For more, visit ADP.com/AI
Video Transcript
Patricia Stark
Well, thank you all so much for being here today.
Jason Albert
Thank you. We are glad to be here.
Patricia Stark
I mean, everyone says that AI is really only as good as its data. So in workforce technology, what does that really mean in practice?
Amin Venjara
Well, I think about the relationship between data and AI. It's like the ingredients and a meal. My sister has a farm, and when I go up there, we pick ingredients from the farm, and then we cook it that evening. And you can taste the difference. That's like the relationship between data and AI. Like if you have really good ingredients, you can make an awesome meal, but no amount of AI is going to make up for poor data.
And that is that relationship of being able to think about when I have AI, what is it? What is it relying on? What data is is feeding into it? What context does it have? Does it have enough understanding of the person that I'm dealing with, of the organization that I'm looking at to be able to think about the work process and the outcome it's trying to achieve?
And oftentimes we put a lot of emphasis on the AI or the models. But so much of that now is we're seeing as it depends upon the data that that that it depends on the what's happening to somebody today. What what pages were they clicking on? What is it. Has it been a busy time of year for them?
What's happening in their personal life? All those kinds of things can come into making the right decision. We know that as as humans, when we engage with somebody, we have to we try to engage and understand, is this person having an off day? Is it the right time to talk to them? How do I talk to them? That's the same kind of data that we all collect when we interact with somebody that AI needs.
And so when you have an AI system, you want to ensure that it has that right capability to have access to good data, market data and contextual data to make good decisions. Yeah.
Patricia Stark
That's a great analogy. Quality ingredients and understanding the circumstance mixed in with some good emotional intelligence.
Amin Venjara
There you go. Makes a good meal.
Jason Albert
Amin, I expect To see you on next season of Top Chef.
Amin Venjara
I don't think you want that.
Patricia Stark
Jason, your role is focused on global data privacy. And of course, governance is essential. What guardrails should employers really have in place?
Jason Albert
Well, when you think about the guardrails that employers should have in place with respect to data, you have to start with thinking about access controls. You have to make sure that the right people have access only to the data that they need for their jobs and for their role. We call this in privacy data minimization, and it's a critical element.
And it's equally true of AI. AI can develop all these great insights. All these great recommendations can act now with agents on data, but it's important to make sure that the AI also only has access to the data that it needs for its task. So I'd start with data minimization and thinking about the purpose you're trying to accomplish and how the data you have relate to the purpose.
Then, of course, you need the appropriate guardrails around security to make sure that the data isn't accessed by bad actors. And you also have to think about what transparency you're going to offer your employees, letting them know how their data is going to be used. Because there's nothing so important in the workplace as trust. And employees will trust employers if they understand how their data is being used.
And there's transparency there about that use.
Patricia Stark
Yes, absolutely. I think trust is more important than ever.
Amin Venjara
Absolutely. It's the foundation, right? Data is the lifeblood of AI, and people won't let AI act on their data if they don't trust how it's going to act. So you're right. It's absolutely essential in this new age.
Patricia Stark
Yeah, absolutely. And Naomi, once that foundation is set, how do you turn reliable data into a product that actually improves a specific business outcome?
Naomi Lariviere
So for me, I think when we're talking about data, data alone isn't the sole answer. You know, we talked about it. You have to have context. You have to understand the employee, the organization, the workflow, what they have access to, what they what they don't have access to, and really what the moment is meeting. So if you understand all of that and you've got data paired with context, then you also have to think about, okay, well what problem.
Am I actually trying to solve. Because if you look at how AI can improve stuff, you can throw AI in anything, right? But I think what's important is what is the problem. Do you understand that problem? And do you understand where there's opportunity to augment with AI to help you get through to the right answer, and then that creates the value that you're looking for that outcome.
Patricia Stark
Yeah. So not using AI just for the sake of using AI, right. Right. Amin, all employers are working toward the same goal of wanting to attract and retain top talent. So before AI can improve on that process, what data and context does it really need to understand before it can deliver that kind of outcome that the organization is trying to achieve?
