Human in the Loop. Or Just Someone to Blame?

[00:00:00] Welcome to the Unmuted Podcast. My name is Jeff. I am the host of the podcast where we talk about all things HR, talent, HR tech, world of work, and of course, AI. I'm also the founder of ScaleHR and Social HR Camp. And these episodes live on our website at GoScaleHR.com and also now on Work Defined. It's work without an O. So WRK Defined. And so we're actually one of three Canadian podcast shows on the platform.

[00:00:28] And now you can pick your platform of choice, Spotify, iHeartRadio, Apple Podcasts, and so on. And so this is an amazing topic. Yes, I'm biased, but you know what? I can be biased. So the topic of this episode is Shiman in the Loop or Just Someone to Blame? Okay. So obviously it's about AI, but it's a really interesting topic. And we have a fantastic guest who I'm going to bring on soon. But I just want to frame the conversation first with a couple of points.

[00:00:57] As AI becomes more integrated into the workplace, we're constantly hearing about the importance of the concept of keeping a human in the loop. Okay? It's supposed to give us confidence that someone is reviewing like a person. It's taking a look at what the technology produces, right? Making sure things are fair, accurate, appropriate, they're on point, and so forth.

[00:01:19] But just because a human is actually involved doesn't necessarily mean that they're providing meaningful oversight. So that's the first point. And I think the second point is really is that when you think about what happens when someone is expected to review an AI-generated recommendation outcome product, but doesn't really have enough time, doesn't really understand how the system reached a conclusion, or doesn't really feel empowered to challenge it.

[00:01:47] Are we genuinely creating accountability, or are we simply putting a human name against something the technology has already determined? And if that's the case, what does meaningful human oversight actually look like? I think this is a really, really important topic. And I'm going to welcome our guest, Karis Hogarth, to the podcast. Amazing. Thank you so much, Jeff. I'm super excited to be here.

[00:02:11] Fantastic. You know, it's been a while, actually. I, you know, I've had you on my list of people to invite to the show. And so we finally did it. So, you know, after a couple of years, but all is good. So before we dive into our topic here, maybe we can tell us a bit about what you do. Yeah, for sure. So I'm based in Vancouver, Canada. I am the senior vice president of people at a fiber internet to the home company called Ting Internet based out of the US.

[00:02:39] I essentially lead our human resources function, also includes things like internal communications, health and safety, and the topic of the moment, AI enablement. So my lane is really around, I guess, kind of HR strategy, talent, culture within the kind of tech sector, tech communication sector.

[00:03:01] And yeah, I'm fascinated by the human in the loop topic. It's certainly something that I have lots of conversations about both in work and outside of work. So I'm very passionate about that piece and excited to chat to you about it. Fantastic. And then you're obviously in a community, like HR community, that I would consider very forward thinking. Do you want to come to HR technology, AI strategy and so forth? And I've been involved in the Vancouver community, well, I mean, forever.

[00:03:29] So, and I've always viewed it as being a little bit further ahead than some of the other communities that I've done work with. So when we hear about the phrase human in the loop. Okay. And so, I mean, I've been talking about it quite a bit and I know that you have also. And so when you hear the phrase everywhere, when organizations talk about a responsible AI, what does it actually mean in practice, particularly in HR? And why do you think we should be questioning whether it's actually working?

[00:03:57] Yeah, this is great. I think at its simplest, it's about making sure that there's a person or a human as part of a process. And, you know, engineering has been doing this for some time, for many decades. And it's become, you know, more synonymous more recently within the AI part of work. And so essentially, if the system is doing something or it's making a decision or it's producing an outcome,

[00:04:28] there's a human that checks it or approves it or that can step in at any time. And so basically, that's what that phrase means. And I think there's, you know, there's in the loop when somebody is actually, you know, holding up a process to make an approval or make a decision or click a button. There's on the loop, which means that there's a system running. Maybe there's a dashboard there, but the human is monitoring at all time and can step in if needed.

[00:04:53] And then, you know, if there's no human in the loop or out of the loop, it means that unless something, you know, goes terribly wrong or blows up, a human isn't isn't part of that process. And it shows up everywhere in HR. I think, you know, there's AI for almost almost every part of human resources out there on the market these days, whether it's around resume screening, interview scoring, whether it's about creating performance review summaries,

[00:05:21] compensation recommendations, looking at flight risk scores. So there's so many different areas within, you know, the HR process that the AI shows up. And making sure that there is a human in that loop, and quite often it's the human resources human that's in that loop, is something that we're seeing more and more of. And I think it's often said now as, you know, you kind of asked why, you know, why we should be a little bit skeptical about this.

