More and more organisations are adopting the use of AI in their hiring practices. As this trend continues, there are growing concerns about the consequences of such practices on talent diversity. Platforms like Thinklytic develop their machine learning algorithms by collecting assessment data from current job incumbents. These algorithms are then used to identify which job candidates are most like the current top performers in the organisation. There is an understandable worry that such practices are creating excessive homogeneity in organisations. Surely, if you only hire people like your current top performers, then you will eliminate the advantages of having a cognitively diverse workforce?
At Thinklytic, we understand these concerns but feel they are often overstated and stem from an incomplete understanding of how such approaches really work. Here are some of the key points to keep in mind when thinking about incorporating AI into the hiring process…
Not all diversity is good diversity
Some kinds of diversity are desirable and some are not. In fact, some degree of homogeneity is likely important for many of the traits and abilities that are measured in the Thinklytic platform. Presumably, you wouldn’t want a large degree of diversity in emotional intelligence in your organisation (with some people being very high and some being very low in emotional intelligence). Likewise, it doesn’t make sense to want a large amount of diversity in general intelligence, learning agility, ambiguity tolerance, problem solving ability, etc. Many jobs require employees to have a specific type of homogeneity to be successful. For example, if you are hiring safety inspectors for a mining company, you would be unlikely to seek out diversity in risk aversion. It would be counterproductive to hire a couple of highly risk-prone safety inspectors to balance out the risk averse ones.
The odds are that your top performers are already somewhat diverse
Some forms of diversity are obviously important. For example, it would be a shame to hire only very extraverted people and filter out the introverted candidates who might have a lot to offer an organisation. In most organisations, the highest preforming employees will already have some degree of variance in their relevant traits and abilities. As a result, the prediction algorithms that are developed will take into account the different ways that a person can be a top performer. For example, many top performing engineers may be quite diligent and detail-oriented by nature, while other top performers might be less so but make up for it with their creative problem solving ability. Our algorithms allow for such differences to be accounted for.
We leave out the demographic data
It is important to recognise that machine learning algorithms do not take into account specific demographic variables that would lead to systematic discrimination. For example, we do not include a persons’ gender, age or ethnicity in the algorithms – only the traits and abilities that are unique to the individual. Therefore, it is theoretically possible to train the machine learning algorithm exclusively on white men and then select only Asian women from the candidate pool. What matters is the match between traits and abilities of the top performers and the job candidates – regardless of what demographic groups they may belong to.
You’ll actually identify more candidates who don’t fit the typical bill
Many hiring process already lead to homogeneity. People tend to hire people who are like themselves. However, they tend to make these decisions on very superficial and uninformative criteria. The data that our machine learning algorithms leaves out is just as important as what is included. There may be great job candidates out there who don’t necessarily have the “right” degree, background, experience, demeanour, interviewing style, etc., but who would still be a great fit for the job. AI algorithms enable organisations to identify people that do not necessarily seem to be a great fit on the surface but who have all the traits and abilities to succeed in the role.
AI Prediction scores are only one source of information
No assessment method is going to be perfect. Every method will have some limitations. Human beings are extremely complex and it is unlikely that we will ever be able to make perfect predictions about how a job candidate is likely to perform in a given role. However, we believe that more information is almost always better. If you have concerns about homogeneity in your organisation you can always choose to include other information in your hiring decision in addition to AI prediction scores. For example, if you feel like the top performers your organisation are too averse to ambiguity, you can look for people who might have slightly lower AI suitability scores but have higher levels of ambiguity tolerance.
The concerns about AI algorithms in hiring are understandable. And while such concerns are not completely unfounded, they are often overblown and based on misunderstandings. We feel strongly that the advantages of using machine learning in selection far out way the disadvantages. Machine learning offers a dramatic improvement to selection over more traditional approaches. If you’d like to find out more about our use of AI technology, please contact us at firstname.lastname@example.org.