Amin Venjara
Yeah, I think it's useful to again, think about a specific example. I think about hiring when you want to hire somebody, an intelligent system is going to know what's happening in your business. Are you in busy season? Are you do you have a- the need for workers? How can I tell if I have that signal? Is there a lot of overtime happening?
Did you have a lot of turnover that happened in the organization? I can show you that there's a need. Maybe there's not enough recs that have been open in your organization for the demand that you have. So sensing first that there's an issue and giving you the data to see that, and then being able to say, okay, you want to act on that.
Great. Let's open up that requisition to be able to hire the right person in the right role. Let's have the job description built in the way that matters. Leveraging the tone of your organization, the data that matters, the compensation for this job in this location that's appropriate for that role. So then now I can be calibrated and compliant also with the with the regulations in my state or locality, and then be able to sort through lots of resumes, put it out in the right job boards.
See what the results that come in and get the right candidates. But also make it easy for those candidates to apply. So the data that's needed always is going to follow the problem you're trying to solve. And I think putting that first and the outcome that you want to have first, and then ensuring that that data flows along so you can have the right context for AI at each moment.
Patricia Stark
Naomi, could you walk us through a job change or a payroll adjustment where AI identifies the required steps, roots the work, and flags the issues? And where does the human expertise really need to enter there?
Naomi Lariviere
Yeah. And actually, Amin started off with his example around hiring. You know, if you think about the job of a recruiter, where do you want them to leverage their expertise? That's really in those consequential judgment decisions. But the reality of a recruiter job today, it's a lot of coordinating, a lot of busy work. I've got to go find candidates.
So I'm going to some of the big job boards to look for the right candidate. Then I have to, you know, email them, call them, get them set up for interviews. And I'm coordinating an internal like panel of people and like, you know, meeting calendar invites all of this stuff. It's a lot of coordinating just basic tasks. And that's actually when you think about what is the problem you're trying to solve for a recruiter.
It's I want them to spend the time, like really evaluating the individual for the right role. I don't need them spending a lot of time coordinating the components of that. And that's where AI can help and leverage. So when we think about what will help a recruiter, we're thinking about taking the busy work out, helping them to have a more streamlined process for setting up interviews, coordinating the feedback that we get from the hiring manager and the other people in the process so that they can spend the time really evaluating that individual for the role and being able to help the hiring manager make the best decision for the right role.
Right person. Right compensation.
Patricia Stark
Yes. And have the energy to be really fully present at that time, because they're not spread to thin doing all of those other tasks.
Naomi Lariviere
Exactly. And that's what recruiters actually want to do. They don't want to, you know, think about outlook invites. And did it go to the right person or email address. Like that's where AI can really help take that friction away for the for the recruiter.
Patricia Stark
That's great. That's going to be music to people's ears. So Jason in the recruiting journey, so what boundaries should govern the collection and the use of the candidates data?
Jason Albert
Well, when you think about this in the context of the recruiting journey, you have to think about what data you're collecting in the purpose for which it's being used. You're trying to recruit candidates, you're trying to find the right candidates. So first, when you collect information collected directly from the candidate, you should tell them what you're collecting and why.
If you're collecting it from a third party, you're searching a job board. For example, if you reach out to a candidate, you should make sure that they understand how you source them. Beyond that, you shouldn't keep the information beyond what you need. If I am a candidate and for whatever reason, I fall out of the running, maybe I want to be considered for other jobs and you could put me into sort of a job bank, and that's terrific.
But I also but that ought to be my choice. Maybe I don't want to be in that job bank. Maybe I'm like, great. But like, this was the only role that I was interested in. And you don't need to keep my information going forward. So again, I think it's that transparency about the collection and use using it for the established purposes of recruiting and respecting the candidate's wishes.
Once that recruiting journey comes to an end, successful or otherwise.
Patricia Stark
Yes, yes. Very good points. Amin, once you attract top talent, the next big goal is to obviously keep them. So how can data identify the right moment for a little bit of that nudge, you know, like prompting a manager to have that career conversation.
Amin Venjara
Yeah. And you think about all the things that happen once you get hired, like, what are the things that have to happen to create a good work experience? Just start from that, that that employees journey in the beginning, in the first days of onboarding. Think about how confusing it is to start at a new company. Who are all these people?
