Under normal circumstances, gestures and posture provide clues to gain an impression of applicants. In a video call, all of this is reduced to a small section and flattened into a two-dimensional image. Recruiters, who could otherwise rely on their gut feeling, can no longer be sure of their judgment in this situation.
People and companies that offer face reading promise to help in these cases. They read hidden emotions from faces that recruiters want to know about and have experienced a modest comeback in the last 20 years. That's why we want to outline the approaches and methods used in face reading below.
Methods based on the teachings of physiognomics from antiquity and those of Lavater, Lambroso or Huter are more appropriately classified as esoteric/new age. They claim a direct connection between the shape of the face and head and certain character traits, which cannot be scientifically proven and is based on cognitive distortions.
Less dubious is the basic assumption of pathognomics, namely that life circumstances, emotions and the resulting facial expressions are reflected in the appearance (wrinkles etc.) of faces. Which is a truism.
Two main forms of scientific study of the face, however, are the Facial Action Coding System (FACS) and the study of microexpressions. The FACS assigns codes to the various movements of the face, which can be used to describe facial expressions in a kind of notation.
Microexpressions, on the other hand, are extremely fleeting facial expressions that reveal the emotions even of people who have their facial expressions very well under control. They often last only half a second and were first discovered in the 1960s in slow-motion recordings of psychotherapeutic films. Anyone familiar with the TV series Lie to Me will know what we are talking about here. In it, Tim Roth plays a deception expert who uses microexpressions to uncover the lies of suspects for the police.
Microexpressions, on the other hand, are extremely fleeting facial expressions that reveal the emotions even of people who have their facial expressions very well under control. They often last only half a second and were first discovered in the 1960s in slow-motion recordings of psychotherapeutic films. Fans of the TV series Lie to Me know what we are talking about here. In the series, Tim Roth plays a deception expert who uses microexpressions to uncover the lies of suspects for the police.
In principle, face reading exists on the one hand as a human consulting service (and corresponding training) and on the other as a technology-driven software product.
There are people and companies who advise their customers by observing the facial expressions of applicants or negotiating partners. This results in proper psychological profiles.
Software providers whose products analyze videos and claim to decode the emotions of applicants in real time also promise to achieve this. One example of this is the company Human, which uses artificial emotional intelligence (AEI) for screening and creating personality profiles. Among other things, microexpressions are detected and interpreted. In 2018, Human established a partnership with the recruitment software platform Workable. Human's KEI was integrated into the Workable platform and is used to profile candidates. It is designed to compensate for recruiters' cognitive biases and ensure objectivity.
The Israeli start-up Faception ("a Facial Personality Analytics company") offers a similar product. Using machine learning, the software analyzes faces and reveals their personality in real time - or so they say. The company makes its software available to companies and security authorities. It is even said to enable the identification of paedophiles and terrorists.
Unfortunately, the company does not provide any relevant data on its website that would allow a judgment to be made about the effectiveness of Faception. For example, evidence of the identification of actual terrorists or paedophiles and clear statements about the probability of true and false positives would be desirable. Faception claims that the AI identified the majority of the Paris attackers as terrorists, all of whom looked Arabic - as presumably did the faces with which it was trained. But the question is whether the AI would also have identified Ulrike Meinhof or Anders Breivik, for example, as terrorists.
Physiognomics and psycho-physiognomics should be treated with the utmost caution, as they are the product of inadmissible generalizations. They may occasionally be correct, but they harbor great risks. For example, there is a danger of creating self-fulfilling prophecies: a person, who is pre-judged as "untrustworthy" based on the bone structure of their face or similar characteristics and is subsequently treated as "untrustworthy", will behave accordingly in the long term.
In this way, potentially valuable employees are either not hired in the first place or are labeled and forced into roles from which they can no longer free themselves.
Personality profiling using video analysis and artificial intelligence also still has to prove its effectiveness and harmlessness. Here too, there is a risk of self-fulfilling prophecies and inadmissible generalizations.
Another problem that can arise is false positive results. Faception, for example, claims that the algorithms used are correct for some of its person archetypes in 80 percent of cases. It is not clear what exactly this means in practice. In any case, the figure of 80 percent leaves room for 20 percent incorrect results. In a large application process with hundreds of applicants, this can result in a considerable number of errors. It gets even worse if someone is wrongly classified as a terrorist or pedophile on the basis of such a personality profile.
There is also a risk that artificial intelligence could systematically disadvantage certain groups of people due to coding errors. This is what happened in Texas, which uses risk assessment software with a face reading component in criminal cases. The results of this software are made available to judges. The software estimates the likelihood of recidivism and thus influences the further treatment of the accused and, in some cases, the level of the sentence. An independent study found that the software systematically disadvantages African-Americans and particularly fails in predicting violent crimes.
The first beauty contest ever to be decided by an AI was also plagued by such biases. The AI chose predominantly white women in the category of the most beautiful people. Here it becomes clear that AIs that analyze faces (instead of just recognizing them) are currently achieving less meaningful results, especially across cultural and ethnic boundaries.
Furthermore, AIs that analyze faces are not yet particularly good at reliably determining the underlying emotions. For example, they cannot distinguish whether a vertical crease between the eyebrows means that a person is angry or whether the person is just concentrating hard on a task. An application video could be an occasion where this problem arises.
Although people's faces allow conclusions to be drawn about their emotions, drawing conclusions about their personality is currently fraught with more potential harm than good. This is because there are too many gray areas in facial expressions that an AI cannot distinguish clearly.
At best, psychologists can make valid statements about people's emotional state based on their facial expressions. By analyzing microexpressions in particular, it is possible to clearly identify whether a person is lying.
In order to recruit managers with particularly high levels of responsibility, a thorough analysis of their micro-expressions is therefore a very sensible investment. Because of their high position, they can cause a lot of damage in a company.
For the time being, however, facial analysis using AI is still far from the point where it should be trusted without reservation.
That's why we're glad that we at metru use the AI Four Precire. It analyzes something that the candidates will actually do in their prospective job, which is communicate. After all, language is another area in which no one can pretend for long. And every person has a personal speech pattern that is as individual as a fingerprint. Apart from that, speech analysis has the considerable advantage that the AI is not confused by external factors such as skin color.
This post was first published in August 2021 in our old Recruiter Blog.
%201new.png)
.svg)
.svg)
.svg)