[00:05:50] It shows up as the kind of the catch-all answer. So, you know, if there's a new piece of technology being deployed or there's, you know, AI being introduced to a process, and someone says, it's okay, there's a human in the loop. Like, that is going to solve everything. Everybody can calm down, sit down, chill out. Not going to be a problem. There's a human in the loop. Don't worry about it. But I think, you know, without having some skepticism around that, it's, you know, it's not actually doing, I think,

[00:06:19] what a lot of people expect it's doing just by kind of thinking about it from that perspective. So, you know, there's no real law enforced in Canada right now. There is in Europe more so. GDPR requires it for some things. There's the EU AI Act that, you know, has some segments of that in terms of high-risk processes or systems. But it's, yeah, it's definitely something that I think we need to be skeptical about. Yeah.

[00:06:46] And I think you said something that I took away from what you said is it's about process. Because I know that a lot of HR teams may not really have well-entrenched processes. And so the technology is really allowing you to manage that process. It's the enabler of doing something, right? And so what I find is that the organizations that have a really strong process, they're able to pinpoint exactly where the human in the loop is supposed to happen. And so they're basically embedded in it.

[00:07:15] So you've given it enough time, enough guardrails, enough structure so that the person can actually do it. I think when you don't have that strong process, then I think that's when you start getting into trouble. That might be oversimplified, but that's just my general take on it. What do you think? Yeah, I think you're right. I think that there's, you know, there's pieces of that process where this breaks down, right?

[00:07:41] And I think that, you know, the first thing that that human in the loop needs is the time and the space to be able to make that decision. If things are moving quickly, if there's a process that's happening and that human doesn't actually have enough time to be able to do that, then they're just clicking a button. They're not actually taking the review necessary to do it. I think the other piece is quite often in these processes, the human is delivered an output or a decision,

[00:08:09] and they're not necessarily given all of the context or the information behind where that decision came from. And so really, they're not making a decision with all of the facts and the data in front of them to be able to make smart judgment at that time. I think the other thing that's super interesting, and this is, you know, this is kind of getting into more psychology here, but there's automation bias. If you have a system, a process or a tool that 99% of the time is making correct answers,

[00:08:36] you're more likely to trust that machine or that computer or that process or that system. And that 1% of the time when it gets it wrong is easy to overlook or is easy to miss because you're assuming it's going to be correct. So you've got that automation bias where we're trusting the machine to make the right decisions, and they're probably going to make it more often than we would as humans. So we put more trust in those outputs, right?

[00:09:01] And then quite often, the human that is put in that loop or the human that has to, like, come in and step in and call out if there's an issue or an error doesn't really actually have the power to be able to do so. So they need to be empowered to do that, and it needs to be in an environment that's safe for them to be able to push back. Right? If they're not, if they don't feel that, you know, if they feel they're going to get blamed for holding up a process or being the bottleneck or whatever that might be,

[00:09:27] then, you know, they're taking that blame and that doesn't give them, they don't feel empowered then to be able to kind of step up and wave a flag and say, hey, something's not right here. So that can be a bit of a challenge too, right? Speaking of that, because that's like what is into my next question really is about when you have a person that is accountable for overseeing AI, but they may not have the control over it, okay? Where does the human oversight typically break down?

[00:09:52] And then what happens when people don't have the time, information, authority, knowledge to challenge what AI recommends? Yeah, and I think the interesting thing here is there was a study that was done, and I know I'm totally guilty of this, and I'm sure most people are. I think it was called The Biggest Lie on the Internet or something like that,

[00:10:16] and basically it went in and it evaluated how long it would take to read the terms and conditions on some of the most used apps or like social media or, you know, you're signing up for your tellers. You know where we're going with this, right? And I think 99% of us will just click I agree without reading a single word on that sheet. Yeah, we're all guilty of it.

[00:10:44] And they did a review of how long it would take to read these terms and conditions. The worst one was over 30 hours. If you were to sit down and read AT&T's terms and conditions, it would take you over 30 hours to do it. Even TELUS, I think, is 8 or 10 hours if you actually sat and read the entire document. And so nobody does that. And so that's where I think that skepticism I have about the human in the loop comes in because, you know, it's all – I don't think anybody's trying to deliberately deceive us

[00:11:14] with those terms and conditions, but in actual practicality, it doesn't work because nobody's actually reading them. So that's the same fear I have around that human in the loop, being there to be able to make that oversight. Like, if it's 4.55 on a Friday afternoon and, you know, you want to go out and meet up with your friends, go home to your family, whatever it might be, and you're the person that needs to make that final decision on something, it's always going to be a challenge.

