The Legal Brief: The golden rules of AI in the workplace

Speaker A: This is a Momentum Media production.

Speaker B: Welcome to the Legal Brief, exploring legal trends and recent case law that HR

Speaker A: professionals must be across.

Speaker B: Hello everyone, and, um, welcome to another special episode of the Legal Brief, produced in partnership with our friends and at Kingston Reed. My name's Jerome Dorasami. I'm coming to you today from Camera Eagle Land in Sydney. And I'm very pleased to be joined for this conversation by Michael Stutley, who is a partner at Kingston Reed. Michael, hello.

Speaker A: Hello, Jerome. Thanks for inviting me along.

Speaker B: Thank you for being here. Today we're going to be talking all things AI, some of the inherent risks from its use, uh, and how to strike the right balance in using it effectively and meaningfully, uh, without putting your team and the broader business in any danger. Before we get into that, Michael, I've got an exciting announcement, uh, for you listeners. Now, over the past year, dedicated listeners of this show will have heard, uh, from a number of different voices at workplace law firm Kingston Reed. Having had a number of senior lawyers and partners join me on this podcast to unpack a range of challenges, uh, facing HR professionals like yourselves. Now, these conversations have highlighted just how much is changing in the workplace as legal developments and the adoption of AI present new opportunities and risks for employers to navigate. And that's why Kingston Reed is hosting the Future Workplace Law Summit in September of this year. It's going to feature legal experts and an outstanding lineup of guest speakers. Um, and this summit will offer practical insights into the evolving environment and the challenges facing HR and legal professionals alike. If you want to learn more about this Summit, visit, uh, www.kingstonreid.com or search future Workplace Law Summit. Sounds like it's going to be a really exciting one, Micha. I might have to get down there myself.

Speaker A: It is, Jerome, and I hope to see you and your listeners down there as well.

Speaker B: Yeah, not just the listeners, but I'll, uh, bring along my team too. Now, Michael, we're obviously talking all about AI today and how to strike the right balance in the workplace. Um, before we get into that though, can you tell us a little bit about yourself and the work that you do at the firm?

Speaker A: Sure. So, I'm a partner, uh, at Kingston Reed. I'm based in the Perth office. I spend quite a bit of time in our, uh, Brisbane office as well. And largely the practice is centered around employment, uh, workplace relations, industrial relations, safety and global mobility, which, depending on the year, can either fluctuate between something which, um, takes up a lot of time or, um, depending on the government of the day, little time and what is

Speaker B: it about this area of law that is, uh, so uplifting for you? I've asked a number of your colleagues this question over the past year and I'm always, I always feel a bit warm and fuzzy hearing the answers. The passion that yourself and the other partners at Kingston Reed seem to have for the workplace law space, uh, is really cool. Um, but yeah, what gets you out of bed in the morning for this

Speaker A: work, it really is unique, isn't it? We are people people, so we love getting out there and talking to people, we love getting involved with business and operations. And it's quite a unique area of the law because you don't really get that opportunity anywhere else. So being able to engage with our, uh, clients on a day to day basis, understand their business, understand the drivers, understand the difficulties and really add value, I think that's what gets us out of bed.

Speaker B: Yeah. So, Michael, let's talk all things AI and uh, as a starting point, I think it might be worth getting your sense of the lay of the land. You know, we've just started the new financial year and obviously AI, uh continues to be one of the biggest trending topics in any workplace conversation and certainly across the broader market as well. But when it comes to its usage in businesses big and small, and given the extent of usage, what's your sense of whether everyone is striking the right balance? Certainly there's some outliers at either end of the spectrum, but what's your general take on the state of affairs?

Speaker A: So the state of affairs, I think where we are right now is an experimental phase. You see, uh, HR practitioners and lawyers and everyone else using AI are really understanding or coming to terms of what it can do. It was all introduced, you know, a couple years ago when Chat GPT was released and it was really quite magical at the time. It was something that could generate responses based on an input which didn't require you to enter a Google search term and go through threads and forums and try and find an answer. So I think we're still experimenting with what it can do and how we can use it. And I think that that's where some of the danger lies because there's still a bit of uncertainty.

Speaker B: Tell me about what you're seeing as some of the hidden or perhaps very visible risks, uh, from AI use across the workplace right now.

