Skip to main content

Diversity in the workforce – Artificial Intelligence (AI) comes to the rescue

3 Tools to Improve Industry Diversity & Inclusion

According to a study published by McKinsey & Company, every 1% rise in the rate of diversity is associated with an increase in revenues of between 3% and 9%. We all know how important it is to have diversity and inclusion in the workforce, and not only for financial reasons. Yet in the US, some 97% of companies fail to reflect the demographic composition of the country in their senior leadership and workforce. 

Most companies want to increase diversity, but unfortunately, the talent sourcing tools that are available today are highly limited, and even discriminatory. There are several reasons for that: 

  • Limited talent pool - Traditional talent sourcing tools (like LinkedIn) have a limited reach to talent. Candidates are often active on multiple sites and leave important data, to which single source tools lack access. Diversity specific job sites and resume databases are limited to active job seekers, eliminating passive, qualified, and diverse talent.
  • Discriminatory search techniques – most talent sourcing tools use an antiquated search technique called Boolean (keywords). This technique is highly discriminatory because it favors candidates who put the right keywords on their resume. For example, some research shows that woman tend to write resumes differently than men. They may provide descriptions rather than measurable numbers or specific skills. These differences may impact the way women are found by search tools and the way they are evaluated by HR and Hiring Managers.
  • Discriminatory profiling - some talent sourcing tools allow you to filter candidates by gender and ethnicity. Filtering by a specific gender or ethnicity may be considered discriminatory because it favors diversity candidates over other candidates.
  • Discriminatory job requirements – unconsciously, job requirements and job descriptions may be discriminatory. People tend to hire people with a similar background (e.g. same university, same social background), or unintentionally have requirements that limit the participation of women and minorities in specific jobs.
  • Photos and names create bias - social sites like LinkedIn encourage you to add a photo and to expose your name. These items may indicate gender and ethnicity and therefore may create bias during the candidate selection process. 

One could assume that AI and Machine Learning (ML) technologies would help recruiters in addressing these challenges and make the candidate selection process completely blind to gender and ethnicity. However, several studies reveal that AI algorithms, if not designed properly, can be discriminatory. If they mimic human behavior, they can also mimic discriminatory human behavior. 

Even using facial recognition algorithm to identify gender and ethnicity, may be problematic. According to a New York Times article, “Algorithms falsely identified African-American and Asian faces 10 to 100 times more than Caucasian faces, researchers for the National Institute of Standards and Technology found.” Moreover, given the subjectivity of the interpretation of visual appearance, it can be inaccurate and problematic to have an algorithm determine ethnic association. 

The easiest way to eliminate bias in the selection process is to eliminate photos and names from the profiles, which most tools, like LinkedIn, do not do. 

The second option is to push diversity candidates up on the candidate “hit” list, either by filtering them out or by giving diversity candidates extra points that will prioritize them over other candidates. This option is problematic because candidate selection is now biased against non-diversity candidates and may result in the selection of candidates that are not the best fit for the job. 

This is where AI and ML come to recruiters’ rescue. AI and ML have opened tremendous opportunities in increasing diversity participation in the recruiting process and in minimizing bias, without lending an artificial advantage to specific talent pools. Here are some examples: 

Using AI to expand and diversify the talent pool – the size of the talent pool is critical in enabling greater participation of diverse talent. AI enables sourcing of talent data from multiple sources and delivers exponentially more talent, that otherwise may be missed. 

AI and ML also help in expanding the talent pool by uncovering talent that Boolean search misses. AI and ML automatically create highly sophisticated searches, composed of hundreds of keywords that are practically impossible for humans to enter. For example, they can add many more relevant job titles as search parameter, based on an analysis of millions of profiles. Thus, candidates that have rare job titles, but the right skills, will be included in the search results. 

In addition, AI adds skills that candidates fail to update on their profiles. This method increases the chances of candidates of all genders and ethnicity to be found by recruiters. 

Using AI to identify gender and ethnicity - AI and ML can be used to classify candidate by gender and ethnicity. The algorithm can use a multitude of elements to make that determination, such as photos, names, country of birth and activity on specific social sites. The difference between AI-powered methods and traditional methods is that AI-powered methods are not used to search or filter by gender or by ethnicity. They are used merely to recommend changes to the job requirements, as described below, in a way that will increase the participation of diverse talent. Under this method, gender and ethnicity classifications are never disclosed to the recruiter. 

Using AI to increase diversity participation – In addition to expanding the potential talent pool for jobs, AI uncovers profiles of diverse talent that traditional, Boolean search methods fail to uncover. 

We know that women and minorities may have profiles that contain less relevant information. For example, they tend to enter fewer skills and fewer details regarding their experience. Using predictive analytics, AI will add “Derived Skills”. These are skills that are accurately “predicted” by AI algorithms. Derived skills will be added to all candidate profiles. However, women and minorities are likely to benefit more from that capability, and as a result, will come up in searches that they would otherwise not come up in, had Boolean search methods been used. 

The most exciting capability of AI and ML in increasing diverse candidate participation in the recruitment process is a “recommendation engine”. This tool will recommend to the recruiter specific changes in job requirements that are likely to increase diverse talent participation. For example, the “recommendation engine” can suggest an increase in the radius of the acceptable locations of potential candidates, a change in skill requirements, or years of experience. 

The recruiter or hiring manager can accept specific changes and immediately increase the percentage of diverse candidates in the top 100 matching candidates for the job. This method is particularly effective because it gives all candidates, regardless of their gender or ethnicity, an equal opportunity to be considered. The job requirements are changed for all candidates, but the percentage of diverse candidates is increased. 

AI can also give recruiters an insight on how many diverse candidates are in the talent pool for similar jobs in the market. In some professions (e.g. nurses), there may already be many women, and there may not be a need to make any changes in the job to increase the number of women considered. However, the changes suggested by the AI software may enable an increase in the percentage of Black candidates. 

AI and ML provide exciting new opportunities to make our workforce more diverse and provide all candidates true, equal opportunities. 

Popular posts from this blog

Why Men Still Get An Unfair Advantage Finding New Jobs

A large scale research of 10 million candidate profiles conducted by Talenya reveals why men still get an unfair advantage finding new job opportunities.   It’s 2020, and women are still being paid 79 cents to the dollar as compared to men, with women making up only 40% of managers. Deep structural problems exist, starting with the hiring process. The problem is twofold - arising both from the way candidates are recruited as well as the presence of implicit bias when viewing candidates. It isn’t that there aren’t qualified women in the pipeline, it’s that recruiters find and accept candidates who write resumes and professional networking profiles in a certain way -- the male way. Discrimination based on how people write their resumes is apparent in both human recruitment and improperly trained AI-powered recruitment systems. Amazon scrapped their recruiting AI after finding that there was gender bias, specifically related to certain language used by male candidates as opposed to fema

The Limitation of "Boolean" in Talent Sourcing

I did a search on LinkedIn for a “Java Software Engineer” in New York City. I entered that job title as a keyword (under Job Titles) and LinkedIn suggested that my talent pool was 2,059 candidates. Then I added a skill and my talent pool decreased to 1,956 candidates. When I added another skill, my pool increased. This is the nature of Boolean search. Every candidate that has at least one of the requirements is brought up in the search results. If you want the skills to be additive (X and Y), you need to write a compound Boolean search string rather than just adding the skills from the LinkedIn menu. I wanted to reduce my targeted talent pool and added “years of experience” range. The pool tanked. The same happened when I added “education requirements”. It was not clear what I should do at this point. I didn’t know what to change in my search in order to maximize my pool while maintaining the quality of the candidates in my search results. Was it a specific skill or the combin