[00:11:43] So I think that's part of it. But there's also – there's a pretty tragic example that happened, I think it was back in 2018, and it was a self-driving car. It's when they were testing them in Arizona. And there was a human in the car that was there to provide oversight, and they were there to, you know, monitor, watch, see what happens, how that self-driving car operates,

[00:12:11] and if there was an emergency or an incident that they could take over. Unfortunately, a pedestrian was killed. And the problem that we have here with that empowerment and accountability is that Uber actually were not found criminally liable for this, but the human in the car did have to – did end up – I think it was pleading guilty to dangerous – it wasn't dangerous driving, but it was something along those lines. And I think, who is ultimately accountable?

[00:12:40] Should it be that human in the loop? Should it be the people that built the process? Should it be the company that owns or subscribes to that tool? Where does the actual liability and accountability sit in that human in the process place? If the human is the one that's being blamed for this, then they need to have the amount of time that's necessary to make good decisions. They need to have all of the information available to them to have good judgment around it.

[00:13:08] They need to work in an environment that encourages them to speak up when something's not right. And that's not always the case. So, yeah, there's some real risks there. That's a very obvious and extreme case where loss of life – I think most of us would say that that's not the norm. Obviously, when you're in HR, it's usually not like a life-and-death type of situation, but it's still a good example to use. And so if simply having a human involved is not enough, obviously,

[00:13:36] what does a meaningful human oversight actually look like? And how should organizations design processes that give people the ability, confidence, and authority to challenge AI, particularly when the stakes are high? And that's it. And I think a lot of it is about understanding how high those stakes are. So if you think about it from the perspective of – I don't know if you watch Netflix, I do, and Netflix is now recommending things for me to watch based on what I've watched previously.

[00:14:05] If it makes a bad recommendation, it might be mildly inconvenient for me, but it's not something that I'm going to get super upset about. And so having a human in that loop that's checking every recommendation that comes through to its customer seems like a gross waste of time and resources. But if you're working in the medical profession and this is about medical diagnoses

[00:14:28] or this is about how much drugs to prescribe somebody or something like that, then that is going to be very, very critical. And you want to make sure that that human is absolutely the final decision maker on all of these pieces. So that level of how high is that risk I think is definitely the first part in evaluating that.

[00:14:52] So when you think about it from an HR perspective, a compensation recommendation has real impact on somebody, right? That's a genuine impact, but it is something that can be reversed if there is a way for that person to appeal or what have you. But ideally, you want to make sure that that human is available and very, very much centered in that process to be able to make decisions around people's compensation or performance recommendations or promotion recommendations

[00:15:21] because those actually do have real and lasting impact on individuals within the workplace. So there's that piece of it. I think that's important. And then the time, the information and the power to be able to do that. As long as the person isn't felt rushed in making those decisions, as long as they can see where that information and where that data has come from and they can really trace it all the way back to the data that's being presented,

[00:15:50] that I think is super important. And then that there's no penalty for being able to call out something that is incorrect or wrong. I think that environment in which people are operating is so important. And I think, you know, you have to get all of those things, really. It can't just be one of the three or four things that are important. It has to be all of those things in order for it to feel like a real decision. I totally agree with all that.

[00:16:18] And I think a lot of HR teams just having conversations with my peers is that one of the big challenges right now is balancing time. Right. And so you obviously don't want to stretch things out for two years because then things don't get done. But the advantage of having AI is that you technically and theoretically can speed things up. And so you should be able to have enough time and not have to negatively impact the delivery and productivity side of things. So obviously don't want to have compensation process go two years.

[00:16:47] I mean, that'll be catastrophic, but you don't want to rush it either. So I guess there's a bit of a balancing act, I think. Yeah, absolutely. I think it is really about saying on that scale of risk from, you know, low risk to really high risk. What is the best use of AI in that process? Is it to make the decision or is it to facilitate the administration of the decision once a human's made it? So do you use AI 30% of the time and humans 70% of the time?

[00:17:17] Or is it a really, really quick one? Like things like, you know, being really specific around HR, like terminations. AI should never make the decision about whether somebody should be terminated or not. But can we have AI help draft severance letters? Absolutely. Can we have AI, you know, help build, you know, a framework or some talking points for a termination conversation? Absolutely. Should the AI deliver that termination message to the employee? Absolutely not.

[00:17:47] So it's really about thinking where is the right balance of using AI to be able to help with efficiency and drive productivity and be able to, you know, bring in some of those administrative tasks that can be challenging? And where does a human absolutely need to remain front and center so that it's not impacting negatively the experience of an employee in that workplace? So just shifting gears here, I want to just focus on something that you talked about earlier in your work at Ting.