How do I deal with my benefits? How do I get acclimated? How do I be successful in my first 60, 90 days at the job? There are nudges that- but at the same time the manager is busy. The manager is trying to do their own job. Keep keep everything running. And some nudges that you have at the beginning is to say, hey, how's it going?
Just that simple reminder to say, hey, how's it going? Can I help you with that? We all forget once we're doing a job, we forget what it means to do that job. So those nudges that can be delivered at the right time, in the right way, through the right channel. Maybe it's mobile, maybe it's email, maybe it's on your watch.
Whatever it needs to be to help you to engage. But then as you progress. Okay. So it's been a year. It's been some time. Maybe somebody it's a year anniversary. Can we help remind you to recognize that anniversary? Recognize someone's birthday. Recognize when maybe somebody spending a lot of time on overtime and that person needs a break. Maybe they've been spending a lot of time late nights or weekends.
Those are the things to recognize and be able to pay attention to. So so nudges have to be put it the right way. And the data actually doesn't have to be that complicated. And but we as there's all things as we as human beings that we know we're supposed to do and we forget, we forget sometimes to call our parents.
We forget sometimes to invest in retirement. We forget sometimes to go to the doctor to keep ourselves healthy. These nudges help us to be more human. Do the things that we want to do to be the best version of ourselves. That's what good data can do to create a great work experience.
Patricia Stark
Yeah, and to not let really even something very small or big or fall through the cracks, that could really make a big difference.
Amin Venjara
Absolutely. Absolutely.
Patricia Stark
So, Naomi, what human understanding does a manager need to consider before deciding how to respond in that scenario?
Naomi Lariviere
You know, I think the example that Amin gave, which is onboarding, it's a really great example of understanding data plus context and the work that an individual is doing. You know, Amin, pointed out that, you know, coming into a new organization, learning new people, their names, their faces, you know, what they do. That is a very stressful time for any individual.
When you know, for a manager, it's really important to remember the stress of the new job, and you want them to be like reviewing, you know, the moments that really matter to that employee. Because we know that a large percentage of employees turn over if they have a bad onboarding experience. So if you have the manager and and they have the right data in front of them, they have the right gentle nudge to actually go, oh, hey, it's been a week.
How's it going? You know, you know, do you have a buddy? Do you know do you understand this process yet? You know, is there anything I can help you with? That's where, you know, data plus the work context plus a nudge can really help a manager be more successful in bringing that employee into the organization.
Patricia Stark
Yeah. I mean, because even think about when you go out to eat, if someone doesn't check in on you, once they've delivered your food and you need something, it's a it's a problem.
Naomi Lariviere
Yeah. You don't want to be calling for your waiter to come to your table or like you're sitting there going, like, where are they? This is an opportunity to be really intentional about these moments that ultimately do play a critical role in someone's satisfaction with their job.
Patricia Stark
Absolutely.
Amin Venjara
And I think what comes up even more to that is as you think about ongoing, like, okay, somebody's been in the role for a little while. Are you checking in on their career progression? You get so engrossed in the work, you may not think about what's happening to their development, right? Because you just forget you get in such a pattern.
But also, you know, again, things like this, like the ratio of positive feedback to negative feedback needs something to be like 4 to 1 or even 8 to 1. We know that with our kids. Yes. And we know that in the work environment too. But we need to be reminded about that because we can apply those kinds of insights, but then make it part of our reality, make it part of how we operate every day, make it part of our practice.
Amin Venjara
And that's what can make AI human and helping us to become the best version of ourselves.
Patricia Stark
Yeah, I love that. Because, you know, sometimes when people are feeling siloed or on their own at work, you know, no one may ever ask them how they're doing. And it can make such a big difference to find that out. Yeah, yeah. That's great. And, Jason, how can organizations ensure that AI tools stay on the right side of being helpful rather than becoming intrusive?
Jason Albert
I think there are a couple of concepts that really help with AI tools staying on the right side of being helpful. One concept is expectations and the other concept is delight. So in terms of expectations, the AI tool ought to propose nudges or things that are in line with people's expectations. So like after a week, how's it going?
