI itself. Um, as ultimately generative AI is only ever going to be as effective as the instructions and the context that you give it. And I'm sure there a number who have joined the webinar that can attest to that already. Um, but for those unfamiliar, I think starting with the definition of what prompting actually is is probably a good place to start. So Google defined prompting as I'll read it verbatim: the act of providing instructions, cues or encouragement to guide an action or elicit a specific response. And I think if you apply that logic to generative AI and what we know about generative AI in so much as that it uses pattern recognition and prediction to provide useful responses, we can interpret Google's definition of prompting as follows. A vague, weak, or ambiguous prompt when engaging with an AI service will usually produce a vague, weak, or ambiguous result. And I think if you use a detailed and concise prompt, you will likely get something more useful out of the AI um at that time. And an analogy that I like to use, and excuse me, but I'm a bit of a layman, that I use to describe effective prompting is a bit like ordering a sandwich from a restaurant, right? The more specific you are about the sandwich, sorry, the with the toppings, the style, the fillings, the more likely you are to get what you wanted from the sandwich. And I think as an example, picture yourself walking into a into a sandwich shop. Traditionally, you would not walk up to the server at the counter and say, "Make me a sandwich." I think the result of doing something like that would be something you didn't really want. Um, and you might get lucky, which does happen from time to time, but I would rather not leave that to chance. And I think the prompt, make me a sandwich, would then be considered a weak, vague, or ambiguous prompt. Right? You haven't told the server what bread you wanted. You haven't told it what filling you need on it, whether you wanted hot or cold, does it need to be vegetarian, do you have any allergies, what is the size of the sandwich? And I think good prompting will always give direction, context, intent as well as sources. And I think most AI, certainly from my experience at least, most AI users have their own organic promptic style that grows and develops with the use of AI as it's a bit of an art form and quite personal. But the information added will largely follow the same structure when you're using effective prompting. And for me personally, I kind of created my own acronym for this. I use task t a sk um when using prompting and if we apply the logic of task to the sandwich analogy, the t and task stands for logically enough task what you're getting the AI to do i.e. Make me a sandwich. The A stands for aim. What aim am I trying to fulfill? It's late. My legs are tired, so I don't want to walk far. I'm hungry. Maybe I want something saucy, something hot with no veg. Provide if you provide the context or the why element behind the request, you'll usually be more successful. And you you you shouldn't be afraid to use a lot of detail as well. Some of my AI prompts more recently have been up to a page in length. Um, mainly because I use dictation for it. But it's it's really been helpful to me. Um, moving back to the task acronym, if you use the S being the source in task, you should where must the AI look to get the information about your request? Where must it get your sandwich from? Is it going to go to Subway or some other fancy sandwich place on Oxford Street? You need to tell the AI where it needs to go to get the information about the request or in this case obviously the sandwich. And the K in task stands for key expectation. How would you like your sandwich delivered? Do you want it delivered quickly? Do you want it in a paper bag? Do you want napkins with it? Do you need cutlery? This will tell the AI how you want the results delivered to you and how to how the AI will communicate the results to you as the prompter. And hopefully by the end of that using task the end of it will be you know the information is provided um in a in a in a way that is effective for you and more importantly useful to you and this is honestly the tip of the iceberg when it comes to prompting you know you can get pre-made AI prompt sandwiches in the form of agents and templates which you know help you form and produce uh run routine tasks and provide uh quite accurate results against those manual routine tasks. tasks. And what I would encourage you to do, especially if you're just starting out with AI, is to do your own research and educate yourself into prompting styles that make sense to you, but particularly to your data. And I think a good place to start is to actually ask the AI service that you're currently using to provide you with guidance and methods of improving your own prompting style. So yeah, go and go and make some sandwiches, I guess.

