Episode 6 (Season 9) : AI in HR: The Risks Most Employers Miss

Paul: And so what the I did was wrote it for him. They looked at it, went, oh my God, that’s so good. Probably changed a few words and now it went. Problem is, is that they scared people over a law that did pass, but the governor did not sign. We here at CEDR had several hundred clients who then. Not all of them, not all, but several dozen of them.
Paul: Call in and go. What’s with this mandatory, sexual harassment training that everybody’s got to do? How did you guys miss that? And we’re like, we didn’t miss it. The governor ain’t in your state. Sign it. He he’s never going to sign.
Voiceover: You’re about to listen to an episode of. What the hell just happened? Join Paul Edwards and his guests as they discuss interesting HR topics and solve.
Voiceover: Some of our listeners submitted questions,
Paul: and occasionally I’ll go off topic to talk about whatever I want. Barbecue, space exploration, growing your business and things that interest all of us. not talked about AI in the past, I’ve really been waiting. I’ve been spending the last year and a half or so before we could do today’s podcast, because it’s taken society a little bit to discover ChatGPT and the other, language learning models and to begin to start to use them for HR stuff.
Paul: And I now have enough material, Jennie for us to do this podcast. And I feel like it’s a little bit of an emergency podcast.
Jennie: I’m excited for this podcast because I feel like this is a this is a win for me because we’ve spent the last two years of Paul obsessively talking about how great AI is. And today we’re going to talk about
Paul: how bad it sucks.
Jennie: Oh, I was right now how it’s you need to be very, very careful about it. Again. I’m in charge of compliance.
Paul: Yeah. Jenny’s in charge of compliance.
Jennie: It’s, it’s not a be all, end all solution. And actually, for HR and employment law, it’s it’s not something you should use on your own. It’s not something we are able to use and we know HR.
Jennie: Why is this an emergency?
Paul: AI is now becoming way more prevalent amongst our members. Remember, we have the CEDR Solution Center and when I say prevalent, I mean they’re trying to use it to solve HR problems and to write stuff for them. You know, I’ve seen some Facebook groups out there giving away prompts, Jennie, like, write me this policy about X, Y, and Z and make sure that it doesn’t violate any of my state laws and make it comply with any other laws.
Paul: It should. It can’t. It can’t ever.
Jennie: I my team will ask. It will we’ll test it to write a policy on this law. And it’ll be wrong
Paul: and it can be very wrong. So again, we’ll give some examples here. First thing AI generates convincing answers, not necessarily correct ones. I want you guys all to keep in mind that HR decisions and actions you take demand accuracy and understanding how the laws apply.
Paul: You’ve got to have a really strong understanding in compliance. And then finally you need context. So context is probably the most important thing. And if you leave out a piece of context, it can give you a completely wrong answer. And in fact you can include context and cause it again give you a really wrong answer. So you know, again, I just want to tell everybody out there and we’re going to put this up in our air base camp, which is Cedar’s Facebook group.
Paul: Kind of got to stop using AI to give you answers. Yeah, yeah, yeah.
Jennie: Because it’s going to get the HR concepts wrong very often. You know, you can’t rely on for getting the law right. And also it’s not taking into account the context. So who you are as a business, the realities of talking to a human being. We’ll put in, you know, write me what should I do about this employee who did x, y, z?
Jennie: And it’ll be like, well, you should have a verbal conversation or you should issue a write up missing the fact that you’re a health care practice and they just injured a patient.
Paul: Right.
Jennie: We don’t need to start with a verbal warning. Yeah, this is pretty serious, right. But it doesn’t have the concept of the reality of what’s happening there and how serious this issue is.
Paul: And on the other side, it may get you to do something kind of serious when you should kind of back, back off and take a moment. Do not ask it a question about how to handle any issue with a pregnant employee. No, just just don’t
Jennie: Can I can I tell them how you tried to use it for that?
Paul: Yeah. Yeah, yeah,
Jennie: I think Paul is writing a blog or something. And he thought, let me just see if the AI tool would be helpful for this.
