AI in the Workplace: Opportunities, Risks and Employment Law Considerations

David: Hello, welcome to the MFMac Employment Podcast. I'm David Hossack, and I'm delighted to be joined today by Alan Delaney.

Alan: Hi, David.

David: Before we get started, just a reminder that we send out a free employment law e-news every month. If you don't already subscribe to that, please email us at employment@mfmac.com and we'll get you added to the list. So today we're gonna be speaking about artificial intelligence and how that's likely to change the world of work, and what the employment law implications might be flowing from that.

And I think it would be daft to assume that nothing's going to change.

Alan: Oh, I think that's absolutely right, David. There seems little doubt that AI is going to change the world of work as we know it. If I look back on my lifetime, we probably haven't seen a change of this magnitude since, the rise of the internet, and if anything, I think we expect it to be even more significant, than that.

The Prime Minister's described AI as the defining opportunity for our generation. However, as you know, David, with great opportunities come sometimes great risks as well that will need considered and managed from an employment law perspective, and hopefully we'll get a chance to touch upon some of that in this podcast.

David: Yeah. It seems only a short time since we had the first podcast on ChatGPT when it was a very new thing, and things have moved on hugely since then. And I think what we see in many businesses is that, employees are using AI routinely in the course of their duties and I think there was an interesting survey by the CIPD last year.

Alan: That's absolutely correct. Last year, I found it very interesting to see this myself, and last year in the autumn, CIPD published their labour market outlook, which listeners may well be familiar with. This surveyed two thousand employers, and this one had a particular focus on AI. And I guess the thing that caught my interest was that of two thousand employers surveyed, seventy-six percent, so the vast majority, said that employees were already using AI tools at work.

This was particularly pronounced within the public sector and within the large private sector employers compared with, for example SMEs. But the thing that really struck me from this particular survey, David, was that most employees, fifty-eight percent, were reported as using freely available public AI tools.

So we're talking about ChatGPT, as you've just mentioned. Of course, others are available, Gemini, Copilot, Grok. With only a smaller number, forty-two percent, using enterprise or customized versions, in other words, versions that had been particularly paid for by their employers.

David: So in this sector, for example, there might be a legal one.

Exactly that and why that is a real concern actually is that there is obvious risk for employers where public AI tools are being used, particularly because the terms and conditions associated with those publicly available AI tools tend to mean that the information that is being put into them by users is no longer private and confidential.

It can be used for machine learning purposes, and this gives rise to very real and obvious concerns around things like personal data, trade secrets and other confidential information. So that's why many employers are best advised to consider paid-for enterprise versions.

So the genie's already out of the bottle, Alan, and many employees are already using AI, using these brands that you mentioned. What do you think employers need to be thinking about to manage the risks flowing from that?

Alan: Well, I think that one thing that really comes through loud and clear from that survey last autumn that it's never been more important for staff to have a clear policy that they can turn to, to tell them what are their dos and don'ts in terms of using AI in the course of their duties, particularly around, as I've mentioned, confidential and sensitive information.

It may seem like a pretty obvious point in this, but we've seen this before, haven't we, David, in terms of social media use whereby the practice, if you like, went ahead of the policy and many employers found themselves scrabbling around after the horse had bolted from the stables to try and get a good policy in place.

So if you haven't done so already now is the time to do that. I suppose the other benefit in a policy as well as the kind of dos and don'ts, is it can reiterate, pretty important messages here along the lines of the one that was given by the CEO of Alphabet. Of course, Alphabet owns Google last year where he warned people not to blindly trust everything that AI tells them.

And we've already seen, haven't we, a number of particularly embarrassing examples of this, including even within the legal profession where there have been situations of lawyers who have been very unfortunate in placing case law that has turned out to be fictitious case law before judges in courts where that case law, in fact, had been generated by AI, but not based on any actual case, a case that had been invented, or the term is often hallucinated by AI, which for a lawyer like myself or you, David, is a particularly toe-curling situation to find oneself in.

David: Yeah. Let's just make something up.

