Artificial Intelligence & Machine Learning: The Good, the Bad & the Ugly

For the past 5 years or so, big data, human resources (HR) analytics, and predictive analytics have been new concepts floating around the HR world. Of course, with technology changing at a mind-blowing speed, two new approaches are creeping into the HR profession – artificial intelligence (AI) and machine learning. Let’s look at the benefits and pitfalls of these new techniques.

AI is the movement toward “smart” machines and computing systems being able to carry out tasks the way that humans would, except much more efficiently¹ – think surgery-performing robots, self-driving cars, even the filter that sends suspected junk mail to your spam folder.

For example, IBM Watson Talent Insights uses predetermined algorithms to find patterns in the data that emerge automatically, without the need for a human to make predictions about potential relationships to look for in large datasets¹.

Machine learning is an application of AI applied to the data analysis processes¹. Developers create algorithms (i.e., math equations) that can be applied to data to make “smart” decisions and arrive at specific conclusions. Developers tell the algorithm what to look for and what to do with the information, and then the algorithm completes analyses without further instruction from the developers.

Google uses machine learning to promote hiring and retention initiatives by using algorithms that predict which candidates are most likely to succeed in their new role after being hired and which employees are likely to want to leave the organization in the future, respectively².

AI and machine learning techniques can provide great value for HR departments because they can eliminate the need for human processing of rote tasks and mitigate the risks of human error or boredom in things like data entry, resume screening, etc.³ These techniques may be especially interesting and appealing for organizations that do not have the capacity or capability to conduct in-depth analyses on their own and are looking for quick “data-based insights.” However, without a proper understanding of the analyses behind the scenes and of organizational and employee behaviors, blind algorithms can have potential pitfalls.

No One-Size-Fits-All Algorithm
There is no “one-size-fits-all” algorithm that can be used for processes across organizations. Relying on a single algorithm, or even collection of algorithms, to take into account the dynamic features of an organization is risky. A variety of factors are important in understanding employee experiences at work, and algorithms developed at one point in time, or for one organization, may only provide accurate insights for that timeframe or context.

High Correlations: Flawed in Predicting Behavior
Organizations should approach talent management with theory-driven ideas about factors that affect the employee experience. Most machine learning algorithms look for correlations to “predict” variables of interest. Herein lies the problem – algorithms based on finding high correlations are inherently flawed in predicting behavior; correlation does not imply causation AND a correlation is not a prediction as it implies no order of occurrence between two variables. For instance, an algorithm designed to monitor turnover risk could find that people whose names begin with “M” are likely to quit after 3 years. However, anyone can easily see how this is illogical and happenstance. There could be some other factor that isn’t captured in the data that is causing these people to quit after 3 years – the fact that all their names begin with “M” is a coincidence that the algorithm interpreted as an important pattern. Let’s not forget that part of our role as HR professionals is to mitigate risk (e.g., discrimination), and a purely machine-driven approach introduces significant risks for an organization.

The Human Element
Organizations that rely too heavily on AI or machine learning are at risk of removing the expert human element from the equation. Acting on insights gained from algorithms may cause organizations to ignore the how and why these relationships exist in the first place. Skilled and trained researchers are necessary to making these algorithms work for your organization. A trained practitioner, hopefully an Industrial/Organizational psychologist, should develop an algorithm using your organization’s data, to account for your organization’s context. Or at the least, an expert should be closely monitoring and interpreting the elements going into and out of an automated system to ensure accurate and appropriate conclusions are generated.

The HR industry is just beginning to latch onto and understand analytics and arguably, most HR departments are not well-advanced in this space. Making the leap to AI and machine learning could be a potential mine field for organizations that do not have a solid analytic foundation.