>> Brilliant. I'm not going to be able to get the sandwiches analogy out of my head, Alistister, every time I use AI. I I think I think your your point about going and asking AI um is is really important is once you've got all of these other things sorted out around the compliance and you know that you're you're safe in using it and you've got the permission of the relevant people in your organization. The way that the way that I learned um how to prompt was by prompting and by testing a result and changing my prompt and seeing what changed in the result. Um, and I think sometimes that trial and error is is is really important. And I think there's also an important point around uh bias with that as well. And >> if your prompt is saying, I'm feeling really positive about this. Can you tell me what to do about it? It will give you a positive response. Whereas if your prompt says, I'm feeling worried about this. Can you tell me what to do? It will give you a negative response. And um you need to be cautious around how you're um again in that prompt engineering what you're encouraging as as the word that you use Alistister what you're encouraging um the AI to give you. Um, we could do a whole hour on prompt engineering and we may do at some point because um >> I mean just for my benefit because it's fun um but uh probably not fun for anybody else. Um what we really want to talk about though is practical uses and what people can actually do. So let's let's again pretend that uh everything is set up. We've got policies in place. We've spoken to our um compliance people in chambers. Everything is ready to go. We've spoken to it. Um and we've uh implemented properly implemented AI into our um systems. It's been tested. Um we know it's secure. We know it's safe. Um and we know that access is restricted in the right way. Um what do we do now? How do we actually use it? Um, and what are some practical uses that that perhaps uh we might not have thought about before? Kayleie, I wondered if you could um start us off to put you on the spot there as to some practical uses um in chambers for for these to >>

>> Yeah. So for me, I guess it's more about what you can do safely. Um, so the ideal uses for me are around templates, internal communications, um, and policy drafting. Ironically, use it to do your AI policy if you really want to. Um, the template side of things, I've been using it, um, for anyone who does any court protection cases will know that you have to have ridiculous enhancement reasons to be able to get 100% uplift. And I know it crosses onto some other legal aid matters, but court protection is my bane of my existence. Um, and after the CCMS system went down, they became much stricter um, on what they'd accept and we had them refusing on really ridiculous reasons. So, I've been able to put the whole of the um, legal aids I can't remember what they're called, guidance, let's say, um, on enhancement justifications into co-pilot and I've told it to create a template and now I upload just some like I just put in into my prompt some um, basic details about the case. It's completely anonymized, but it's basic details. Bearing in mind, we've had the security work done around this, so I know it's a safe environment. Um, put that in and it will then give me the enhancement reasons to add on. And at the moment, we've got 100% success rate on then getting 100% uplift on every case we've put in. Um, we've had cases rejected previously because they have had the same reasons on each council's claim because anyone who does legal a billing knows how difficult it is to get the information out of council in the first place, let alone trying to get various different council to give enhancement reasons. I can run those enhancement reasons through the system and say, "Give me three variants of this and it will do it for me and it takes 30 seconds instead of 3 weeks chasing." um internal communications. Sometimes I can sound really firm when I'm sending out memos to the team to say don't do this and I can ask AI to make me sound much friendlier. Um, or much firmer if that's needed. Um, I also went through a phase previously for internal communications of if there were issues I asked pilot to draft me tales of wo based around what had gone wrong and what the outcome was. And I think reading it in a funny way can make it sink into your brain a bit quicker on how you should be doing things to avoid problems. Um, which I found really really helpful. Um, the policy drafting um I think saves me hours. Um, I can pull policies from other places that are relevant to what I need and ask it to draft me a policy. Um, there's no client data involved in any of that. So I know that I'm not going to be breach uh breaching anything. I try to avoid anything at all with personal data going into any kind of AI. Just because it can do it, it doesn't mean that you should do it. Um, and I think I just try to encourage everybody to take that approach that just because it can, it doesn't mean you should. Um, just try and do anything that doesn't avoid client data.

