It is a fact universally acknowledged that a manager in possession of too much data is in need of an algorithm.
The number of start-ups pitching proprietary algorithms that claim to use artificial intelligence and the sum-total of human knowledge, processing petabytes of data it in a matter of minutes, have multiplied over the past few years.
Nowhere is this more apparent than in the growing field of predictive hiring. Recruiters, managers, entrepreneurs, employers – have little time to wade through and assess the hundreds of resumes they receive to fill a single position. Their goal is to find the person who provides the “best fit” for the job. Time is of the essence here. So what if the science is a little dubious?
The market is significant. Fortune estimates that there are at least 75 start-ups competing with established incumbents in a $100 billion growing HR assessment market.
When it comes to assessing people, the most important asset of any business, there is no way of knowing for sure if a person the algorithm picked was better than the person the algorithm rejected. There are studies pouring out of research houses and academia that conclude: “algorithms beat instinct” showing statistically that machine-picked hiring out-performs human decisions. But such studies rely on internal performance assessment systems gamed by “corporate types” in search for promotions. These systems are not known for spotting future leaders.
Analysing The Mind
“Analysis” as Freud would have defined it – getting inside someone’s head – is an altogether different proposition.
It is an inexact science which, at best, and even with the tremendous advancements in psychology and neuro-sciences, remains judgemental and fraught with error.
The human mind is extremely complex. Every person is unique. We have tools and case studies and copious research on physiology of the brain but even the most experienced of professional psychologists will find it difficult to pronounce definitive judgements on an individual’s ability to lead or grow a company by studying their mannerisms, body language, facial expressions or choice of language. Yet this is exactly what some of these “breakthrough” hiring algorithms do.
Personality tests and psychometric tests have been around for many years and have been the norm in large organisation in selecting people. These tests identify, if they identify anything at all, personality traits that are obvious (someone is good in mental arithmetic or likes to be around people or leans towards being an introvert). They also produce a set of questionable observations that are difficult to verify. What they do reveal are a few things about how the human mind works and how we establish our analytical preferences. First, we prefer simplicity of explanation. Second, we prefer to categorise things, in this case people, slotting every person on the planet in one of a handful of personality types. And third, to make life even easier, especially when pressed for time, most of us are willing to sacrifice transparency for convenience.
Speed Over Science
Recruiters in fast growing companies are amazed by the speed with which predictive hiring software platforms solve the insurmountable task of sifting through a large number of applications narrowing the field down to a manageable short-list of usually not more than 1 or 2% of the total.
Software companies claim not only is this process highly productive, it is also fair. It filters out human bias and errors of judgement born out of mental fatigue. People produce inconsistent results when they are tired. Research has shown that judges give harsher sentences before lunch when they are hungry. These become softer after snack breaks. The task of intelligent algorithms is to reduce such inconsistencies.
Weapons Of Math Destruction
Cathy O’Neil, author of the book Weapons of Math Destruction, quotes the example of a San Francisco based start-up called Gild which goes through thousands of sites and collects “social data” identifying, among other things, who a person is connected to. If a person is connected to someone important in the high-tech world, their score gets a bump up. Also, if they share characteristics of talented peers, their score goes up again. For example, it found that “a bevy of talent” frequented a certain Japanese manga site. Its algorithms incorporating this correlation attributed a higher score to someone who visits that site. If an excellent programmer had other more important things to do like taking care of the kids or spending time with family or reading an interesting book, they would not get a bump up in their score.
When assessing a “good fit” – predictive hiring platforms look for similarities between candidates and the people they will be working with. This results in people hiring clones. People feel comfortable with colleagues who have similar preferences and similar backgrounds to them. People do the same when hiring people. They tend to hire clones. The algorithm simply automates that process incorporating the same bias in its logic.
Sometimes the marketing message is more powerful than the algorithms. A Seattle based start-up called Koru offers “predictive hiring” in which it deciphers the culture at a client organisation using a 20 minute survey and then sifts through resumes of candidates to see who provides a good fit. When assessing people it looks at the first 7 years of their career. It is difficult to understand how such a summary approach can claim to offer assistance in meaningful decision making. But it has a powerful and effective marketing slogan “Grit Over Grades” and a growing portfolio of HR professionals in large companies who believe in its approach.
The Science Of Inflections and Microgestures
Some start-ups have taken the science of human assessment to an altogether different level. HireVue, a company based in Utah, videos candidate interviews. Its algorithms evaluate a candidate’s “microgestures” to identify their personalities and inclinations. It notices where facial expressions contradict the choice of words. In a promotional video on its website, HireVue says 25,000 unique data points become available from an interview including all types of facial expressions, intonation, inflection, word choice, and content of each answer. All this helps a company “simplify” their recruiting process and “modernise (the) candidate experience” – now that is a choice of words that needs deciphering.
Another one, Interviewed.com offers “assessment-powered hiring automation” software which claims to help its customers “hire 10x better”. It uses machine learning and natural language interface technologies to assess a candidate’s personality. Algorithm logic is based on assertions such as those who say please or thank you frequently have empathy and therefore are better suited for customer oriented jobs.
“In the United States, 72% of CVs are never seen by human eyes”, says O’Neil. The more candidates these software programs eliminate, the better it is for the recruiter. Applicants now aim to game the algorithms “sprinkling their resumes liberally with words the specific job opening is looking for,” adds O’Neil. “Those with the latest information learn what machines appreciate and what tangles them up, and tailor their applications accordingly”.
But the most concerning aspect of machine assistance in predictive hiring is that the opaqueness of these algorithms cannot be questioned and their conclusions are beyond reproach and beyond appeal. These conclusions cannot easily be tested in the short term. The company will never know if the machine disposed off applications from some future stars nor whether it short-listed those who would contribute little to the company’s future.
The fact that such algorithms save an enormous amount of time and save recruiters from indulging in the dull and tiring task of reading resumes. Things we don’t like tire us just like a character out of a Jane Austen novel: “Nothing ever fatigues me but doing what I do not like”.