After a year, it's your year anniversary. You know, maybe it was my birthday, but it would be weird if it, for example, remembered my anniversary with my wife. Like, that's probably a little intrusive. That's sort of like, well, now what's this thing doing? So I think it's keeping in line with expectations and making it seem part of the natural flow of things, rather than something that comes from sort of left field about, well, why did it say that?
Why did it know that? And then I think delight AI can still surprise us. But, Amin, to your point about career progression, maybe after a year I makes a suggestion. Hey, Jason, you've been in this role for a year. Here's some coursework you can do to prepare for your next role. I might be delighted, like it's already thinking about what I can do next.
It's helping me prepare. So I would say it needs to align with people's expectations or create moments of delight. And that's the real way to stay on the right side of things.
Patricia Stark
Yeah, yeah, boy, that's making me think about all kinds of things with the way that I use AI, right? I love when it members something that I didn't, but if it's listening to me and offers something that I didn't ask for. Now I'm not really happy, right?
Amin Venjara
I think there's another dimension to this about just being able to trace back and kind of audit and to see like, how did we get here? Like what? What set a reasoning did you do or what. How did this conversation unfold with the with AI. And that gives me that sense of trust. Right. Where I can not only have expectations of that, it's aligned exhibition and it's delight, but I can like interrogate and audit what happened.
And that feels to me more like I'm in control, right? Because I can trust this, that it's doing something. But I can, you know, you can inspect what expect, right? We say that. So if I have these expectations, I can inspect it and know that that really happened the way I wanted to, to then build that trust over time.
Jason Albert
Amin as the compliance person. I really like the emphasis on auditing here.
Amin Venjara
So there you go. There you go. I'm here to serve.
Patricia Stark
So Amin, compliance involves changing laws, local jurisdictions and countless exceptions. So how do verification and validation help turn that complexity into AI recommendations that an organization can safely use?
Amin Venjara
Yeah. So compliance and jurisdictions this is some of the challenges that business owners face, right? They want to be able to run their organizations serve their customers. But then they have to face the these these complexities of, you know, I'm not trained in tax law and I'm not trained in labor law, and I don't know what it means to do it in this jurisdiction versus that jurisdiction.
Amin Venjara
I don't have to deal with that. Yeah. I just wanted to make it easy. And I want to be able to run my business. And how so? How can I help in this? Well, on the, the the validation, it's having this trust layer right where I can have the, the verification to know that, you know, if this is the knowledge base of what all the laws are, the jurisdictions are to ensure that this is accurate.
This is true. This is grounded in what is actually the the correct application of that law and that that compliance to my specific fact pattern, my organization, my locations, my employees. But then at the same time, it's saying, great. Now that that's true is taking is how I reasoned through that content. Can I validate that? That's accurate. Again, inspect what you expect and be able to see that happen each time, so that now I can gain that confidence to know not only that what's coming through is accurate and the foundations on which it's built is accurate.
The data that's gathered. But then to see that the reasoning in which it is applied is also accurate. And I can inspect every step that can then stop, build that trust to know that it comes from a foundation and a place that's going to be accurate for my business. Yeah, yeah.
Jason Albert
I'd like to add to that. In saying this, these rules that we're talking about are so important and you can't really make a mistake. Right. Because the consequences could be severe. And so that trust layer is important. And this is complex stuff. There are thousands of jurisdictions. There are thousands of rules and regulatory interpretation and decisions and things like that.
So Amin, to your point, this validation, this understanding that's making sure the AI isn't hallucinating, it's making sure that even where it's not hallucinating, that it's interpreting things correctly, having that sort of human oversight so you can trust the recommendations that it's making is just vitally important. Because we're again, we're in this HR employment space that's so critical to businesses, that's highly regulated, that's highly specialized.
And so being able to when you look for use of AI, when you look for offerings in the market that be able to have that level of trust is absolutely critical.
Naomi Lariviere
Yeah. I would maybe add that you want it to also be effortless. Like today, a lot of, HR practitioners, they spend a lot of time like searching websites and everything. But when it's like right in front of my face and it's coming from trusted sources, it's there. It's got the context of what I'm trying to accomplish. Then I've got that intelligence right there at my fingertips.
