One decision about your workforce can have a massive impact on your business—whom to hire, whom to promote, and when to cut back—can be the difference between a successful and unsuccessful business venture. With so much pressure to make the right workforce decisions, HR teams are starting to use Artificial Intelligence (AI) to guide decision-making processes in order to increase accuracy and efficiency, while decreasing the risk of error.
For example, AI can be used within an HR department to identify a pool of applicants that are most qualified for a job, without letting biases about a person’s age, gender, ethnicity, or background influence the decision. Or it can analyze payrolls within an organization to spot salary discrepancies to help close the gender pay gap.
And while AI systems can identify a solution faster than the human brain, how much of the decision-making process should be delegated to AI? Are there some decisions that should always be reserved for humans? My father Earl Hunt did pioneering work in AI and noted, “We discovered pretty quickly that computers aren’t very good at mimicking how the human brain actually works. But they are very good at solving complex problems that the human brain could never solve.â€
So this all begs the question: What’s the difference between letting a computer or a human call the shots? Well, let’s take a look at what those differences might be, and what they could mean for your business:
What kind of problems we’re looking at
Let me start off by saying, AI renders numerable benefits—like tackling the complex problems my father described, but humans have one thing that AI doesn’t: adaptability.  We’re naturally good at identifying problems worth solving, and there is no problem we cannot meaningfully think about even if we cannot always solve it. In contrast, AI can only solve problems that it is told to solve by humans. More importantly, it can only effectively address problems that meet certain specific conditions, which are, interestingly enough, the kinds of problems that humans commonly struggle to solve.
How and why both approach problems differently
AI and humans not only solve different kinds of problems, but they also approach solving them very uniquely as well. Humans are adept at developing solutions without fully defining the problem, which makes us amazingly flexible and considerably prone to errors. For AI it is the opposite because the first requirement for the use of AI is to have a precise empirical definition of what you are trying to model, identify, or predict. For example, in order to use AI to predict employee performance, you must first define what you mean by “performance†in highly measurable terms.
That would never happen if a human being were to look at a problem. That’s because humans are comfortable making decisions based on very small amounts of data, since we find it hard to process large amounts of it. Sometimes, we can take that small amount of data and make intuitive impressions that are amazingly accurate, while other times they are totally wrong. In contrast, most AI methods require very large datasets containing matched pairs of “predictor data†and “criteria data.†For example, to model work characteristics causing employee stress, you would need predictor data that measured characteristics that might cause stress, as well as criteria data on these same employees that measured stress levels. Not to mention, you’ll probably need data on thousands of employees for an AI program to create an accurate model.
AI is very effective at sorting through large datasets to find small bits of meaningful information, but it cannot find meaningful information unless it is reflected in mathematical patterns in the data. Of course, AI is amazingly good at finding treasures of useful information in massive piles of garbage data, but it can’t find treasures in data that is entirely composed of garbage.
AI is also a powerful problem-solving method for situations where outcomes have precise empirical definitions, there is access to large matched predictor-criteria datasets, and small but meaningful relationships exist between predictor and criteria data. Thanks to the amount of data made available through human capital management (HCM) technology, we are finding more and more situations that meet these conditions. Â Examples include modeling relationships between applicant characteristics and post-hire retention, job characteristics and employee turnover, and employee work characteristics and absenteeism and healthcare costs. But there are many HCM situations that do not meet and may never meet these conditions; for example, using AI to predict notoriously poorly defined outcomes like performance and potential.
It is important to know when AI is likely to work and when it will not work. The kinds of problems AI is good at solving are usually the kinds of problems humans find difficult or impossible to solve, and vice-versa. AI should be positioned as a tool to support better decision making, but it cannot be a replacement for human intelligence.
Why it’s important to define a difference between the two
Anyone who has read science fiction knows the danger of turning over too much power to machines. Though both man and machine solve problems in different ways and offer unique benefits when it comes to problem-solving, many people harbor concern about computers making decisions that determine what they can do in life. This includes decisions that impact our employment and careers. Saying a company is making HCM decisions using AI could create anxiety for many employees and candidates, so it’s important to draw a distinction between the two, and show the value each bring to the decision-making process. Start by avoiding using words like AI that people associate with science fiction, and instead used terms that are more boring yet more accurate such as “mathematically developed models.â€
Manage expectations around artificial intelligence in HCM
AI is about using complex mathematical techniques to address specific types of HCM problems that impact organizational performance. These mathematic models do not work everywhere, and sometimes there’s just nothing that can replace the human touch, but in the right situations, they work extremely well. So, in order to be able to successfully leverage by AI and human thinking during the decision-making process, it’s important that you first and foremost set expectations. Avoid hyping up AI, and focus on educating your business on how to combine complex math, sophisticated computer programming, large data sets, and of course, the human brain, to improve certain types of HCM practices. That way, both resources can be used to their potential, and you can make the best decisions possible for your business.