>> Brilliant. Thank you. I think the um the point about templates is really uh good as well. And the um the idea that there are probably so many things that we do repeatedly um that you know I'm typing the same email four times a week or five times a week and you know actually is there a way that we could do a a saved reply function that kind of automatically um populates every time that someone sends a subject access request for example that your initial reply kind of fires off automatically. Um I think there's there's a lot in that space, isn't there, that um if we really kind of boil down every task that we complete every week, um there'd be a lot of repetitive things. Anna, could I come to you and uh hear any thoughts from you as the practical tools that we could use?

>> Yeah, absolutely. I think my advice for anyone who is just starting out with AI and not really sure how to approach anything is to have a sit down either with a group of your peers or a sit down with yourself um and see if you can identify frustrations or things that take you a lot of time, those repeatable tasks. So everything that's already been mentioned here, those emails you have to send five times a week, um internal coms is is a really good place to start for all the reasons that Kayle's already mentioned. Um, and like see if you can either write down everything that you do. Why does it take you so long? Why do you have to do it so much? Um, and kind of give that job to your AI tools straight away and say, "This is this is something I I have to send out uh these chambers coms. Um, they currently like here are the things that change. Um, and I like how how can I improve this? How can I improve my time to time to do this?" um and see if you can work with your kind of your treat AI as a bit of an expert in that that regard. It's being used um and most of the models will have been trained to maximize efficiency um and and to improve workflow. So even by just describing it to AI, you'll come up with um with with advice. Um, internal coms and writing styles is a is another is another like I think we've already touched on it, but that's that's one of the most like practical use cases in terms of immediate time time taken back um and time reclaimed from those um sort of changes updates um and improving improving writing styles. Um I've already mentioned um Excel earlier um and upskilling in those areas. Um I've had some successful use cases in terms of diary management as well but obviously in specific cases. So if you're using uh for your outlook if you're using Outlook calendar um particularly for internal events you can use the copilot there as well to improve how your your diary management um and maximizing your efficiency there too. Um with our private AI tool um by we we've talked about giving AI the right context um you can if you're if you're worried about the relevance of an AI response because it's that often happens because the scope of what the AI searches is too is too big. It might be integrated with web tools and then it's searching for the top 20 results and then trying to synthesize the top 20 pages into your into um a practical response as well. Um use if you can use tools um with folders then you are restricting the context to that you provide to the AI tool and therefore your response is going to be much more relevant. Um so where um practically that means uploading your your last five five chambers updates or uploading your um specific work pieces that you want to work on. Um and that will that will improve the relevancy of of the responses that you're getting back to. Um and so hopefully um it it nothing's 100% but it will reduce hallucinations. So that poorly generated AI output um quite significantly because you're restricting it to those um to those folders.

>> Brilliant. Thank you. The um the idea and of of sitting down with yourself and writing down everything that you do in a day and kind of thinking what can AI help me with is exactly where I started with AI and and and kind of thinking this is a repetitive task or something basic that could be taken off my hands and I'm now sitting here thinking maybe uh we could develop an AI tool that could just follow me around for a week and um see everything that I do and then tell me all of the things that AI can take off of my hands. Um, there you go. A free business idea for someone. Um, Alistister, could you uh take us through your thoughts as some practical tips that we can >>