Naomi Lariviere
And if it's it's coming from somebody I trust, then I have the confidence that I can use my judgment to make the best decision for whatever the scenario is. And I think, you know, with with this, if you do it right, it can be effortless and it can be easy. And I'm using the expertise that I have, which is my experience and my judgment.
Amin Venjara
I think another another thing about compliance is that we all know two things. One is in law, sometimes there's a clear answer, sometimes there's not. And when new regulations come out, you need clarification. You need to know and to spend some time clarifying. What does the agency mean about this? At the same time, AI well, all of us use AI and we see it's very happy to give us an answer.
It always wants to please. Right. And sometimes overly so. Yes, the mark of a good system is to know your limits, to know when when the intelligent thing is to not give an answer, to say I don't know. That is an intelligent system to know where to stop. And I think it's really important, especially in this, this, this of human capital management where we're dealing with people's lives, we're dealing with compliance regulations that affect the business is to know where those limits are and to know where to say, I don't know or this is not clear yet, so that you can make the right decision and the human judgment comes in.
Patricia Stark
I would love to just dig a little deeper into that, because when I use AI and ask questions and I am clearly seeing that it really just wants to please me and I'll say, no, really, don't try to please me right now. Right? But it's never told me I don't know. And hearing you say that now, that concerns me.
Amin Venjara
That's right. And that that's the nature of the- That's the nature of how these models work. Because there is this next token prediction. There's always this next token I'm going to give you. I'm going to tell you something else. And it wants to be helpful. It's they're designed and trained to be helpful that way. So building harnesses that it can actually limit that and to constrain it for what's needed for, for high stakes situations is where you have to build that balance and that just doesn't come out of the box.
You just don't get it. That's and knowing the these natural contours of how these models operate is important, to know how to apply it to a situation that's as high stakes as as HR.
Jason Albert
Yeah. Amin, this reminds me, we were talking before we were filming the session, and I was talking about how I've taken up a lot more tennis recently. And so I was doing lessons with my coach one Sunday, and we know the coach in the next court. And he's like, well, she's really being tough on you. Jamie, you shouldn't be so tough on him.
Jason Albert
I said, I'm not paying her to be nice, right? Right. I want her to coach me. I want to get better. I need her to tell me directly, like what works and what doesn't so that I can I can get better. And again, in these compliance areas, you need an AI. It can be friendly, it can be nice, but it needs to give you sort of the firm guidance to make sure that you stay on the right side of the line.
Naomi Lariviere
Yeah, I think I definitely think it's, you know, to your point, these models will tell you whatever you want to know. They have all of this information. They've been trained on the data that tells us. But what they're missing is do they have the data. Do they have the work context. And do they have the guardrails to really shape it so that you actually have a clear and confident response?
Patricia Stark
Great points. So Naomi, walk us through a job change or a payroll adjustment where AI identifies the required steps, the routes routes the work and flags the issue okay. And where does the human expertise enter into the process.
Naomi Lariviere
Yeah. Well today, you know a job change. Someone's getting promoted. I'm going to promote you, Jason. You just got a chief I-don't-know. We’re promoting Jason. Right. And, you know, that is for some of our practitioners. It's not something that they do on a regular basis. And, you know, a lot of HR practice is based on memory.
And, you know, they might have a SOP in, in SharePoint or something that tells them. But, you know, work is nuanced. There are things that you might do for promoting an attorney that is different than promoting a product manager. And, you know, remembering all of those steps and everything can get kind of challenging. So where AI can really help in the job change is giving them a very clear structure on how the workflow is going to actually happen.
Do they have the right permissions, and are there things that they should do versus not do? So for example, if I'm the HR practitioner who's promoting Jason, I may not have access to the compensation element. That might be Amin’s job in this scenario. And what I want the AI to do is take me through all the pieces that I can do and then tell me what happens next.