Paul: Yep.
Jennie: And he wrote a blog answering a question, and it wrote a really well written answer. And it took into account the ADA, EEOC guidance, a whole lot of stuff. But it when it ran through my eyes, I was like, Paul, it missed the Pregnant Workers Fairness Act, which is a federal law that’s been in existence for several years, which makes the answer completely different now.
Jennie: And so it picked up on other federal laws, but not this one, which is the most on point one. And because we are all human, Paul missed it. And Paul’s an HR expert. And again, it was based on guidance that you had been using for, you know, 15 years now, just in recent years it’s different. Yeah. And he totally was like, oh my God, you’re right.
Paul: Yep.
Jennie: It just missed that little concept right there. And I think the AI convinced you it was right. Paul: Okay Jennie, so this is confirmation bias and action. Yep. Yep.
Jennie: And it took me multiple prompts to even get the AI to figure it out.
Paul: Right. So the problem here is we’re prone to our own bias. And my bias there was I wrote this.
Paul: It must be good AI amplifies your bias. So if you say I have a pregnant employee who’s late too often, she’s missing shifts, she’s taking too many breaks, and we have these other problems with her. It’s going to, follow my bias there. And literally it’s going to, it did write up a PIP, which is a terrible thing for small employers to be using.
Paul: It warned her that if she was late or missed any more shifts, she risking being fired, and then it address some of those other problems which in the end look like you took a pregnant employee and put her in a closet away from everybody else. And so I just want, you guys know, someone used a prompt to do what I just described, and then sent it to us and said, is this okay?
Paul: We put this together, and we immediately were like, that’s written by AI. Yeah. And no, it wasn’t okay. And thankfully they were a member. But good gosh, it looked
Jennie: it look good.
Paul: It was written so incredibly well. And it knew she was pregnant. It went, it was told. If you’ve been listening my podcast, you must know that once someone falls in this temporary disability of of maternity, they get to be late.
Paul: They get to miss a few extra days. They get all kinds of accommodation afforded to them pretty much carte blanche now. They can’t not show up for work. That can be they can’t be late.
Jennie: And, you know, if I’m over here at compliance, technically, technically it has to be communicated actually for a medical reason and all those things.
Paul: Nonetheless, the AI, amplified their bias of of their frustration. It picked up on their frustration and went, you know what? I’ll handle this and
Jennie: get rid of this person.
Paul: Yeah, we’re going to make this person miserable. Yeah. So be careful about the bias there. Jennie, you remember this? We get these all the time. Payroll company.
Paul: I love this giant payroll company. I’m guessing someone in sales doesn’t understand anything. Did understand and kept hearing from everybody. Lot of different laws got changed this year by a legislator. Everybody’s handbooks need to be done. You should use that. And so they went to AI had AI, I’m going to guess the prompt, write me a scary email that I can send all of our potential leads in this state that will get them to call in and named a few things for them.
Paul: And so what the AI did was wrote it for them. They looked at it, went, oh my God, that’s so good. Probably changed a few words and out it went. Problem is, is that they scared people over a law that did pass, but the governor did not sign. Yeah. We here at CEDR had several hundred clients who then.
Paul: Not all of them, not all hundred, but several dozen of them. Call in and go. What’s with this mandatory, sexual harassment training that everybody’s got to do? How did you guys miss that? And we’re like, we didn’t miss it. The governor in your state.
Jennie: Didn’t sign it.
Paul: He’s never going to sign that into law. And so again, it wants to read the beginning, skim the middle and then read the end.
Paul: And then it wants to draw conclusions, which if you’re writing a novel and you’ve created a long outline for a chapter, you put a lot of stuff in there and you want it to help you write. That’s probably, a pretty cool kind of tool to have. You might come up with some, neat stuff, but if you’re trying to write policy and you need something to be clear, and giving you guidance, then we don’t.
Paul: Jennie you don’t skip the middle, do you?