Alan: Well, absolutely. So again, I think the key question that a policy can really help with is letting workers know what are the dos, what are the don'ts, how can they get the balance right? Hopefully at the same time, it's not just a case of drafting a policy, putting it on the internet. It ideally would be backed up by some good training so that everybody is clear how AI can be used to best effect to enhance existing roles.

David: Now, you're referring to enhancing roles there, Alan. If I can be devil's advocate for a moment, we read an awful lot about AI replacing roles. Is that the case?

Alan: That was looked at also by the CIPD survey that we're speaking about. And whilst I do think we need to take survey evidence with a bit of a pinch of salt, it did find that one in six employers, so some seventeen percent, did expect the use of AI tools to reduce their headcount in the next twelve months.

Now, personally, I think that is a little bit pessimistic, but perhaps there's more comfort to be taken from the fact that just under half of those responding said they expected AI to make no difference to headcount in the next twelve months at all. So some expect it's going to make a difference, others less so.

I think looking more laterally, however, here to this particular question, maybe of more interest and concern to employers is not so much AI replacing jobs, but to make sure that jobs are apt enough such that they can change over time, as they may need to as AI tools become used. So what are we thinking about there? By way of a concrete example, we're thinking about things like contracts of employment, we're thinking about job descriptions. Are these sufficiently flexible to allow changes to be made to roles as might be required as AI looks to, for example, enhance roles to take away the more routine work and allow workers to focus on the more interesting and challenging work, which of course is what we hear all the time in terms of what AI can do for us.

David: So one needs to be careful about making judgments about what's going to happen, but we'll probably see some reduction in workforces, but it may actually, as you say, make some jobs more interesting.

Alan: Yes. I think the evidence from the United States would seem to be that it is typically the kind of entry-level jobs, data input jobs that are particularly affected but I've got no doubt it will transform work for all of us.

David: And if we go back to specific employment risks likely to arise, what sort of things do you think employers need to be thinking about?

Alan: Good question. I think a really obvious risk, and actually personally what I think to be a really interesting one, is around AI decision-making.

Now, if you think about employment law, this could relate to recruitment, it could be promotion, it could be training, performance management, or even decisions around termination of employment. And of course it's very easy for employers to look at AI and see the obvious benefits of introducing a greater level of automated decision-making, for example, efficiency, speeding up decisions or reducing labour costs.

We've heard how quickly AI can be used in the recruitment context in particular. Again, interestingly, and going back to that point about enhancing rather than replacing roles, AI can sometimes read patterns that humans might overlook in terms of decision-making such that decisions become actually more accurate.

A really interesting example that's often cited was research that showed doctors who used AI to assist them assess ECGs were actually more accurate in interpreting those, ECG charts. So that's very interesting. And then finally, we've got things like consistency of decision-making.

Again, sometimes something that can be quite difficult when it's been done by managers on a, on a subjective basis. AI may well be able to check against all similar previous situations and ensure a greater level of consistency in decision-making going forward. However, of course, with those benefits come a number of risks.

David: One of the things I'm thinking about here is recruitment. And, as you know, AI is often used to sift, come up with shortlists of people, and even make selections of employees. And I think the challenge there is that you can't actually say what criteria have been used in reaching that conclusion because you can't see how AI works .

If I were to say to you, Alan, "Was AI used in connection with my recruitment?" and your answer is yes, and my next question would be, "Well, what factors were used in terms of, applying those criteria?" And I don't think you'd be able to answer that question.

Alan: Yeah, I think that's a great point and I really think this is something that's going to come into sharp focus in Employment Tribunal cases that are inevitably going to arise over the next few years.

To take your first point it's the opaqueness, isn't it, of the decision-making process. It's going to be difficult to understand how and why a decision has actually been reached in terms of using an AI tool, and of course, that's a very real concern for any decision-maker that needs to sit in a witness box and give evidence to an employment judge, to be in a position to explain how the decision was arrived at.

It's going to be a real challenge. Similarly, I guess, a very well-known concern in respect of the use of AI is the possibility of bias or discrimination because it, of course, depends upon the learning input that's been used to train the system, and at worst, this could lead to a continuation of what is actually an existing discriminatory, state of affairs.