So if I can't change the pay information, I want that to go to Amin so that he can complete it. So that's giving me enough information, that's giving me the clarity of what I can and cannot do. And then it is also telling me what the right next step is now, where I think the human element of it really is woven in, is knowing what the boundaries that I have as Naomi, what I can and cannot do, what my permissions are, and where my expertise has a limit to where it has to go to somebody else.
Patricia Stark
Very good. Jason, how do you ensure that the process is accountable while still protecting sensitive employee information?
Jason Albert
So, you know, in terms of making sure that the process is accountable? I think Amin that it really well when he said inspect what you expect, right. You've got to have an audit trail. You've got to understand how the AI reached the conclusion it did, what information had acted upon, how the controls it was given in terms of its prompts, in terms of the protocols that it was subject to.
I think that's really important. Beyond that, when you are doing this type of auditing, right, you don't need to necessarily know the identity of the person who is going to be promoted or a series of people. So you can separate that out. So you can even look at sensitive information that's de-identified or even anonymized relative to the individuals.
And you can judge the value of those decisions again while protecting their personal information. And then beyond that, obviously, we have done a lot of compliance work well before AI or even computers. When you think about preventing discrimination and doing bias analyzes and things like that, and those techniques are still applicable and important in an AI world. And so, you know, with all of those things, it's definitely possible to hold AI accountable to be able to understand how it reached decisions, to be able to determine that something went awry without compromising individuals personal information or putting that at risk.
Patricia Stark
Terrific. Thank you. Amin, how can AI powered storyboards, translate workforce data like turnover costs, pay equity, or future labor needs into a useful business story.
Amin Venjara
Businesses are all about stories. It's the stories that we want to tell our customers about how we can help them. It's the stories of the employees and how they work in the roles they play within an organization. And oftentimes data can feel dry. It's like metrics and charts. And what does all this mean? I think bringing the stories to life helps me to understand what's really happening and connect it to a broader context of what's happening with the business, what's happening with my employees, and how can how can this drive an impact.
So let's take an example like the turnover cost, right. Turnover employees are either this the first or second most important resource in the organization. Right? In terms of where your costs are going and how you're running your company. And so you want to understand turnover. But what does that take. We'll have to look at turnover by different cohorts.
And I might have different expectations. People who are working in the store or the manufacturing plant versus the corporate office, it's going to be different. And so I have to be able to segment those groups. I have to be able to pull that data, and I have to be able to compute the cost, like what is there? What is their salaries.
Put that all together. That's a lot of manual work. In order to just to get the insight that you want. This is where having powerful data and AI together brings in controlled environments, allows us to now be able to bring that picture. And but turnover data is like backwards looking, right? It tells me what happened, who left, how much did it cost me?
I also want to look forward. Who's likely to leave. What risk is, is do we have how much would that cost us. And some jobs it will take longer to hire for it given the market conditions. Maybe it's more expensive to hire in the certain locations. I want to factor all that in as I look at the risk.
And so putting that together as a story, so that when I look at it as an organization, I can go backwards looking at what happened. How much did that cost? And I don't have to do all the digging. It's all brought for me, but also forward looking. I have a picture of this is not a certainty, but helps me give me an understanding of what do I need to be on the watch out for, what things can I look for?
And putting that in a story I can understand, speak to my executives about, and make good decisions about what's happening in the organization. Stories are the foundations of every company, and data and AI help to make that easier.
Naomi Lariviere
And I think maybe one other point to add to that is it really if you're doing AI correctly, you're shortening that decision, the distance between that question and the next decision that you have to take, because it's just there. And like if you're transparent and how you're using it, AI can really help you get from my question to my answer rather effectively.
Patricia Stark
Excellent. So Naomi, once a leader decides to act on those insights, how does a tool like the Agent Control Center help leaders determine which agents are enabled, who can use them, and which actions still, though, require human approval?
Naomi Lariviere
Yeah, I think, you know, we've talked about it a little bit throughout this conversation about transparency with AI. And one of the biggest questions we get are what you just asked me, which is like, what AI is turned on, how are you using it? What Agent Control Center really offers our clients is the ability to have that right at their fingertips.