>> Yeah, absolutely. I think and they honestly touch very much on what Anna and Kayie have said as well as yourself. You know, it's you do have to sit down and look at the repetitive low value tasks and figure out where AI can actually save you time. And I think AI is a tricky one sometimes to integrate because often you don't necessarily see what AI can do until you start sitting down and working with it on a day-to-day basis and you you build up that creative energy. And I think even starting out with the simple things can be very very powerful. You know, Anna and Kayie both touched on it the summarizing of meetings and transcripts which you can do in pretty much any modern web service like Zoom or Teams. Reducing meeting notes which are quite time consuming uh such as for practice meetings as an example. The power of analyzing long threads is something that's also for me personally I feel like underutilized a bit um and you can apply that to documentation as well. Uh email searching which was a recent one that someone provided me feedback on. And they said it was actually a lot easier to run a search for something um in one of the clarking mailboxes using copilot versus trying to actually search the mailbox itself. Uh and also once again the drafting of responses especially for tone. Um I personally have been doing a lot of dictating nowadays cuz I actually speak logically enough faster than I type. So uh still very much a one-finger typer but it's it's it helps me speed up that process as well. I can get out a lot more information faster and I certainly think in Clarking rooms that could be quite helpful. But I also think the other area of benefit with starting small really is it helps build confidence and you get used to the how AI tools operate and and understanding as Kayle's been touching on how to work with them in a protected boundary, what you can do with them, what you can't do with them. Um, I think that will also help you build up this this ability to see use cases from your day-to-day day-to-day tools. And actually speaking with a team of uh not Clark specifically in this case, but administration personnel recently, they mentioned that they had quite a burden um on operational reporting. Um they have a very structured process which is highly manual but it's hugely time consuming and and what what came out of that discussion was that there were data was being collected from a number of different locations and I had diaries spreadsheets finance systems and internal documentation and all four of those resources were uh communicated with and pulled together into a central report. You know, this was producing producing health and safety reviews, financial tracking as well as operational summaries for management committees right and I think this is where co-pilot can actually or or any AI tool not any AI tool but you know trusted AI tools as long as you've gone through the due diligence they can come in and do some of this heavy lifting for you and also really importantly by not replacing the individual or the task but outsourcing it as Kaye said and accelerating how quickly information can be brought together and analyzed. You can you can do things like analyzing exported data from diary management systems. That was one comment that came from one of my clients. And you can summarize documents um and data sets, but also as Anna mentioned, spreadsheets. You know, I think AI is underutilized in the in the Excel community. I use it not as much as I could, but I certainly find it really useful. Uh it can also highlight inconsistencies which I think is really important in reporting. Um, and you can also apply consistent language across reports and documents. Um, especially if you have a specific tone and style that you like to use, you can get the AI to adapt that to your your specific method of drafting these reports or or it could be anything really um any sort of data structure. And also once again, one of the my favorite things on a personal level to use it for is to review policies and flag gaps. That's something that's that's that's really important. But the other thing that's also quite important is that you don't have to completely change the systems that you use just to get the a benefit from using AI. I said in most cases the data already exists as what Kaylee mentioned her use case and it's simply a matter of getting the the AI to interact with that data um in a bit of a different way. And this is where you can use I made a note once again after a conversation with another client. Yeah, they they pull quite a bit of data out of the um their case management system um and they use it quite a lot to interrogate what's actually going on within that environment. And I think it was quite helpful to them as well to do that. But I think yeah, that's probably where I'll stop for now. I think that hopefully has provided some insight.

>> You can tell with all of us that we all get a bit excited about it, can't you? And it's um I think we could all go on for a long time kind of talking about um different opportunities with this. I think Allison you mentioned about tone and style and that the more you use these tools if they're implemented well the more you use these tools the more they'll get used to your tone and style and and just like we were saying the more we use them the more we will understand how they work. >> The more we use them the more they will understand how we want them to work as well and how we want them to sound. And that's what I think is really exciting is if we set them up properly. Um, they could eventually they won't sound like AI. You can you know, we can all spot a LinkedIn post that's drafted by chat GPT a mile. Um, but there will come a point where that becomes really difficult because they start sounding like us as people. Um, just to summarize some of that. So um Anna's talked about um using AI for um enhancing Excel. For those of us that use Excel, for those of us who it's our favorite software package available, hands up. Um, then uh you know there's some really exciting things you can do with that and it can really enhance that experience. And Anna also mentioned about summarizing documents and you know we I'm sure many of us receive long bundles of information that um we don't want to be able to pour through or even you know summarizing the BSP guidance on AI usage. Um, uh that could be really helpful. Um, Allison mentioned practice meetings. I think that that's a a real key area that I think is largely untapped at the moment but I'm really excited to see what AI can do with practice meetings moving forward. It's something that many people in this in in this audience will um have to be undertaking for their barristers or if they're on the administration side maybe um doing some backend admin um for those and I think there's a lot that AI could could assist with on that and again I think it's about experimenting it's about talking to some members who are ready to embrace it and saying, you know, are are you happy for me to run a couple of tests with AI to see if um if this works for you um reporting as well mentioned. I think that's that's key. Something that I do every single day is use one of my AI tools to to help with reporting. Kaylee mentioned about internal coms. Um again, for those of us that have to report to boards or have to report to management committees or anything like that, um using AI to help us with some of those reports where if you can give it all of the context, you can give it the background. It can save hours and hours and hours and policy is a huge thing. Um, we've we I pride myself on the fact that we've got um uh 100 policies or plus in in chambers and um uh and thanks to AI um I shouldn't say that because my compliance administrator I think is on the call but um thanks to my compliance administrator assisted by AI we've been able to to really