You know, one need only think about the battle to secure equal pay over the last fifty years or so. And in fact, there is at least one quite prominent historic example from the last few years involving a large employer that in twenty eighteen actually stopped using a kind of algorithmic recruitment tool because at the time, because of the data it'd been used to train, the model tended to favor male applicants over female applicants when recruiting for jobs.

And the reason why it did is it'd been trained on the company's recruitment data over the previous ten years, which itself had a bias built into it, unbeknown I should say, to the employer in question. So some real risks and challenges around that that we're all going to have to try and navigate over the next wee while.

David: So at the very least, it's a perpetuation of bias and perhaps an amplification of that.

Alan: That's right. and I think it's going to be even more of a challenge perhaps than it always was. I think back to, goodness, going back twenty-five years to when I was finishing university and being interviewed for traineeships.

I remember, my first encounter of psychometric testing and, now could a decision-maker in that situation be able to explain why it was a decision had been reached in relation to me? But back then that was just one of a number of tools that was used, and I think the danger and risk these days is that perhaps over-reliance is being placed on AI in that context.

David: I suppose with psychometric testing, you were able to at least explain how that worked. But with it's very difficult to get anyone to explain how it's working in a particular situation.

Alan: Yes. And actually a related point is that let's say you buy a, an AI recruitment tool and let's say a challenge is brought around bias or discrimination, if you then go to the supplier, is the supplier going to be willing to reveal what it may consider to be trade secret information in terms of how the decision has been arrived at, how the model actually works? Lots of interesting challenges.

David: It maybe wouldn't matter if the software comes with indemnities, so they pick up any employment costs that flow from discrimination cases and so on.

Alan: Good luck with that, but yes.

David: And I know, Alan, that as well as being an employment lawyer, you've got a very soft spot for data protection.

Alan: Well, you know me too well and we couldn't do this podcast, it would be remiss of us not to say that there are, of course, specific GDPR issues that need to be kept in mind in terms of the use of AI, particularly when it comes to personal data, which is very valid, of course, in the recruitment context.

And we're thinking about things there like transparency, so what is said within existing privacy notices, not just for job applicants, but for staff as well. They need to think about data protection impact assessments, or DPIAs as they're called, for any high-risk processing. And in some cases, there also remains some restrictions on solely automated decision-making, albeit less so since the new Data Use and Access Act Twenty Twenty-five.

But drawing this section together, I think the key takeaway point for me would be this. Since many employment law decisions, for example, around decisions to dismiss someone in the context of Section Ninety-Eight of the Nineteen Ninety-Six Act require individual circumstances to be properly considered and Employment Tribunals, of course are likely to expect no less in the future. It will certainly help manage the risk of AI use for it to become an aid to decision-making rather than a decision maker in itself.

David: Now, moving on to perhaps the end of an employment relationship. What we're seeing these days is Employment Tribunal applications are often being drafted with the assistance of AI, and I think that's quite problematic and indeed may be a relevant factor in terms of looking at the delays that are now in the system.

But before we even get to that stage, you can see AI might be playing a part in terms of grievance processes and so on. What role do you think AI has to play in dispute resolution in the employment arena?

Alan: Yeah, again, a really interesting question there, and certainly employers have to deal with and grapple with grievances much more common than they do with Employment Tribunal claims. We can already see the influence of AI when it comes to the nature, and the length of grievances that are now being submitted by employees. There has been if perhaps one describes it as a unfortunate tendency for these grievances now to become much longer, more complex than they've ever been before.

I've heard stories of grievances going on to fifteen, twenty pages citing, for example US Supreme Court case law authority, things that really just aren't really helpful, I don't think, to the employer understanding what the issue is that the person has a grievance with or how it might be resolved. I struggle to see that there's really any benefit in someone submitting a grievance that they don't really properly understand, or that at worst contains references that just don't make sense in a employment law context. And I think that really is the risk there.

AI can only take you so far. It's based on the information that's put in, and if that information isn't accurate or doesn't explain the context in which the question is being asked, then unfortunately the output is going to be affected by that.