They can understand what agents are turned on in their system. They can see what the agent actually does, what is its goal or its mission like? How does it help me do my job? It helps me also understand if I have that agent turned on who has access to run that agent. Are there special security permissions and individual needs to have to be able to do that?
Naomi Lariviere
And and for a client, that really means they have control over it. It's not something that's happening to them. It's happening with them. Now, the other thing that Agent Control Center also helps a client with is understand what is happening between the agent and what a human actually did in the system. And that's really important for in this day and age with AI, because AI does make mistakes.
And what you want to have a good understanding for is what did Naomi do versus what did this agent do. And I want to be able to see that. I want to be able to understand all of the decisions that made, all the reasons it made those decisions, and who, at the end of the day, made the final call, me or the agent?
Patricia Stark
It's really just like that age old paper trail. We need to be able to track back and see who was accountable, what happened and why.
Naomi Lariviere
Yeah, and especially in this profession, I think it's really important that they have that level of visibility because otherwise, what's the AI doing? You know, what does it have access to? What you know, who's controlling the AI.
Jason Albert
Actions are always taken by something. The world doesn't live in a passive voice. Things don't just happen. And it's important to know when things happen. Who or what was the actor again for that accountability again for that auditing and ultimately for that compliance?
Patricia Stark
Yeah. Absolutely. So Jason, as agents act across HR, payroll and other business systems, what controls are essential to ensure that permissions, purpose, limitations and human accountability travel with that data.
Jason Albert
Right. Well, when you look at AI acting across these various systems, it's important to make sure that it's appropriately constrained so it doesn't engage in unexpected actions. Right. To give a silly example, you don't want an AI chatbot that's designed to help with employment decisions to suddenly give you the best set of predictions for tonight's baseball games. Right?
That doesn't cause a lot of harm, but you can imagine other things that might cause more harm. And so when you really think about this, it's important to have those types of controls. And we have them now, particularly with respect to AI agents with these model context protocols or MCPs that govern what data the agent has access to, how it can act on that data and what it can do.
So at the first instance, it's really about constraining the agent and establishing the limitations of what and how it can act so that it acts in ways that are expected, so that what it can do is appropriately limited and scoped to what's intended. And then you have human oversight of those protocols. You have human oversight, ultimately, of those decisions.
You decide what things the agent can do, what things at work can only make a recommendation. And so to me, it's less about sort of controls traveling with the data. Many years ago in privacy law, there was this desire to perhaps have data coded or encoded with limitations that went along with it. But in this modern world, there's so many different things you could use data for.
It's really about controlling the agent and making sure it's access in the first place. It's appropriately constrained and acting that way, and that's how we control agents and govern them.
Patricia Stark
Very good. Thank you for that. All right. Now we're going to go and do a little repeat and complete round okay. So this will be for all of you. And we can go right down the line and start with you Naomi. Workforce data should always....
Naomi Lariviere
Be paired with context work and human judgment.
Patricia Stark
Very good. Workforce data should always.
Jason Albert
Be used in alignment with the employees expectations and understanding okay.
Patricia Stark
Workforce data should always.
Amin Venjara
Workforce data should always be tied to business outcomes.
Patricia Stark
Excellent. All right. Amin, this one's for you. Humans must make the decisions when.
Amin Venjara
The stakes really matter.
Patricia Stark
Great. Naomi, I would allow an AI agent to act independently if.
Naomi Lariviere
The stakes didn't matter.
Patricia Stark
Very good. Jason, a trustworthy AI product, makes it possible to.
Jason Albert
Gain new insights and be able to act in new ways, while respecting the trust of the people that are impacted by those decisions.
Patricia Stark
Okay. And now this is all again, the most important workforce AI rule is.
Naomi Lariviere
Start with the problem.
Patricia Stark
Excellent.
Jason Albert
Ensure appropriate human oversight.
Amin Venjara
Trust.
Patricia Stark
Trust. Very good. Well, I thank you all for wonderful insight and a fun conversation. And let's do it again sometime soon.
For media inquiries, connect with our media relations team for press materials, interviews and expert commentary or reach out directly to a member of our team.
Stay informed! Sign up for email notifications to receive updates from our Newsroom.
Facebook
X
YouTube
LinkedIn