Is Recruitment on the Brink of Extinction?
Last week the Chinese tech giant Baidu published a study in the TMIS (Transaction on Management Information Systems) Journal titled: Person-Job Fit: Adapting the Right Talent for the Right Job with Joint Representation Learning.
It's a bit of a mouthful (and the abstract isn't much better), but the long-and-short of it is that their teams have created a tool that utilises cutting-edge tech to match jobseekers with job listings.
In other words, they made a digital recruiter.
Right now, the technology is just at testing stage, but it's already looking pretty powerful. It uses a specific neural network model that they've dubbed the PFJNN (Person-Job Fit Neural Network). This doesn't just match keywords together, for example looking at a resume and seeing "Java developer", then looking for job ads with "Java" and "developer" in their title/description.
Instead, it actively analyses the language of both the resume and listing, looking for common trends that infer greater meaning – what they call "semantic representation". An example they give is that the combination of phrases "product development procedure" and "documenting" within a listing infer that the role will require a person with high levels of program management experience. Rather than simply searching for someone who has "documented" something in the past, the PFJNN can therefore target a specific range of job titles instead. It's that logical inference that sets the concept apart from previous automation attempts, letting it find skills that aren't necessarily a one-to-one match.
It isn't all roses, however, as a neural network is only ever as good as the data that it has access to. The team at Baidu openly admit that there are currently major problems around education-level, for example, because most roles they trained the PFJNN on asked for "bachelor degree or higher", leaving little room for nuance.
Still, if the main concerns are simply over data-set limitations, then over time these will become increasingly less relevant. That is, after all, the great power of additive algorithms like neural networks: they're always learning and improving.
So, you might be thinking: will the entire hiring industry soon become redundant? The next in a long line of services to be automated away from humans and placed solely in the hands of AI? In other words...
Are We Worried About Automation?
Sorry, did you want a longer answer than that? Fine, the reality is that we're not worried about these types of tools because of these reasons:
- Soft Skills
- The Human Factor
Thanks to these three core factors, it's going to be an extremely long time - if ever - before a machine can accurately fill roles as successfully as a human.
It's probably the most universally relevant reason presented, but tools like the PFJNN can only really analyse hard skills. Sure, it can determine that the role will best suit someone with high degrees of emotional intelligence, but how can it pinpoint that in a CV?
Tools like Textio and Facebook keep proving that we can derive a huge amount of secondary information from someone's writing style and word choice, from gender bias to political leaning, but they can't accurately predict elements that remain in flux, like emotional state. Your writing captures what you are at the moment you write it, not in general.
As a result, getting a complete overview of someone from a document as heavily curated as a resume will be hit-and-miss at best. Even in person, gauging skills like empathy, teamwork, ambition and communication ability can be tricky, so it will always be a lot harder for a machine to form those kinds of deductions from snippets of curated information.
Heavily related to soft skills, even a very average recruiter will do a certain level of applicant screening. It helps to judge elements like character and "comms", whilst ensuring that the jobseeker's listed talents actually match their real-world ones.
A world that relied purely on an algorithm to deduce job competency is a world that can be gamed. Learn the right keywords and sentence structure and, you too, can become a CEO overnight! It's a great pitch line, but does it sound like a productive hiring market..? Exactly.
No matter how good we get at predictive matching or automated data analysis, humans will always come up with new ways to break the trend or distort the model. Sometimes that will be accidental, sometimes it will be deliberate, but we've seen it with every major technological innovation to date, so systems like Baidu's won't be immune.
As a result, you'll always need a person to screen out the lies, interrogate the soft skills and generally make a judgement call. PJFNN and similar services will make that list of screening calls more accurate, and compile them faster, but actually letting it hire people is likely to be a costly mistake.
The Human Factor
I could have just as easily called this "the company factor", because that's equally as valid. What I mean is that no two roles are going to be identical; no two companies the same; no two applicants truly comparable.
Soft skills and screening go some way to alleviate this, and again, we can tell a lot about subconscious company culture through text analysis, but personalisation remains inherently quicker when people are involved. No piece of software can ever be totally flexible, nor able to be customised to every single user's exact needs.
At some stage, knowing the company that you're hiring for, knowing the job market you're hiring in, knowing the specific regional quirks for the area you're based in – these factors all add up. Neural networks are great at finding the trends, but trends belie averages; they struggle to account for the outliers.
Would it be possible for that to be replicated by software? Sure. But the sheer number of variables involved would either make it prohibitively costly or incredibly complex and hard to test, making the output increasingly questionable. Worse still, algorithms are never perfect and frequently adopt the biases of the data used to hone them or the people creating them, and no one wants to discover their expensive hiring tech is, for example, discriminating against women.
The reality is, it's always going to be easier and more cost effective to create a one-size-fits-all service to sit alongside a person with the specialist knowledge to further curate that list.
The Future of Hiring
That isn't to say that neural networks like the PFJNN won't end up replacing a large chunk of traditional hiring and recruitment work. These tools will certainly be adopted by some recruitment agencies, as they represent potentially huge time-savings.
Today, thanks to the proliferation of digital job boards and personal CV websites like LinkedIn, there's more readily available hiring data than ever before. The problem is finding out where to look in that huge haystack for the specific needle you need to fill your role.
Any system that can scan through those digital archives and come back with a shortlist will be invaluable, and the more accurate the tool is at deciphering context from both sides of the hiring equation – jobseekers and companies – the easier it will be to operate.
That could mean some significant changes to the industry as a whole. In the future, one person might be able to do the work of five very average recruiters today, thanks to massively reduced (and more accurate) sourcing times.
But, at Talent Point, that's something we're already working towards. Our entire business model is based around utilising rich data upfront to plan hiring strategies, making it more accurate and efficient to source top talent. That's why our turnaround time is already much lower than an average recruitment agency.
Could it be improved further still by tools like Baidu's neural network? Of course, and you can be certain that we'll be looking into them as soon as they're publicly available. But we don't see them as a threat, more an augmentation; something to cut out the time consuming, low skilled elements of what we do, letting us focus on the parts that most need our attention.