Speaker A: Yeah, look, it's really new versions of old risks rather than entirely new legal risks. And it's a lot of things that existed, but I think we've got to go back a step and you look at the Fair Work system, right, and the Fair Work act, it's very human centric, right? So it's, it assumes there is human decision makers behind each of the various things that can happen and go wrong. And where AI comes in is when you use it to summarize information or get through large data sets.

Speaker B: That's all.

Speaker A: Okay, we understand that use case, but where we can sort of be led astray is we start to use AI as a decision maker. So we start to rely on it to give us answers to complex problems and then just assume that that answer is correct or that it's not in some way bias in terms of what it's relied on or that it's misinterpreted some of the information that's been provided. So AI as a decision maker is one of the, the topics. I think that is going to have increased attention going forward and it's particularly going to play out in the courts.

Speaker B: No, certainly. And, um, there's already been some use cases of misuse of AI for lawyers and other, uh, similar professionals. Can I ask though, I think there's a lot of societal or cultural pressure for workers to be using AI right now, whether that's intentional or not. And so there would be a lot of workers out there who feel obliged, if not forced, uh, to use these new technologies because otherwise they're going to get left behind, they may lose their jobs to AI. And so from my vantage point, it is kind of understandable that workers may just try to use it for anything and everything, including being a decision maker, uh, given that they may fear the ramifications of not leaning in. So to that end, it seems like it would be a really hard balance to strike and certainly one that employers and HR teams, you know, need to be on top of in, I guess, offering the right kind of reassurance.

Speaker A: Yeah, look, that's a really good point. I think that there shouldn't be a fear to engage with AI. Uh, AI can provide enormous benefit to HR practitioners or any professionals really. But where we need to be cautious is this shift in, or movement between using it as an administrative assistant or an assistant to get through large volumes of documents or materials and understanding quickly the state of play to outsourcing entirely your judgment or delegating tasks to AI to complete without any human oversight. So we often use this term and you hear it around a place of human in the loop. And it's so critical because there should be a human in the loop at every stage when you're using AI and the reason you need a human in the loop is because you need to be able to withstand external scrutiny later on when it comes to the use of any of the AI responses.

Speaker B: Certainly, uh, and just on that point of needing a human in the loop, one area of certainly workplace law, but just general workplace, uh, process where humans are required, um, is in internal investigations. Now, as I understand it, AI will be very good at summarizing large volumes of information, but, uh, workplace investigations will require a human touch. What are you seeing in this space?

Speaker A: You're spot on. So investigations are one of these fast growing areas in terms of the use of AI involved in investigation processes. Now, you're spot on when you say that it's really good at summarizing information. And yes, sometimes you can have investigations with incredibly large volumes of documents and materials. And it's a really good tool to summarize that and prepare a chronology and understand who the key people are and who you might need to speak to in terms of witnesses and other participants. But that's where it stops when it comes to investigations. Because investigations aren't just based on summarizing information. They're about credibility and context, competing evidence and procedural fairness. And it's about observing behaviors through interview processes and really understanding what's going on and making findings of fact based on judgment and all of this. There's a balance to be achieved, and in fact, and there's an entire test on the balance of probability that's required in an investigation process. Now, if you completely outsource those inherently human functions to an investigation, then can you rely on the findings at the other end? And can you then use those findings in terms of a workplace process? If that leads to a dismissal, then how is it going to be justified? How are you going to withstand that external scrutiny that we spoke about? So I think yes, you can and absolutely should use AI, but it's not about replacing the human element of an investigation process. One really positive way to use AI in an investigation would be to put together your allegations and then have AI interrogate those allegations for you. Are they specific enough? Are they really getting to the nub of the issue? Do we make sure that we engage adequately with the complaint and the material that we have available? So use it as a tool or as a colleague to assist you in that process, but don't completely outsource your role in terms of making findings?

Speaker B: Yeah. So just to clarify then, or to put it another way, it's not about having certain processes within the workplace be siloed off from m AI, it's about knowing how best to utilize that tool so that you can still achieve the right kind of outcomes. Uh, and not just simply look for the easy way out.

Speaker A: Exactly. And achieve those outcomes in potentially a shorter timeframe. Because one of the biggest criticisms with investigations, and there are a lot of them throughout workplaces, is that they take too long. And that's not a, uh, criticism that should be directed at anyone involved in investigations. They do take time, and a lot of that time is thinking time. But there's also a lot of preparation time and a lot of sifting through information and understanding what's relevant and what isn't relevant. So yes, it can potentially reduce that time, make some of that process easier and ultimately fairer.