Jennie: Do not skip the middle. The middle is kind of important. Yeah. And it’s I think that maybe what happened, although it’s still made things up with something I was just telling you about how we decided to test it. Like, hey, how would I answer this question about handling, it was a termination or an ex-employee or something, and we fed it the documents and we said, just look at our handbook and just look at our confidentiality agreement.
Jennie: We came back with a very well written termination.
Paul: It was incredibly well written
Jennie: really good. And
Paul: the writing was good. I want to be clear
Jennie: The writing was good. Yeah. And he quoted references to our non-compete agreement and non solicitation agreement from our handbook and confidentiality. Those things do not exist in either of those documents.
Paul: That’s right.
Jennie: They cannot for different reasons.
Jennie: And this is not the podcast for that. Those types of agreements if they’re allowed to be had go somewhere else, they simply don’t exist there. And it just made it up.
Paul: So it even
Jennie: if, like a member had done that, they could have issued that letter. Yeah. And then that actually could have caused them problems.
Paul: It really could have caused a lot of problems. Oh, I want to share this one on the ozone concern. Okay. Do you remember that. Remember that one when it came in the notes.
Jennie: So we’ve actually had multiple questions about ozone related things in the past few weeks I actually don’t know which one you’re talking about. And it’s weird how they come in in trends like this.
Paul: Well, this is one where I say in in my presentations. I titled it What to Do when an employee brings an AI generated concern to you two different entities, the employer and the employee went to an AI tool and asked it about ozone. So the employee was pregnant and concerned about her workplace and she was like, oh, we use ozone.
Paul: So she asked ChatGPT about ozone. And it came back and went, oh my God, it’s going to kill you. Do you want me to write a letter to your employer?
Jennie: It’s worse than WebMD
Paul: I’m I’m I’m paraphrasing. Yes. Don’t even, I’m paraphrasing, but it wrote a letter which, quite frankly, was scary and looked like it had been written by an attorney.
Paul: But contextually, she didn’t say, I work in a dental practice. Our doctor uses an ozone treatment with a recovery system. Didn’t give it any of that information. And so what it thought was, is Jennie in, you know, when you, like, rent a car and you agree not to smoke in it or stink it up or anything, so they have a machine called an ozone machine.
Paul: They put them in hotel rooms. You set it in there for 24 hours, and it puts out this layer of, ozone gas that neutralizes the smell everywhere. And it’s really bad for humans.
Jennie: So the AI thinks this person’s sitting inside of that room.
Paul: They think they’re sitting inside of this room. The AI did not have the right context. Paul: The doctor goes to it and gives it more context. And oh,
Jennie: because the doctor runs it through and uses it to write a response.
Paul: He wants to write a response with it, and he came back and said, they’re being ridiculous. There’s nothing to worry about with the ozone stuff. Well, the real answer to this lies somewhere. Not even in the middle.
Paul: She’s expressed a concern. You can’t be dismissive of it. We realize now that she didn’t write that letter something wrote it for her, and it didn’t have context. We need to take the emotion out of this now. Okay? You’re not being attacked. It’s just a, person trying to figure out how to be safe in the workplace when they’re pregnant.
Paul: Doctor is, like, on guard. He’s kind of pissed because he thinks he’s. She’s trying to sue him. We’re not experts in ozone, and we don’t need to be. We can tell him. Look, go. Go to the manufacturer. Tell them what the problem is. This thing must be FDA approved or something. If you’re using it in practice, get the literature on it.
Paul: Tell her. Thank you for bringing this concern. The AI didn’t say anything to him about telling her. Thank you for bringing this.
Jennie: I’m going to investigate this. Or in the meantime, since you’ve expressed his concern, I’m going to move you out. So you’re not around the ozone while we come up with a solution?
Paul: Yeah. While we come up with a solution.
Paul: And then the next thing is, is you can provide that information to her and say, Jennie: Also ask her to go talk to her doctor.
Paul: That’s the next thing. Take all of this information and go talk to your doctor about it and give them all of it so that they understand we have the recovery system. And in the end, I can tell you what we ended up with, which is exactly what you just described it.
Paul: It happens for about eight seconds, 2 or 3 times, or not at all during the procedures. And he would just have her step away. She didn’t even have to get out of the room.