enhance our uh our policy offering. Um, there's there's other things as well. There's um, if any of you have ever used Google LM, uh, that's one of my favorite tools if you're talking about non-confidential information. But things like the BSB guidance, um, that came through and I was about to get on the train for an hour. So I just stuck it into Google LM and said, um, uh, can you make this a podcast for me? And minutes later, I had a 30-minute podcast of two people chatting through the guidance from the BSB. Um, so while I'm on the train, I don't have to look at my phone, I can just listen. Um, and it and it it's not perfect, but it tells me kind of large amounts of that guidance, which is brilliant. So, do go and check that out.

And something that I really love um using AI for is my task manager. Um, those of you that know me know that I'm kind of driven by efficiency, wanting to always find those uh, those extra five minutes somewhere. Um, and I manage my tasks religiously and kind of spend more time putting tasks into order than I actually do doing the tasks half the time. Um, but now AI can do all of that for me and it knows my most efficient points of the day. It knows what my diary looks like. It knows when I need a coffee and it will assign me tasks um depending on on those those things in my in my calendar. So, there's some practical tips. Hopefully, that's helpful um to people.

Before we started today, we did when we sent out the invite, we did ask if anybody had any questions. We had a couple come through at that point. One of them that I think is particularly pertinent was around um somebody asked about Lex and what um the plans are at Lex to implement AI. What we've done is we've taken that and extrapolated it to talk about what are the plans at Lex and other um case management system providers um for AI. So we have approached Lex um and Lex have told us that they'll be offering an AI option later this year. Um, they've recently purchased or acquired Uncover AI um as a company. Um, the AI option they're planning to release will be on their council product um and would enable uh barristers to uh use AI for case documents and collaboration. Um, Lex were very keen to point out that um they know people want it. They know people want AI implemented, but they're really keen to make sure they're doing it properly um and they're doing it securely. They know the importance of the data that we're we're all storing there. So they're they're keen to make sure that it's um it's not rushed through and I think we all we can all understand that and and attest to the fact that it's really important these things are rolled out properly.

We've asked um Advanced as well about MLC and Advance do have a standalone AI product. Um, it doesn't integrate with MLC yet um but but may do uh one day. Uh, so if you're on Advance, do do speak to them about their um their standalone product and for those chambers on Barbook um, we understand that Barbook have big plans for AI. Barbook has been built more recently, it's been built with AI in mind. Um, I don't have any details to share of that offering at the moment but exciting things are coming, I understand. And I think that um, I don't know if there's any views from the panel, but my my sense is that if AI was integrated into those systems, kind of our everyday case management systems, that's a huge win, isn't it, for for clerks and chambers. Any thoughts?

