Recruitment technology has the potential to improve hiring manager decision making, and to improve decision making for internal promotions – and potential ethical pitfalls can be mitigated by interrogating tools before implementation says Franziska Leutner
The potential of decision making tools and algorithms for recruitment lies in their ability to minimise the impact of bias compared to human decision makers. Rather than circumstance and candidate characteristics, decisions are made based on standardised criteria. As a decision making aid, recruitment tools help hiring managers and those deciding on promotions to evaluate all candidates based on the same criteria.
They also help detect skills and attributes that are relevant to job potential that a human interviewer would be unable to pick up on. For example, personality is an important predictor of job performance, but interviewers are bad at picking up on it during interviews.
Recruitment technologies therefore have great potential to improve hiring and internal progression decision making. Here are some principles for applying recruitment technologies ethically:
1Benefits to job seekers: Any new technology or tool deployed in the recruitment process should introduce a benefit to job seekers – in other words, job seekers should be better off if it is used than if it is not, and ideally they will reap significant benefits from this new tool over and above alternatives. It is also easy to understand what these benefits would normally entail, namely the chance to be evaluated based on skill and potential, a more meaningful and rewarding job, a career upgrade, the ability to unleash and grow their potential, as well as the ability to understand their talents and potential more clearly. In other words, are all job seekers (including traditionally underrepresented groups) improving their career prospects because of the new tool that is used to recruit them?
2Informed consent: The second principle is an important booster to the first. Ensuring that there is a transparent transaction with candidates, where they are fully aware of the potential consequences of the recruitment process, and willingly opt-in because they have made the educated and rational choice to be examined or vetted by a potential employer or recruiter, takes care of most of the potential ambiguities. Note, this principle includes the necessity for good choices to exist. If women can choose only between recruitment practices that are all biased against them, there is no meaningful choice. Recruitment technologies increase the presence of good choices in the recruitment market: Standardised recruitment processes such as structured interviews used in video interviewing analytics compare favourably in terms of fairness to the common unstructured interview.
3Confidentiality, anonymity, and data protection: Recruitment technology must protect and preserve the confidentiality, anonymity, and data protection of candidates. The number of people who have access to recruitment information should be limited, and the information should be kept only for the purposes of the recruitment process. Data should be erased or deleted once the assignment is over, and no other parties should be given access to the data later on, unless candidates agreed. Needless to say, this also implies preventing data breeches and hacks, such that any information disclosed during the recruitment purpose is only used for that purpose and ceases to exist after that. If organisations or recruiters feel the need to reuse or recycle these data, it should be justified vis-à-vis candidates’ benefit and approved or chosen by them through informed consent. In some instances, recruitment data can, and should, be examined at the aggregate level. For example, if companies were interested in testing whether X or Y tool predicts future performance, they will want to mine group-level data and should not be interested in identifying individuals. Equally, if companies want to monitor whether their selection process is bias free, they need aggregate level recruitment data.
4Feedback and self-awareness: While not generally a main criterion for ethics, it is ethical to return as much information as possible to candidates, irrespective of whether they were offered a job or not. There’s no significant cost to providing feedback to millions of people if you have done it for one (except of course cloud computing servers). This provides great opportunities for organisations to engage in ethical behaviours with rejected or not selected candidates, explaining how they evaluated them, and why their particular scores or profile was deemed a poor fit with the role, or not as strong as other candidates. To be sure, rejected candidates may always challenge unhappy decisions and question whether the score or desired talent profile was indeed correct, but to the degree that organisations use validated tools, such as science-based assessments (with or without AI), they will also be able to address these concerns and defend their decision. And of course, being rejected for no reason or told that ‘there were others better than you’ is neither satisfying nor ethical, especially compared to a transparent explanation of the company’s rationale.
5Explainability: Recent AI ethics frameworks tend to converge on one important dimension, namely explainability. We could apply this to any innovation in recruitment. Explainable AI or explainable data science imply that we should not just be content with predicting things – and understanding that if you get X you may get Y – but also understanding and explaining things. So, predicting that someone will be a high or low performer if they are hired is useful, but in order to leverage this prediction in an ethical way, we also want to understand and explain why they will be in one category or another. This principle puts a limit to blackbox algorithms that fail to explain the recruitment process. For example, just because academic studies have shown that ‘liking’ curly fries on Facebook is indicative of having a higher IQ doesn’t mean we understand why, or can use curly fries Likes in the recruitment process. In short, we should favor tools, signals, and methods that boost our understanding of what someone is like, rather than simply increasing the probability that they are the right candidate, not least because this will also help us explain our decision to them, and make a rational decision on why they should join or not.
In summary, recruitment technology has the potential to improve hiring manager decision making, and to improve decision making for internal promotions. Potential ethical pitfalls can be mitigated by interrogating tools before implementation: Is this tool improving the process for job seekers, adn are job seekers aware of what it is doing, how and why so they can provide meaningfully informed consent? Is data secure? Are candidates learning from the data they are providing, i.e. do they get feedback? And finally, is it clear how the technology is working and why it is being used?
Franziska Leutner is the co-author of The Future of Recruitment: Using the New Science of Talent Analytics to Get Your Hiring Right by with Reece Akhtar and Tomas Chamorro-Premuzic. Out now, published by Emerald.
