Machine learning (ML) is the study of computer algorithms that allow computer programs to examine data, reveal patterns within data, and improve understanding of data trends through experience (Mitchell, 1997). It is a branch of artificial intelligence (AI) and just one of the many ways to show how computers can behave like the human mind.
In the HR space, ML methodology is one of the newest approaches for learning about employees and job candidates. For example, organizations are using these methods to predict turnover, recruit job candidates, and auto-score interviews. Here is a simplified explanation of how a hypothetical staff-member, Jamie, might use supervised machine learning methods to predict turnover:
Jamie collects a bunch of information from employees – how far they live from work, how happy they are with their pay, their level of commitment to the organization, etc.—and whether the employee left the organization or not. Jamie makes a few decisions about how the computer should handle the information (e.g., selecting a random forest algorithm) then feeds the information to the computer. The computer runs over each example, looking for patterns and learning similarities between employees who did and did not turn over. The learned pattern lives in the machine learning model. Jamie takes that model and applies it to a new set of employees – except this time, leaves out the turnover result. The computer uses what it learned from its previous experience with employee information to predict whether these new employees will stay or leave the organization. If the model generally predicts correctly, Jamie can reuse it on separate groups of employees to predict turnover.
Recommendations for Integration
Although ML can certainly enhance HR processes, it is not a golden ticket that can solve all our problems or replace a well-developed system. Instead, think of it as an added measure that supplements an existing process. And since that measure will impact important decisions, it must be held to the same standards of excellence as any other personnel-related assessment. That means scrutinizing the methodology and rigorously testing the quality of the measure while applying theoretically sound logic along the way.
Below are questions you should be able to answer before using ML methods:
Does the method make sense for our goals?
Implementing ML methodology for its own sake – or because it’s trendy – is not necessarily valuable and can actually create problems when used haphazardly. It is important to fully understand your existing HR process, areas for improvement, and access to relevant information before deciding whether ML is a viable solution for your business. Speak with subject-matter experts who understand your business and can best advise you on appropriate means to achieve your goals.
Does it fit with our current, high-quality systems?
There is not enough research to definitively say how results from ML models should be used to inform personnel decisions. Therefore, it is safest to integrate findings into a larger system. The results should serve as an additional data point that make an existing process more robust. Paring the model results with separate assessment results, for instance, will likely be more telling than model results alone.
Is the method transparent?
When using ML methods (or any components of AI), be transparent about its usage. Explain what stakeholders should expect, how the information will be used, why you trust the model results, and what it means for the business. This will alleviate concerns, questions, and/or uncertainty about the experience.
Is the model biased?
Though ML models may seem unbiased, and even impersonal, they are not. The models are built by humans and humans have inherent biases. If you include data from measures that are biased, the model’s results will be biased. If you allow the machine to learn from data that is not representative of all employees or job candidates, the model will not be impartial when applied to a more diverse group of people. It is important to continuously look for evidence that human biases have crept into our methods, models, and results. Then, eliminate them as completely as possible.
When implementing any tool into your HR system, you must determine whether the tools are empirically sound, rational, and fair. Take steps to discover whether ML methods 1) are right for your intended purpose, 2) complement your current system, 3) are clear to stakeholders, and 4) are largely free from biases. From that point, you can more confidently deploy cutting-edge technology and ensure you are gaining and maintaining the best talent for your organization.
Source: Mitchell, T. (1997). Machine Learning. McGraw-Hill, Inc. New York, NY.