What you need to know about AI candidate screening
The 21st century has seen AI (artificial intelligence) taking over almost every aspect of our daily life. Andrew Ng, the founder of Google Deep Brain called AI the “new electricity”, emphasizing its ability to transform major industries like electricity did 100 years ago.
And just recently, AI showed to the world another one of its capabilities; screening candidates for job interviews. It works by assessing resume data and skills of applicants to deem if they are a match for the job. It proved more effective compared to the traditional Applicant Tracking Software (ATS) which required manual intervention and inputs.
But is the idea truly sustainable? As recruitment is both a personal and professional thing, it has raised many technical and ethical questions; all which now require a clear explanation as it involves jobs and livelihood.
Using AI to find candidates
Ask any HR manager or recruiter how they feel about screening candidates, and you’ll get one common answer: it’s tedious. Entelo’s 2018 Recruiting Trend Report noted that hiring managers spend an average of 13 hours weekly on screening candidates for just one role. Seventy-percent of the survey’s respondents agreed that productivity can be increased by automating the screening process.
The ATS software did little to improve those numbers either as it still required manual data inputs and filters, meaning that recruiters still need to source for candidates themselves and refer to hiring managers for advice.
Instead, offering the potential to learn the type of talents that the company needs, and look for them automatically from the applications that come in, AI certainly seems like a lifesaver for many.
AI’s machine learning ability allows it to learn and get more proficient as it repeatedly does the task. Furthermore, it can also be programmed to be legally compliant and non-discriminative, so it avoids bias related to demographics (age, ethnicity, language etc).
One successful example of AI candidate screening is when IBM deployed an AI screening system in 2013, where the company claimed it managed to reallocate 80 percent of staff in a business arm that was forced to close down.
The glaring difference between both ATS and AI is what drives stakeholders towards the latter. Many companies are already opting to automate their screening process and with the wide variety of AI-based HR tools available, it’s not difficult to find the perfect tool for every company.
AI screening drawbacks
Being a human-made system, the AI studies data that’s input into its database to make decisions— and it is designed to analyze patterns. Thus meaning, if it is fed with biased data, or if it’s not programmed to exclude certain physical traits like race and gender, it becomes a biased machine that disregards fair treatment.
Take Amazon’s AI screening system, which was recently discovered being discriminatory towards women because it was being fed with resumes from a 10-year long period, allowing it to build patterns based on past biases. Amazon has scrapped the system after learning of the serious flaw.
Another underlying issue with AI screening is the accuracy it provides. AI systems are always learning and improving with data, making it prone to exclude good candidates simply because the system can’t find the information it’s looking for.
The ATS generation of candidates knows how to trick systems like these by including keywords into their resume, which will prompt it to recognize them as “potential candidates”. Good candidates who do not know how to work around would miss their chances just because they explained themselves differently than what the system has learned.
Some systems even try to judge a candidate’s personality and skills using algorithms by extracting data from their social media activity and past work. These tools potentially limit the talent pool as it might downgrade suitable candidates based on what it learns about them online.
It’s still far from perfect
While AI seem to show a lot of potential in employee screening, it still has a long way to go if it is to mimic the exact same thing that a recruiter would do. It definitely simplifies the screening process, but it certainly cannot yet work without supervision.
31 March 2020