Speaker B: Yeah. And I guess given the volume of AI inspired applications to the Fair Work Commission right now, it would be somewhat ironic for one of those applications to have come as a result of an AI generated dismissal as well. So, you know, it's probably important for employers to ensure that they are staying, um, above board with all of this.

Speaker A: Yes, really important. And one of the topics that, ah, your listeners will be well aware of is general protections, uh, general protections claims, where whatever is in the mind of the decision maker is critical to the claim, the success of the claim. Now, if you have outsourced this entirely to AI and AI has come up with a decision and it has used its judgment, who is the decision maker? And how can you possibly discharge that reverse onus if you don't have the human decision maker giving evidence about what was in their minds at the time? So it really, it can be a very challenging issue to overcome if you don't get it right from the beginning.

Speaker B: Yeah, um, and I think that leads nicely into another topic I wanted to raise with you, Michael, which is misplaced trust. Now certainly, uh, workers across the board will have a certain level of skepticism, healthy or otherwise, and certainly some automation bias, uh, when it comes to these new technologies. What are you seeing in this space right now?

Speaker A: Jerome, I love that you've raised this point about misplaced trust. So there's two issues that um, that I want to talk about when it comes to that. And I think it's really important that we just start from this place of it being magical, first of all. So where does trust come from? Trust comes from us receiving, uh, a response from AI and it looking so perfect and so well written that it must be true. Right. So we have an automation bias, which is our starting point. For some reason, humans will trust a machine more than they'll trust the person next to them. And I don't know if that's come about over time, but you can trust a calculator to always get the number right. But ultimately what AI is doing is you're putting in a prompt. That prompt is made up of tokens. Each token is a letter, it's a word, it's a comma, it's a space, it's all of those individual tokens combined together. And the ar, the large language model, and the algorithm sitting behind it is interpreting the next most predictable token. So it is so well read, far better than anyone, any other human on the planet that is able to then string together sentences and paragraphs and entire stories that will seem like it is the truth. And the problem with that is that if you rely on it as being the truth without verifying, without checking its sources, without really interrogating or being inquisitive about the response that's come back to you and understanding how it has relied on the information to give you that response, then the problem is you're going to misplace your trust. Your trust is going to be in the machine, which is potentially leading you down a garden path which you ultimately don't want to be.

Speaker B: Mhm. Yeah. I guess that goes back to the conversation we were having a bit earlier about having AI act as a decision maker when it should be more of an assistant. So with all of that in mind, that automation bias, misplaced trust, what uh, do employers and their HR teams need to be doing on this front to ensure that trust is being placed, uh, where it belongs?

Speaker A: Well, it comes back to this human in the loop. So it's ensuring that throughout the process of using AI that you are interrogating the response that is, it's really about being inquisitive. So again, it's not a, it shouldn't be a fear to use AI. Engage with it, experiment with it, but make sure you don't just take a response that's being provided to you and proceed on the basis that is correct. Because AI can hallucinate, AI can get things wrong and it can misinterpret information that's being provided. And medical reports are a really good example of that.

Speaker B: Yeah, no, fair enough. And I think that offers a nice segue into talking about privacy. And I imagine there are certain pitfalls, um, here, uh, for workers using, uh, certain information or inputting certain information into platforms that they shouldn't be. Now certainly there are hard and fast rules around, uh, sensitive client information like medical reports, as you say. But what might be some of the grayer areas around use of the information that they'll have?

Speaker A: Yeah, look, the one question that I always ask is not what did the AI tell you? It's what information did you give the AI? Because that's the critical distinction here. So if you were to input medical information or report from a treating practitioner about an employee's particular condition, you are using that information in a way which, yes, is, um, in one sense connected with the employment and necessary to be able to make decisions about the ongoing employment. But also you are inputting this information potentially into an unknown space. So who has control of that information? Where does it go? What else is it going to be used for? These are the types of questions which we don't properly understand because we don't know what sits behind the other end of that screen that you've inputted a person's medical report into. So it's not just medical reports that are in this category. It might be other sensitive information like tax file numbers and other, uh, personal information, which if misused, can create real problems. Now, a lot of people will say, well, there's an employee records exemption and allows us to use that, but this really is an untested space. And the question is going to become how can you be satisfied that the use of that information does not go beyond what you have used it for to interpret the information you've been provided?

