Frequently Asked Questions
Below are questions I am most frequently asked about Super-Recognizers. Regardless of how you ended up here, this is a good place to start collecting information about the topic. The answers are intended to be short and understandable for non-experts.
Each answer ends with a “too long didn’t read” (TLDR) version. You’re welcome.
You can download the FAQs my answers here; please cite them as:
Ramon M (2021). Super-Recognizers: Frequently Asked Questions. https://doi.org/10.17605/osf.io/gvcuw
For a collection of other experts' replies see here.
The term “Super-Recognizer” was introduced by Russell and colleagues in 2009. They reported 3 individuals who excelled at three different face processing tasks: face matching, face recognition and face identification.
Currently, there is no formal definition of what actually constitutes a Super-Recognizer.
Things are a bit confusing for at least two reasons:
1. Differences in testing
Researchers use a variety of tests, which can differ vastly. For example, you may have to (a) decide whether a pair of faces shows the same person, or (b) learn a number of faces and later recognize them among entirely new ones. While (a) relies on visual perception, (b) also requires visual memory. Even two tests of face perception are not necessarily comparable — a test that uses highly dissimilar faces will be easier than one that uses relatively more similar faces.
2. Terminology used outside of the lab
Outside academia and scientific research others have used the term Super-Recognizer to refer to people who excelled as person processing. For instance, police officers who could track, recognize or identify perpetrators using various types of information including non-facial information have been (incorrectly) labeled as Super-Recognizers.
TLDR: Super-Recognizers are individuals who excel at processing facial information - eg in the context of face matching, recognition and/or identification.
Super-Recognizers are better for face processing in general, and some individuals might excel at other-ethnicity processing. However, recent evidence suggests that Super-Recognizers also show an own-ethnicity advantage. This means that they too tend to find the processing of faces of other ethnicities more difficult — just like typical observers.
TLDR: Exposure matters; Super-Recognizers seem to be more proficient at processing ethnicities that they are most familiar with.
The numbers that you may find in the media are not reliable estimates. Super-Recognizers are currently identified via performance on behavioral tests, where estimates will depend on the
TLDR: To date, there is no reliable estimate the prevalence of Super-Recognizers in the general population.
Based on the available evidence, I do not think that you cannot become a Super-Recognizer. Rather, it seems like you either are one or not.
Training of security professionals, eg passport officers, leads to at best modest effects; more often than not training leads to no effects.
Even professional memory champions — who achieve impressive feats thanks to extensive training with mnemonic techniques — seem to process faces differently than Super-Recognizers.
When typical observers have to match faces for an extended period, their accuracy declines with time - even regular breaks or distractions cannot eliminate the performance decline. We have shown that Super-Recognizers use facial information more consistently in challenging and long experiments, even in surprise tests of facial memory.
TLDR: Everyone can improve their performance through training, but practice cannot make up for Super-Recognizers naturally occuring and consistent advantage.
Super-Recognizers - if appropriately identified - could aid all tasks where high fidelity facial identity verification is required.
While media reports can create the impression that Super-Recognizers are being used widely, my experience is somewhat different: Interest in the topic varies across and within countries; there are no formal or evaluated approaches of identifying and deploying Super-Recognizers.
This is not actually surprising. Although police and security agencies may have common interests, their specific needs and strategic priorities will also differ and thus require individual consideration.
The Berlin State Police and I have developed BeSuRe(TM), a behavioral assessment tool to identify Super-Recognizers among the >18.000 officers of the Berlin Police, designed specifically to meet their deployment needs.
To find out more about this project, here are a few resources:
TLDR: Super-Recognizers could aid areas that require facial identity verification.
Superior face processing skills could theoretically be helpful in a number of professions, but Super-Recognizers are a new phenomenon and the available scientific knowledge is extremely limited.
Super-Recognizers are not an officially recognized profession; I have never heard of any employer looking to hire a Super-Recognizer. Government agencies have sufficient manpower to find people with superior skills among their employees.
Even if governments could hire civilians for security-relevant tasks, there are many open HR relevant questions: How should Super-Recognizers' abilities be verified, and what can they (not) be expected to do? What are the limits of their capacities, and how do they change eg with aging?
TLDR: Super-Recognizers are not an officially recognised profession; I have never heard of any employer looking to hire a Super-Recognizer.
Researchers have looked at potential ties to other aspects like intelligence and object processing. Currently, there is no strong evidence to suggest any such relationships, but previous studies typically involved relatively small groups of Super-Recognizers.
So, it is possible that future studies with larger samples of Super-Recognizers may reveal relationships to other skills. I would not be surprised if we found that their visual processing is generally better than that of average observers.
TLDR: To date there is no evidence suggesting that Super-Recognizers have other special skills.
Humans and algorithms each have their own (dis)advantages.
Algorithms are tireless and fast. But, to perform individual face identification, they need to be trained with a lot of images. Also, different algorithms make different mistakes*.
Humans can learn facial identities very quickly, ie they require less “training” than automatic solutions. Human face processing is also very flexible: factors like distance, lighting, partial occlusion, etc seem to have a comparatively small effect.
So, rather than deciding between humans or algorithms, the optimal solution would be to combine the best of each.
* A recent test of contemporary face recognition algorithms carried out by NIST found that algorithms
“exhibit demographic differentials of various magnitudes … the highest false positives are in American Indians, with elevated rates in African American and Asian populations; the relative ordering depends on sex and varies with algorithm.”
Note that the algorithms were tested with largely ideal images (eg mugshots, immigration benefit or visa application photos). The test did not
“[u]se wild images: We did not use image data from the Internet nor from video surveillance. This report does not capture demographic differentials that may occur in such photographs.”
If you know of a test comparing algorithms’ performance with naturalistic conditions, ie “wild images”, please let me know.
TLDR: Automatic face recognition accuracy can vary greatly (across available algorithms), and is less efficient than human face processing (due to algorithms’ requirements and/or development characteristics).