Artificial Intelligence is rapidly entering most of the internal and external processes of companies, to increase efficiency, speed up operation, reduce human errors, implement data driven decision making. It is increasingly playing an important role also in the human resources department, particularly in the hiring process, to screen CV and select the most suitable candidates based on several desired factors.
However, for people who are not software engineers, and sometimes even for them, AI remains a ‘black-box’: once trained and implemented, it is very difficult, if not impossible, to trace back all the processes that led to the final decision, making it very hard to spot errors, patterns and biases.
“Is a résumé screener identifying promising candidates, or is it picking up irrelevant, or even discriminatory, patterns from historical data? Is a job seeker participating in a fair competition if he or she is unable to pass an online personality test, despite having other qualifications needed for the job?”
To answer this questions, Julia Stoyanovich (Wall Street Journal, 22/09/2021) proposes an innovative and kind of funny solution, starting from something very common: nutritional labels.
Firstly, exactly as it is in every product we buy, a job description should list all the ‘ingredients’ the optimal candidate should have,such as degree requirements, specific skills and the years of experience, and how the assessment will be carried out, so all the candidates can know in advanced what the company is looking for. Then, if the candidate is rejected, the AI would use this label to show the candidate which criteria he did not meet (or met less than other candidates), to show exactly what drove the decision. This I believe is the most innovative part of Julia’s solution.
The label could also provide actional information, allowing the candidate to modify some data if needed, and inform them of the possibility of making changes for the assessment methods if they discriminate them (a blind person could ask for an alternative method to submit his application instead of a video interview, where eye contact is a requirement).
The rejected candidate would therefore get a “decision label” along with the rejection, showing how their qualifications measured up to the job requirements; how he/she compared with other job seekers; and how information about these qualifications was extracted.
Until now the focus was only on the candidate’s potential benefits, but this solution would also give employers vital information: since AI’s judgment calls are opaque, employers often don’t know what data are used for screening, and how they are analyzed to come up with the final decision. The labels can show managers the factors that the AI is using to screen candidates and let them decide if those factors need to be changed.
Hiring process is complex: there are several steps and tradeoffs between objective criteria and subjective factors.
AI could help alleviate some of this complexity, but it is important to not forget that this tool works to specification, and these specifications must be clear. We can use AI effectively to identify clear requirements-based matches, but it cannot exercise discretion or apply subjective judgment. These labels may actually help identifying where the automatic and impersonal screening can be used, and when it needs to leave space to personal decision of human resources.
Reference( https://www.wsj.com/articles/hiring-job-candidates-ai-11632244313?mod=management_lead_pos5 )
