Social Engineering Blogs

An Aggregator for Blogs About Social Engineering and Related Fields

Facial Age and Recognition

Sometimes it is very important to evaluate age based on someone’s face alone, but this can be quite tricky.

This is actually a surprisingly pressing issue as age is relevant in all sorts of commercial, social, or political contexts. While it seems intuitive that we should be able to recognize people’s ages pretty easily, this has been challenged, if not refuted, by emerging research. For instance, in a series of experiments by Dr. Colin Clifford and his team, it appears that people tend to be incredibly bad at accurately judging age.

Not only is it often awkward or inconvenient to not be able to judge someone’s age, but age also undergirds a great deal of social evaluations. Group identification, emotional evaluations, and other assorted judgments are heavily determined by our perceptions of age, as should not be surprising to most readers.

Given the importance of age evaluations in social interactions, Dr. Clifford’s team attempted to expose experimental participants to a massive database of passport photographs, tasking them with estimating the age of the person.

The experimental design was relatively simple; though employing some complicated measures to avoid sampling biases and other confounding factors. Essentially, each of the 84 participants was asked to identify the age of almost 400 participants, ranging greatly in age and across gender.

Overall, they detected certain systemically incorrect estimations across the participants. Namely, young faces tended to be seen as older, while older faces tended to be seen as younger. This was actually in line with some previous research that found that age estimations tend to skew to middle aged faces.

Importantly, perceptions of age tended to by heavily impacted by the face most recently seen. This can take the form of bias related to gender, attractiveness, or facial expression. While this research sought to control for that, Dr. Clifford did find that, without controls, such dependency would have significant impacts on evaluations.

One interesting aspect of this path dependency is the impact that previous assessments of age have. After coding multiple faces as being young, participants were more likely to gauge subsequent pictures as younger than they were. This is particularly notable given that that is the opposite of the expected and normal bias of skewing towards middle aged assessments.

Dr. Clifford’s work not only helps demonstrate the challenges of accurately gauging age but also helps shed light on how we can be influenced by seeing other faces. For instance, a bartender who is used to seeing older faces might be more likely to overestimate a younger customer’s age.

While you may not be much better at age estimation than the average participant, it is helpful to keep these considerations in mind. Knowing a person’s age can be a helpful tool in correctly reading their emotions, but it can also help us gauge whether that person is a threat in various social situations.

Complex Deceptive Behaviors

A great deal of popular discussion of how to detect deception rests on specific, isolated factors like eye contract, but the reality is a bit more complex.

This is the case that Humintell’s Drs. David Matsumoto and Hyisung Hwang made in a 2017 study published in the Journal of Police and Criminal Psychology. In this experimental analysis, they had participants engage in a simulated investigative interview which, after being recorded, was analyzed to see which deceptive nonverbal behaviors were exhibited and, most importantly, in what combinations.

Importantly, while many previous studies have found that certain nonverbal behaviors are reliable indicators of deception, these findings have often been difficult to replicate. These studies have focused on vocal fluctuations, body language, and gestures, all of which do demonstrate underlying emotions.

However, Drs. Matsumoto and Hwang emphasize that, because of the complex emotions involved in deception, analyzing just one behavior at a time seems problematic. This is why, in the current study, they sought to see whether looking at clusters of behaviors may help solve this puzzle.

In order to do this, they recruited a series of participants who were all asked to engage in a mock crime simulation. These participants were given the opportunity to “steal” a $100 check, with some told to do so and some to refrain. Both groups were then assigned to mock interviews where they were either told to lie or confess.

With this premise set up, the exciting analysis work began. Each interview was recorded and then analyzed, frame by frame, with machine-learning informed algorithms which sought to categorize individual frames based on certain emotions, including many basic emotions like anger, disgust, fear, happiness, etc.

This allowed the researchers to calculate exactly which emotions tended to be the most common during the interview.  Then, they hand coded a series of nonverbal behaviors, including head shakes, nods, and shoulder shrugs.  This analysis was then combined with sophisticated assessments of vocal pitch and volume, helping create a comprehensive account of the subtle behaviors involved in the interview process.

When comparing these behaviors to whether or not the interviewee was lying, Drs. Matsumoto and Hwang found that it was clusters of non-verbal behaviors that most reliably predicted deception. Liars tended to have fewer head nods and greater changes in vocal pitch, though with a lower average.

Importantly, the types of questions, be they open-ended or more closed had significant impacts.  Liars tended to have even lower pitches during open-ended questions, for example.

These findings have significant ramifications for anybody attempting to detect deception. While many of us are told to focus on individual behaviors, like eye contact or closed postures, these alone cannot fully explain the situation.

Instead, deception seems based on these clusters of behaviors which can be even more difficult to determine. This is definitely on reason why Humintell offers advanced deception detection classes which can be helpful for anyone, but especially any of you who make it your business of conducting lie detection interviews.