>> Yeah, absolutely. I'll jump in there. Yeah, I think it would be. Um, I obviously I'm not a clerk, but I could see benefits um for AI integrating into specifically case management systems. I think it will, without going on a tangent, I think it would allow for data, whatever that data might be, it could be barristers' diary systems, um, even I guess profiles, it could be information associated with cases. If a tool or an AI system was able to look at that data and send out responses to queries, it could be something as simple as uh, we have a a a bundle that's come in. Who is the best barrister to look after this bundle based on availability as an example? That could be some a useful interaction with that AI data. But for me, because that is where a lot of the key data is held, I immediately feel that having some sort of present AI presence within that data would allow you to start looking at that data in a different context. Even with using going back to the reporting functionality, that's another element that would be quite a strong value add to including something some sort of AI tool.

>> Yeah, I think the data the reporting side of it is is key, isn't it? Because that's where all of our data is. It's a manual process at the moment to kind of export some of that and especially when we're trying to compare lots of different reports at the same time. And it also feels like those >> tools are our knowledge base, aren't they, of information. And um uh if we can find a good way of of extracting that into something that is queryable, if that's a word, um, then uh that would be that would be incredible.

So another question that that came through was, we've started using ChatGPT to help with emails, but we avoid including names. Are there any GDPR issues that we need to watch out for? Um, Kaylee, I wonder if you can uh, resident GDPR expert.

Yeah. So, it's not just names. It's anything that would be identifiable information. So, it could be the court reference. Um, it could be anything. If you gave it to a colleague, could they identify it as being um about a specific person? Um, is all I'd say. Just go really, really careful that there's no identifying information in there. If it's clean, there's no identifying information, you're fine. But if there's anything that could be used to work out who you're talking about, avoid it. Um, but that's also with my subject access request hat on because that's driving me up the wall at the moment. So, from two points of view, one, make your life easier if you do get a subject access request, but also just, you know, make sure that you're not putting any kind of personal data into AI if you can avoid it.

>> Brilliant. Um, thank you, Anna. I'm going to come to you with one that has been asked in the in the chat um which is in what way is the paying version of Copilot more secure versus the free version and how is Instant IT's AI tool um AI a private um secure tool?

>> Yes. Okay. Great. So the um Copilot offers is largely on three different plans and I think there are because it's Microsoft and they obviously develop a lot, they also put out different versions of this advice at different times. The the ca the as as it stands, there are three versions of Copilot. Uh, the first one is it's completely free and it's the one that you will have on your personal laptop if you're a if you're a Windows Microsoft person in your in your spare time as well. That is completely free and that one is not secure. It doesn't go like it doesn't comply to the standards of where your location needs to be processed or anything like that. If you're using it on a personal device, generally the advice is to swerve the swerve Copilot in the same way for the same task that you would swerve free ChatGPT for. Um, the the version that's available to currently included on all enterprise and business plans. So that's the majority of people who will be using a device from Chambers. um, that that version of of Copilot is secure because it it doesn't have access to your SharePoint folders. So, it doesn't have access to all that secure data, but it is running within the the confines of your tenancy. Um, which means that the your data isn't being sent anywhere. It isn't being processed anywhere outside of the the EU. Um, then there is obviously the paid version of Copilot. Um, and that one uh does have access to your SharePoint and your folders. It will um so well the the idea being that it has the context to all of the business files that you'll need to uh get accurate and good and relevant answers. Um, however, in order for it to to make sure that's useful, um, you have to run a kind of preparation project where you review your SharePoint folders, um, and who and what the sharing permissions are on each of those things and documents as well. So there's a little bit of housekeeping required um as an example to ensure that all hiring records are restricted to certain certain locations and certain people um and aren't going to be accessible to to everyone um because that could cause a breach of like that internal internal data at chambers could be could be a risk if those permissions aren't set up correctly.

Yeah, I think it's a really important point that there's a lot of concern about what data are we putting into AI that will leave the environment, leave chambers um in this case. Um, but actually data remaining in chambers can still be dangerous if it um is training tools for other people um especially if you've got barristers against each other on a case or if as you say HR records are available to all and sundry in chambers if someone queries um how much does the CEO get paid um and and the the salary records are all available for everybody um that that wouldn't be pretty. Um, your the AI secure tool Anna, the um Instant offer. How how how is that secure?

