The term “artificial intelligence” is bandied about casually in conversations and sales pitches about HCM systems. But is AI really ready for show time? And what does it actually mean? For some answers about how AI can help with hiring, performance reviews and gauging employee morale, Employee Benefit News spoke with Holger Mueller, vice president and principal analyst with Constellation Research. Edited excerpts of that conversation follow.
Employee Benefit News: Is true artificial intelligence used in any of the HR systems on the market today?
Holger Mueller: The limitation with machine learning is that there must be enough electronic data for it. And it must have an understanding of the data: what is good, what is bad. Sometimes we don’t have that. Take weather data, for example. We know it’s 70 degrees, but is that nice or not nice? It depends on the person. So we need to understand the data, we must know what is good or what is bad, and we need a lot of computing power to build models to predict and make the terms.
EBN: How does that come into play in the systems available today?
Mueller: Take, for example, a CEO search. If the data I have is based on the last search five years ago, it will probably not help me to find the next CEO.
EBN: So it would be different if the question pertained to a topic for which there is more recent and abundant data available. What types of questions might those be?
Mueller: It could be: Who do I put on a particular shift? Who works better with whom? Who am I going to promote? What kind of salary increase should I give? Who am I going to transfer? Who’s the best person on the job internally, or who is the best hire externally? Those are high-frequency interactions where there is lots of data, and that can lead to more reliable answers.
EBN: Is it clear when you have enough data to have some faith in the system’s output?
Mueller: Ultimately, with all forms of machinery, when you test these things to make the next decision, we are very good at figuring out if it’s a bad decision, if it’s an OK decision, if it’s pretty interesting, or, “Wow, this is an amazing decision.” So I wouldn’t say, “Oh, we don’t have enough data because we only hired 50 people last year, so don’t even try to apply that data to the 20 job candidates we have right now.” You can apply it to them, and see what happens.
EBN: And hope for the best?
Mueller: Sort of. Suppose a company tests its system for performance prediction, and the model they had was only 70% accurate. Would you delay its implementation until it reached a 90% accuracy level if today your average manager is only making the optimum hiring decision 60% of the time? There is often a quest for perfection when “good enough” is really good enough.
EBN: Why would I want a system that deploys chatbots?
Mueller: The problem is that clicking on things and finding things in two-dimensional structures is not a human way for us to communicate. We only do this because 500 years ago, Gutenberg invented the printing press, and there was paper, and then somebody invented display, and it’s two-dimensional too, and that’s how we try to extract things. If we want to do something between you and me right now, we talk, we communicate, we don’t press buttons in a menu structured to find out where is the right conversation to have. And that’s why chatbots and natural language processing are so powerful.
EBN: What are some ways one can “chat” with a system?
Mueller: You can say things like, “Trade my next shift” or “Show me the attrition statistics from last week,” instead of figuring out where to click and find the answers. Software has been inhuman to humans, not communicating in the way that distinguishes us from the rest of the animals out there — with spoken language.
EBN: How are advanced systems helping managers with performance reviews?
Mueller: They can crunch the self-performance reviews and you don’t have to read them individually. If you’re a warehouse manager and have 100 direct reports and you are forced to read those performance self-ratings, that’s a huge task. So software can tell you that here’s the sentiment of the group, here’s where you agree, here are the people you have to spend more time with, or less time with, without you having to read all of the typed information.
EBN: How reliable are these systems?
Mueller: There is always room for improvement, but I think on sentiment analysis, the industry has reached a level that is beyond our understanding. And where the same technology is very interesting is when it’s applied to analyzing job posts. For example, suppose you post a job and you want to understand why no woman is applying to this job. It can tell you if it was written using male lingo and women reading it think only men can apply. And vice versa. It can also be used to assess whether text will only appeal to a particular age group.
EBN: Talk about how face recognition technology is being used in HR.
Mueller: In video interviews, systems can try to find discrepancies between a candidate’s words and body language, or assess how comfortable a candidate is when asked for certain job skills, experience, and so on.
EBN: And what about using technology to assess the mood of existing employees?
Mueller: Sure. Employees punch in, punch out, the visual software recognizes you, so why not take your mood, right? I mean the mood setup — smiling, not smiling — is very easy, very low-hanging fruit. Some people might have concerns about Big Brother and Orwell, but, yes, it’s technically feasible.
EBN: Is anybody actually doing this?
Mueller: I’m not aware of somebody doing that, but if they have the information, they could do it, so all your people are recognized by the iris scan, by the facial scan, by the facial parts and getting let into premises, so you have a record. You could go back to the picture and say, “Oh, they looked happy,” or “They looked stressed.” The data is there.
EBN: Tell me about systems that identify employees who should be promoted or transferred to new positions. Is that giving too much ground to a machine?
Mueller: No, it’s absolutely a valid idea. When you think about how people transfer jobs in a traditional work environment, it very often depends upon things like who does the person know, the relationships, and not necessarily an objective assessment of the appropriateness of the person’s skills or experience. Companies struggle to put the best people in the right spot. So the more objectively they can put the right people in the right job at the right time, the better you function as an enterprise.
