The misconceptions around AI and unconscious bias in recruiting
AI is not the save-all panacea for unconscious bias in recruiting. It will certainly help management of the process, but human oversight will still be necessary.
There is much misconception around AI and unconscious bias in recruiting. One post from Ideal leads with the grabber headline “How AI Can Stop Unconscious Bias In Recruiting.” Although the post itself is more balanced, the headline is misleading. AI can’t be eliminated it can only be managed. AI won’t stop unconscious bias in recruiting, it can only support the management of the recruitment process and flag up patterns, limitations and issues.
AI potentially supports the recruitment process in two ways:
- AI makes sourcing and screening decisions based on hard data points
Recruiting AI parses huge quantities of data and extracts information based on the parameters laid out for a specific search for the ideal candidate via ATS (Applicant Tracking Systems). It will be bias free, based on hard data and generate candidates who match the specified criteria, regardless of any of the biases which we might apply if we if we did this ourselves. But these data points are imputted by human beings.
- AI can also be programmed to ignore certain data
Bias loaded data such as gender, age, place of birth, post code, university attended and address all of which can be by-passed and excluded. This means that the dodgy or preferred zip code, the university with a poor reputation or an excellent one, age and gender can be pre-programmed to be excluded.
AI can exhibit bias
However studies suggest that AI can exhibit bias especially when dealing with text which is generated by people. Which it always is. We are essentially confirming our own inherent biases when we programme the search. An article in Science quotes research from Joanna Bryson, a computer scientist at the University of Bath in the United Kingdom and Princeton University says.
“Don’t think that AI is some fairy godmother, AI is just an extension of our existing culture.”
The post shares that to test for similar bias in the “minds” of machines, Bryson and colleagues built a test which analysed word-embedding association (WEAT).
They started with an established set of “word embeddings,” basically a computer’s definition of a word, based on the contexts in which the word usually appears.
The “mind” of a computer will be influenced by the cultural settings in which the data is archived. So if your AI is programmed with search strings which include engineer, Director, President, Executive, Manager, Leader which are male coded words, they are therefore still going to generate the same level of male applicants as previously. Unitive’s technology focuses on hiring, with software designed to detect unconscious bias in Applicant Tracking Systems, that read resumes and decide which ones to keep or scrap, based on certain keywords which is a step closer to controlling unconscious bias in recruiting.
Unconscious bias is big business
Unconscious bias is now big business, with a host of companies trying to break into the market. This is perhaps the root of exaggerated claims. Research from companies such as Textio and Kat Matfield offer software to identify gender bias in job adverts and profiles. Kanjoya’s technology, with linguistics experts from Stanford University, established that words such as “assertiveness” are used to describe women are perceived negatively in reviews, but are received positively when evaluating men. Kanjoya claims to identify subtle differences in human emotions and intent in language. Apps such as Blendoor hide a candidate’s name, age, employment history, criminal background and photos, to allow hiring managers to focus on hard data only. Interviewing.io’s platform can mask an applicants’ voice. Automated interviews are becoming increasingly common place to reduce the unconscious bias in recruiting.
The human element can’t be ignored
“The ugly side of technology is that it has dumbed recruitment down. It has weakened the human skills, which made being great recruiters so special. So many recruiters today hide behind technology, and avoid real interaction. They miss the chance to ‘influence’ the outcome of all those critical interactions. They concede the part of the job machines can’t do!”
The reality is that recruiters need to become bias conscious and create a bias aware process where checks and balances are embedded to safe guard both the process and the result. AI is set up to identify patterns of behaviour, but it means that any human bias that may already embedded in an organisation’s recruiting process – even if it’s unconscious – can be absorbed by AI. We need to have recruitment processes which are based on a culture where it is acceptable to challenge bias and perception. We all need to be called out about the biases we hold from time to time, and for each one of us it will be different.
Recruiting AI software will test the process for any inherent bias by evaluating any search results generated, to establish if any specific patterns emerge. This will give any recruiter the opportunity to self-correct and adapt specific elements of the process. But it will still require human input and supervision. Unconscious bias in recruiting will probably never stop, but it can be better managed. To suggest otherwise is currently nothing short of nonsense. While the market is competitive we are likely to see a number of organisations making exaggerated claims with more click bait headlines.