Jennie: Yeah.
Paul: Everybody just agreed. As long as she was five, five feet away and the recovery systems are working properly, that was… you know, probably no danger.
Paul: I just had to share that because both parties got really worked up because both of them asked the same question. But asked and answered it in a different way. We fixed it.
Jennie: Yeah,
Paul: because we’re awesome. I guess I might have a little bit of satisfaction here because it really is struggling. Doing H.R.
Jennie: It it is it really because it misses.
Jennie: So the example you just gave it missed. The human component of the goal is not to write nasty letters back and forth
Paul: or to win. There’s no winning
Jennie: The goal is to make the working relationship work, to be able to work together and find a solution.
Paul: Yeah. Okay. So this is a good example because one of the things that we don’t want to ever do is subject you.
Paul: And we’re probably going to touch on this a couple of times, a subject you to a law or rule or regulation that you shouldn’t be subjected to. So that’s very, very important. And that’s a mistake people make all the time when they’re trying to create their own policies. They borrow something old. Okay, Jeannie, if you borrow a policy, of ours from, well, we’ve been around since 2006, you think would
happen if you got a hold of one of those, many of those policies that we wrote in 2012 from CEDR who’s an amazing company who knows what they’re doing.
Paul: How good would that policy be that we wrote in 2012?
Jennie: We would immediately write a memo to the team telling them that we’ve rescinded the handbook and we will be reissuing new policy. Yeah, because that is so old. And it’s so outdated federally. And and who knows? But it’s in the state. We just had, Victor, who talks to prospective new clients.
Paul: Yeah.
Jennie: Ask someone was asking, you know, they’re no longer remember they used to be, they’re growing, they acquired more locations, can we continue using that same handbook, or do we have to pay for an update? And it’s only a few years old. Yeah, and we’re not trying to sell when we don’t need to sell. But he asked me and I’m like, well, where are they?
Jennie: And he’s like, well, they’re in New York. Like, I don’t need to know how old it is. If it hasn’t been updated in the last year, it’s it’s wrong.
Paul: It’s already wrong. Wrong. Yeah, it’s already wrong.
Jennie: And a big problem we also see is everyone’s very happy to help out their peers. Yep. And they will share their handbook with other people.
Jennie: And we’ve had people come in and tell us, well I don’t need your handbook. I already have one. And we say, well, let us see the handbook will evaluate it for you. And we look at it and it’s our handbook and it’s a great handbook for the member we wrote it for. Right. But for you FMLA doesn’t apply to you.
Jennie: But you promised it to your team. California laws don’t apply to you, but you promised it here. You have made your life so much more difficult. Or the flip side is you’re in New York and you’re using an Arkansas handbook. There is
Paul: a single policy out of their
Jennie: a single policy out of it. It’s it’s wrong and it’s going to get you in trouble.
Paul: Yeah, yeah. So in this instance, someone’s trying to, create a policy because they want to address what happens when someone is on emergency call within the practice, and specifically that they have to come back. They may have to come back to the practice on the weekend. And so this is something that you can do, right? This is something that you can say, look, you can on call.
Paul: There’s some rules about being on call, whether or not they’re actually on call and sitting and waiting or whether or not they’re just on call and can can show up, not going down that rabbit hole. But we just wanted to pick out one of the provisions. And basically the policy as written, altered their entire handbook to say that all travel, all travel to and from the office.
Paul: And this is a problem in this practice, because this practice has multiple offices, and sometimes people travel between those offices and somebody at times people travel directly to the office from home. And what they meant to say was, look, if you get called in and you got to come in for an emergency, we’re going to pay you and we’d like to pay for your travel time as well, coming from the office.
Paul: So clock when you leave the house clock, when you get home. And they wanted. That’s what they wanted to say. But that is objectively not what they said. They subjectively said all travel is now compensable. Now, look, everybody was focused on what happens when someone’s on call and they were writing a policy for what happens when you were on call.
