We are probably years, and maybe decades, from real AI drastically impacting HR’s day-to-day operations. However, there is potential to leverage it for impact, safely, now. This article outlines the safe places and danger zones for utilizing AI, writes Shane Douthitt, Co-founder & Managing Partner, Strategic Management Decisions.
We hear a lot about artificial intelligence (AI) and machine learning these days. So, what is real and what is hype in the world of HR? Like anything new, the potential is exciting, and the risks are scary. Let’s explore both to understand the current state of AI in HR.
The potential for AI is to help employees and managers make faster, better, more objective decisions. Consider all the promise in the candidate selection/hiring process. If a computer can screen resumes, administer assessments, and apply algorithms that predict performance and/or turnover, imagine the potential for reducing turnover and minimizing those bad hires. This sounds pretty amazing. Now let’s explore the reality.
You may have heard about the Amazon AI recruiting tool that showed bias against women. This failure became public knowledge in the fall of 2018. Basically, the machine learning platform analyzed 10 years of resume data in a male-dominated industry and “learned” to reward male resumes over female resumes. But, isn’t AI supposed to be more objective than humans when making decisions?
Before the Amazon failure occurred, I predicted that machine learning and AI in HR would fail before they would succeed. The prediction was based on the following:
Machine learning has significant limitations from an analytics perspective – for example, correlation does not imply causation.
A machine has difficulty applying context (interpreting analysis).
Predicting human behavior is extremely difficult.
Taken together, these reasons explain why a machine would screen applicants in a biased manner. We know certain fields have been male-dominated. The analytics methods would find this spurious relationship and would be unable to interpret the findings (i.e., apply context) as not relevant to the hiring decision. Therefore, this outcome was very predictable.
This example illustrates the inherent risks with AI in HR. Does this mean we should just forget about AI all together? Of course not. There are still valuable applications. We just have to understand the risks and mitigate them in smart ways.
Dominic Holmes, the partner at Taylor Vinters, says, “In my view, the answer lies in the end-users of AI solutions working together with those who create them. Machines have the potential to make more objective, consistent decisions than humans. They can be more reliable, more accurate and work 24/7 if needed, without getting tired or distracted. However, they are not fool-proof and humans may still be required to intervene and manage any unintended outcomes.”
I agree with Dominic’s recommendation. That’s because he’s suggesting human beings be involved to apply context and overcome some of the analytics’ limitations of machine learning and AI.
Let’s apply this rationale to action planning. The action-planning functionality within an organization’s employee survey technology could use AI in the form of an expert system (not machine learning) to prioritize and provide action-planning recommendations based on employee survey data. In this scenario, components of the decision-algorithms are determined by predictive modeling conducted and interpreted by humans (ideally Ph.D. consultants). This critical step allows human beings to apply context and overcome some of the analytical limitations of machine learning. An expert system is not as flashy as machine learning, but it allows organizations to mitigate the risks outlined previously.
The benefit? AI-enabled action planning ensures managers are making the best decisions possible by telling every leader exactly what to work on and what to do about it. In short, the functionality pre-populates a plan with proven actions. This makes decision-making as easy as possible – no analysis paralysis or expecting all managers to be Ph.D. experts. The tool does it for them so they can spend time and energy on making improvements. This increases the impact of common HR practices, such as the employee engagement survey, and safely leverages the power of AI.
Additional HR activities where AI has significant promise are as follows:
Training & development applications such as personalized learning or AI-driven training recommendations
Benefit administration applications such as AI-based automation that can ease plan administration, employee decision-making (e.g., chatbot FAQ’s), and plan implementation
Recruiting applications such as “AI recruiters” to automate scheduling interviews, provide candidate feedback, and answer candidate questions (e.g., recruiter chatbot).
These types of applications can bring value to HR functions through automation and improved decision-making. Although none of these examples are transformational in nature, they do provide a meaningful impact.
As you or your department start down the path of leveraging AI in your HR activities, ask yourself these questions first:
Are there any typical HR legal risks based on the decisions being made (e.g., adverse impact, gender bias)? If so, how can those risks be mitigated in AI decision-making?
How will managers likely use the information and does the application create any legal risks? For example, if an AI tool provides a turnover risk score for individual employees, can this impact how a manager treats him/her? One can easily imagine a manager not considering an employee for a promotion or pay raise as a result of an employee having a high turnover risk score (due to a feeling of betrayal regardless of whether the employee is truly considering leaving).
How transparent is the vendor with the decisioning algorithm and/or process? If they won’t provide details or a validation report, then you cannot trust the underlying decisioning process – especially if it involves a decision with significant legal risk (e.g., hiring, promotions).
What are the trade-offs of using AI versus not? More precisely, what does the organization gain (e.g., faster-hiring decisions) versus the potential risks (e.g., legal risk from biased algorithms)?
As an HR practitioner, what does this mean for you? This is another place where a healthy dose of skepticism is warranted. The truth is we are probably years from real machine learning and AI drastically impacting our day-to-day operations. We should be open to this new technology and its potential, but we must also apply healthy skepticism and ask the hard questions as we continue to improve our field.