>> So, it's secure. It's set up within your Microsoft 365 tenancy. Uh, which means that you're not putting your files anywhere new. Um, you're not putting your files anywhere new. Um, we also ensure that the that the tool runs it with models that are in your location that you specify for data data saving uh for uh inference and processing. Um, so many of the big companies that produce LLMs that have created these LLMs like Anthropic who produce Claude um and obviously ChatGPT, OpenAI created ChatGPT. Um, there are options within all of those paid versions for where you want your loc your data to be processed. Um, as an example, ChatGPT, even if you're on the paid version, you your data is likely being processed globally, which means that if you're once you're running the query, it's being processed in up to 20 server locations anywhere in the world, most likely the US. So you if you do have concerns about where that data might be going, it's just worth like looking into any any tool, just look worth reviewing exactly what that means and and where where it might be going. Um, so our AI tool is secure because we ensure that it stays within certain locations if you're concerned about it. Um, the inference process, even if you the inference process generally is secure, but you are taking taking what the tech companies say for granted. Um, there's no real way to to check um because what they say is your prompt is sent to a server somewhere, then you get the response back and it's deleted from that server. Um, so like the the process itself is fairly secure, but it is based on based on that trust.

Um, brilliant. Thank you. We've haven't got to all of the questions, but there are others which um I'll try and go back to separately uh to this. Thank you to everyone who sent questions in. I'll try and get back to those um privately. I'm sure that all of us on the panel would be very happy for people to to reach out directly um and ask questions. I'm just offering that I haven't asked them, but I'm I'm sure they will be happy. Um, so do do email any of us directly and ask any questions. We'll see if we can help. Um, before we go, we've got two minutes. I wonder if I could ask each of you just to share one uh very quick piece of advice for clerks or staff as to if they were going to take their their first steps with AI today, where would you where would you tell them to start?

So Kaylee, what's your what's your kind of quick piece of advice for today?

Don't treat it like Google. It processes the data. It remembers the data. It creates risk. Um, just look at what you're putting in. Where's it going? And can you rely on it? Um, getting it wrong can affect everyone in chambers.

>> Brilliant. Thank you. Uh, Anna, >> um, if you'll choose a choose a collection of champions or peers and ensure you have a healthier mix of enthusiasts and skeptics because the enthusiasts will be really keen to drive things forward and will come up with lots of ideas and the skeptics will help with your guardrails um and make sure that your implementation is actually useful because they'll let you know about it if it's costing you more time and it's saving.

>> Brilliant. Thank you. Alistair. I think um for those not using it, it's not going away. Um, it's going to be here for a very long time. Talk to the powers that be, get the policies in place, get educated, and then just get started. But also consider the ethics of what you're doing as well as the safety, but also the knowledge sharing. It's really important that especially in the clarking community for things to get progressed, you need to share your insights with the wider community, however that may be done, even if it's, you know, regular IBC patch-ups like this. So, that's what I would recommend.

>> Brilliant. Thank you. Well, a huge thank you to to Kaylee, Anna, and Alistair for your time and your your expertise today. That's been hopefully it's been helpful. Um, it's been helpful for me. Hopefully, it's been helpful for everybody else. Um, what I would encourage is if anybody's got further questions, please do email them through to to any one of us or or to the IBC um and and we can try and assist with those. But also, if anybody's got anything that they would really want um a future chat about AI to to cover, it's not going away. Um, I'm sure we'll we'll keep having these conversations and if anybody has anything they'd really want that to cover in more detail, um, then please do say um, email the IBC and I'm sure we can try and get something sorted. But thank you for attending. Thank you to the three of you on our panel um, for all of your insight. Uh, I hope everybody has a good rest of your day.

>> Thanks everyone.