Paul: But then we come in and pour water on their fire and go, hey, you left out four parts. You need to know what the law says and actually follow that. This might not follow that. And by the way, why would you obligate yourself to pay for all these extra hours? I mean, it could have really been used against them.
Paul: So there’s a good example of how I looks, really. It looked at like a again, like a very well written policy,
Jennie: We’ve had a few very similar okay, member contacting us and show us, showing us what they’ve already kind of done and asking us to just help them.
Paul: They’ve written a policy that I’ve written a policy
Jennie: well, they’ve written and in this case they actually wrote a corrective or termination letter. Paul: Okay,
Jennie: something like that. And they kind of just want us to put our eyes on it, Paul: which is a normal occurrence.
Jennie: Yes. Okay. Great. I mean, if it’s fine, something on your own sent to us, we’re happy to look at it. Yeah. In this situation, they did it based on what they asked AI. And that raised different issues because the AI doesn’t know everything that even our member knows.
Paul: Okay.
Jennie: So it’s missing so much. So these issues happened to be around employee attendance. And employee attendance is always an issue. And there can be very standard protocols about, you know, corrective action. Here’s our attendance policy. You’ve missed X many days blah blah blah. In some situations it is missing that there is a specific state law covering that type of absence.
Jennie: Like in one of them, there is a state law saying you actually have to provide employees a certain number of hours per year to attend to, like child school meetings.
Paul: Okay.
Jennie: And it totally. It didn’t know that.
Paul: Yeah.
Jennie: Because why would it know that. Right. And so they actually were going to write something, someone up for something very specifically protected under the law, which, most employers probably don’t know about.
Paul: Yeah.
Jennie: You know, and that’s what we’re here to track for you. And it’s very particular. But there’s other ones that more broadly, they’re asking like, okay, well do I have to allow an employee to use paid time off for this absence, or do I have to give paid time off for this? Questions like that and consistently the AI is coming back saying no.
Paul: Yeah.
Jennie: You’re not required to offer paid time off at all. Your state or your state doesn’t require PTO, PTO, or vacation. And in each of those instances, the member used phrasing like PTO paid time off of vacation. But in reality, their policies actually had sick leave policy. Why? Because their state or city or county mandates that they offer.
Paul: So the state does…
Paul: So the AI didn’t
Jennie: The AI, because you didn’t ask it about California paid sick leave law asked about PTO. It’s like you’re not required to offer that. And so people relied on that. They would have violated a whole lot of state law, right? Their own policies because because in this happens all the time, you know, you as a business owner, Paul, you may even ask, hey, Jennie how much PTO do we offer a year?
Jennie: And I’m going to like, well, we actually offer vacation time and we offer sick time, separate policies. But you’re just in vernacular saying, yeah, PTO
Paul: and, you know, I mean vacation. But yeah, you’re
Jennie: Your AI does not know that.
Paul: no, I did want to touch on one thing that you made me think about when, when CEDR has a member calling in with these issues, the member is not just getting help with it, but they also get the proper education.
Paul: And what we find is, is that when managers and doctor and our doctor owners who manage their teams are calling in, they’re getting smarter and smarter and smarter, because as each problem comes up, they get another piece of the puzzle put on top of them. The AI is teaching the wrong lesson. That’s that’s the other problem that not only will it give you the wrong answer, but then it will arm you with
incorrect knowledge.
Paul: And then if you’re good at your job, you want to absorb that and use it going forward. And what you’re doing is you’re using the wrong information going forward. Okay. Jennie I really appreciate your, input today. And, hope to see everybody on the next episode.
Jennie: Thanks Paul
Paul: Yep, we’re out of here.
Voiceover: Thanks for joining us for this week’s episode of What the Hell Just Happened to You, Paul and yourself, a favor, and please share the podcast with your network.
Voiceover: If you have an HR issue question or just want to add a comment about something Paul said he’s recorded on your phone and send a podcast and just happened.com. You might even ask if we can play it on the show and you can also visit H just happened.com to learn more about the show and join our H.R community.
Voiceover: Don’t forget to like and subscribe and join us again next week.