Speaker B: Well, certainly if you're using a generic tool that is publicly accessible. So taking all of this into account, what are some of the golden rules for usage here now? Ah, certainly it is hopefully okay for workers to use AI in the right context, uh, but they'll have to be doing so responsibly. So, reflecting on everything we've been discussing, Michael, what do you see as being the golden or, you know, hard and fast rules moving forward?

Speaker A: Yeah. Okay, so if there was going to be some golden rules for me, and in fact these are, uh, some of the ones that I use, first of all, health information, treat it as a red flag category. And so you be extra cautious when it comes to anything that's sensitive or health related. Be very careful about how you use that and what systems you upload that information to. Another one would be to de identify wherever possible. And that, uh, in some respects addresses the first one, which is the health information being a red flag. If you de identify the information and redact personal information, for example names, Medicare numbers, dates of birth, addresses, those types of things, then it does make your engagement with the information a lot safer. And it means that you're potentially not exposing you or the business to additional risk. Keep humans in the loop. That's got to be one of the most well discussed rules in terms of using AI. If you have a human in the loop, the human's going to moderate it. It's going to be able to test the veracity of the responses. It's going to understand the obligations that exist in our current framework, such that there should not be an over reliance on whatever the response is from AI. Preserve the source documents, don't lose them, don't just upload them and then forget about them and think that, well, that the machine's got it and so it'll always be there. If at some stage you need to be able to justify a decision that's been made and you had the assistance of AI, you need to be able to explain what source documents were used in terms of interrogating and getting to that endpoint. You need to really have established governance rules. There needs to be, it doesn't need to be detailed, but there does need to be an understanding, a policy across an organization about what is the acceptable use of AI. What are, uh, the approved systems that we are going to use? And you really need to. You mentioned a point just earlier about public system, publicly available systems and people using personal ChatGPT accounts. This is all saying that if you're not paying for the product, you are the product. So unless you are engaging with a system that you have paid for and subscribed to, and you can verify that the information you putting in to it is not being used to train larger language models, then you need to be very cautious about what you're inputting and then just apply the newspaper test. That's got to be sort of the one rule that everyone remembers, that if you would not be comfortable providing this piece of information to a stranger on the street, don't put into AI.

Speaker B: Well, those are some fantastic rules. Thank you, Michael. Now, at the risk of asking you, uh, to offer some unsolicited or unbillable advice, uh, further to the outlining of those golden rules, what, I guess, broader or more holistic guidance would you offer to employers and their HR teams moving forward against the backdrop of those golden rules, um, but I guess just, uh, some guidance or words of wisdom as to how they can be viewing the importance of their roles as leaders, uh, in the new financial year and beyond as they move to ensure that their entire workforce is using AI in the right fashion.

Speaker A: So, two parts to that question, I think. Um, in terms of HR practitioners and leaders in the workplace. It's such an important role because you are, ah, compiling various pieces of information from various sources and you are guiding, you're providing advice to business in terms of the decisions that need to be made. So engage with AI to assist you in that process, but don't completely outsource the role. That's really it. It just means not delegating your role to the unknown. And in order to do that, you need to then comply with those golden rules or call them what you will. But essentially we just need to understand that AI is here to stay. We can engage with it responsibly and it can be a very effective tool. But that's just it. It's a tool. It's not going to replace human judgment, human decision making. It's not going to replace procedural fairness, steps, delegations of authority. Those things are all going to exist, but they need to exist in harmony. And I think that you can have both. You can have a situation where processes are sped up and things are, uh, fairer because of that process being sped up. But you just need to be cautious.

Speaker B: Now that makes a lot of sense. Well, Michael, I've really enjoyed this conversation with you. Thank you so much for your time.

Speaker A: Thanks Jerome, appreciate it.

Speaker B: And just a reminder, for those who want to learn more about Kingston Reed and its services, you can visit www.kingstonreid.com and if you want to learn more about that future Workplace Law Summit, uh, that I mentioned a bit earlier in the conversation that's taking place in September, you should also check it, um, via Kingston Reid's website. Thank you so much, uh, to you, the Listener, for tuning in. Thank you to Kingston Reed for its ongoing support of hr. Ah, leader. We hope you all enjoyed this conversation. See you again next time.

Speaker A: Sam.

Speaker B: Mhm.