David: I always thought the key points of a grievance were saying what you weren't happy with And what you wanted to happen. And I think what we're seeing with AI is perhaps the stuff in the middle about why it's not good and, as you say, bringing in US Supreme Court cases and so on, rather than focusing on the key elements, what's wrong and how do we fix it?

Alan: Well, that's right and in many ways the nub of a grievance should be "What is the problem? Where are we going wrong? How can we help you? How can we resolve this?" And the focus is very much on resolution of that dispute. And I don't think it's ever been helpful over the years to see a grievance that is overly legalistic. At the end of the day, an internal dispute resolution process is not some kind of quasi-judicial process either.

It should be about fixing what's gone wrong. And I think that's where AI being used, might help the individual narrow down in their own mind the issues that they have a valid concern about versus the ones perhaps that they don't.

But the danger and I think the risk is that they will perhaps make disputes more complex, more lengthy actually more difficult to resolve. And indeed one can't help wonder about, are we going to see a nightmare scenario whereby the employer takes the AI-generated grievance and scans it in, feeds it into its own enterprise AI software, and asks AI to advise on how the employer should respond to it.

There we perhaps have both parties to the dispute, neither one of them really properly engaging with the issues at stake and how they might be resolved.

David: That's not good.

Alan: No, it's a lack of applying one's mind, isn't it, to the problem at hand. It's almost you're outsourcing the problem to someone else.

David: I'm wondering if there's a world here, Alan, where an employee with a grievance might be using AI to say, "I've got a problem here. AI, can you help me write a constructive- grievance letter with a view to arriving at a reasonable outcome here. "

Alan: And the word there is constructive, I think in that sentence. And as I said just a few moments ago, if it helps someone narrow down their concerns such that actually they realize in terms of what they've been told that the employer is quite right in some respects, then it may well mean that a grievance is shorter than otherwise it would be.

But I think the focus is on keeping it constructive. " What is the issue at stake? How might it be resolved?" 'Cause really that's what this is all about and it may well be that although we're talking from a kind of negative context there's absolutely a positive side here whereby AI will help employers and employees to see things from the other's point of view, which I think gets us into a topic I know that you're very interested in, mediation and I know I read something recently about AI and mediation and what role it has to play but query whether or not it's ever going to give the same benefit that a real focus on human mediation could have. But you're the expert on such matters.

David: And that would be the topic for another podcast. But yes, it's a very valid point you raise, and I'm wondering whether also there's another world where employers could be saying to employees, "You can only put a grievance in if you use our constructive grievance software."

Alan: Which is focused on solutions.

David: Well we can think about these things another day perhaps.

So trying to draw this to a close, Alan, there's a lot of stuff out there, and I suspect we probably need to be speaking about this next year or perhaps even sooner given the pace at which this is developing. In the meantime, you might expect me, given previous podcasts, to be asking you a question, and that is, what one piece of advice would you give right now to an employer from an employment law risk point of view? What should they be doing?

Alan: I think it goes back to the point you made earlier, David, about the fact that the genie is well out of the bottle and that being the case the advice that I would give would be implement a policy. If you haven't already done so, please do so now. As you know, Employment Tribunals absolutely love policies.

I've been in Tribunal before where my witness has been cross-examined and the judge basically says, "Mr. or Mrs. So-and-so, can you take me to where it says... Can you take me to the inventory of productions and show me where your policy is?" And it's always particularly embarrassing when, of course, there is no policy in place.

Better to provide the guidance now, even if it's no more than a starting point, so that workers know even just some basic ground rules in terms of do's and don'ts, in terms of using AI because all that's going to happen is more and more workers are going to do so anyway, as I think comes through loud and clear from the CIPD survey.

David: And please, please, please don't use AI to draft your policy. You need to think very carefully about what is in that and to ponder what as an organisation you can live with and what is a good tool and what's a bad tool.

Alan: Couldn't agree more, David.

David: So thank you very much, Alan. As always, a pleasure to have you on the podcast, and hopefully you'll come back before too long.

Alan: Of course. I'm sure there'll be plenty case law generated that will be very interesting to talk about in future podcasts. Thanks very much.

David: And thanks everyone for listening. Please remember to subscribe to our podcast and to leave reviews in the usual way.