[00:18:16] So what your approach to AI literacy, you're prioritizing human capabilities like judgment, critical thinking, discernment, ahead of the technical skills, which is what I think a lot of people are trying to do. Why did you take that approach? And then what can other HR leaders learn from it as they prepare their people to work with? Yeah, I think there was two core reasons why we took that approach.

[00:18:42] And we were very, very deliberate about setting out our stall within our organization and communicating very openly and transparently with our employees about what our AI journey was going to look like. So what were we envisaging for that journey? And one of the things that we were very clear about was that AI was to augment, not to replace. And so that was the first stall that we set out.

[00:19:05] And then we wanted to make sure that we both backed that up in terms of how we were delivering AI literacy and enablement across the organization. But also, I'm a big believer that the tool can be as wonderful as you want it to be. It can be super technical and amazing. But without the right human skills to be able to operate in that new environment that we're all working in now, it's all for nothing.

[00:19:31] And so I wanted to make sure that we started with a real clear definition around what was important to us, what aligns most with our values at our organization. And that was about those human skills. And so, you know, we really did start by looking at what things like curiosity, critical thinking, decision making, ethics, judgment, empathy. What does all that look like now in this AI world?

[00:19:56] And how can we make sure people bring those skills into that working environment to be able to operate effectively, be able to make smart decisions and be able to use the tools in a way in which they're intended? And I think what we have to remember is at the end of every AI process or decision, there's usually a human at the other end of it. And so if we can't remember that and we can't have empathy for that human at the other end of it, then what is it all for?

[00:20:24] Right. So we were really deliberate about building our AI literacy program, starting with those human skills and making sure that that was the foundation from which we were then able to start talking about the technical competencies. You know, tools are changing so rapidly now within months or weeks, even sometimes that tool has changed. But what stays and what remains really constant is things like judgment, critical thinking, discernment.

[00:20:52] You know, all of those things have to remain as part of the foundation and part of the value system in which you're working. So, you know, we want people to be able to say this doesn't look right or I don't agree with this or the context was missing here. We want people to be able to stand up and do that. And I think, you know, having the courage to be able to do that courage is another big one. But creating the environment in which people feel safe to bring that forward was was also really important, too.

[00:21:21] I think the other piece is if you outsource and there's been studies on this fairly recently, if you outsource all of those human skills or all of the decision making processes to AI, the skills that you had in order to be able to do that without AI. So it's a muscle. Use it or lose it. Right. So if you stop being the person using critical thinking, if you stop using judgment,

[00:21:47] if you stop using your curiosity or your creativity, those skills start to fade if you're outsourcing all of that to AI. And I think that's a really dangerous, a dangerous move and a dangerous future if if those skills are lost. So I think, you know, I would hope other HR leaders can look at that and say, you know, let's not necessarily start with the tool, but let's start with making sure the people using the tools have the right skills and values to be able to operate those tools most effectively.

[00:22:16] This has actually been one of my favorite episodes. And I was smiling just a few minutes ago because I was going back to a conversation with a peer of mine that said, I do not want to be in a company that we're building that is full of idiots. And this is what, you know, she was talking about, right? This is exactly what she was talking about. And I laughed. I mean, I got it, but obviously now I get it even more. So just concluding here.

[00:22:43] So we talked about a lot here, but if our viewers can take just one thing away from anything that you've talked about, what is that one thing? I think it would be to have a healthy dose of skepticism around what human in the loop means. So taking human in the loop on face value, I think is dangerous. And I think being able to ask questions around who is that human? Do they have the amount of time necessary to make decisions?

[00:23:12] Do they have all of the information required to apply critical thinking and judgment? Do they have the power to say no? And are you teaching them and training them and educating them around the necessary skills to be able to do that effectively? And I think being skeptical in this instance is the best foot forward. Thank you very much. I really appreciate you being here. So Karis Kogarty with Ting based in Vancouver. So thank you very much for being here.

[00:23:41] And so for all of our guests, this is the Unmuted Podcast. I'm Jeff. I'm your host. And we talk about all things HR, HR tech, hiring, firing. Well, maybe not firing. But hiring, leadership, and the world of work and so forth. And so please share your comments. And so Karis, thank you very much for being here. Thank you so much for having me, Jeff. This is super fun as always. I love chatting with you. And that's a wrap. Thanks for listening to the Unmuted Podcast with Jeff Waldman.

[00:24:09] Real conversations, zero corporate filter, all about the future of work. If you enjoyed this episode, subscribe, leave a comment, and share it with someone smart. New episodes drop every week. Until next time, stay unmuted. This is a wrap. Let's go. We'll see you next time.