School of Human Sciences
Face and Voice Recognition Lab
Institute of Lifecourse Development
University of Greenwich
A major profile article on super-recognisers was published on Saturday 30th October in the Washington Post (Cimons, 30 October 2021) featuring the work of the Face and Voice Recognition Lab at the University of Greenwich.
No US police deployment of super-recognisers
One of the main points covered in the article was that no US police force officially identifies super-recognisers for deployment to specialist roles that could capitalise on their skills. This seems strange when the University of Greenwich has signed research consultancy contracts with over 10 different police forces in Germany alone. These projects involve testing of all interested police employees to identify if any might be super-recognisers (see Davis, 2019, 2020 for information on the development of these procedures).These projects have resulted in almost immediate positive results in terms of crime detection. For instance, the blog below describes how a Stuttgart Police pilot project led the Deputy Prime Minister of Baden-Württemberg state to endorse the University of Greenwich procedures for police forces across the entire state.
Stuttgart Police Project: https://www.superrecognisers.com/post/super-recogniser-police-assist-in-stuttgart-riot-investigations
The deployment of super-recognisers should be beneficial in ensuring that true offenders are convicted. This is because super-recognisers are more accurate than the majority of the population at correctly identifying people they have previously seen. Crucially, they may also protect innocent suspects inadvertently caught up in a police investigation, as they are also more accurate at recognising that they have never before seen a specific face.
Some US police officers have been identified as super-recognisers after taking the University of Greenwich tests that are included in our collaborations with the Association of Super-Recognisers (https://www.associationofsuperrecognisers.org/) and Super-Recognisers International (https://superrecognisersinternational.com/). All, however, completed the tests as private individuals.
UK Case example: PCSO Andy Pope (Super-recogniser)
More information can be found in previous blogs about Andy Pope (West Midlands Police, UK), the super-recogniser who was also interviewed for the Washington Post. Andy who has identified more than 2,000 suspects is a Licensee of the Association of Super-Recognisers and has contributed to our lab-based research at Greenwich.
As revealed on BBC1's Crimewatch Live show in March 2021, Andy also possesses superior voice recognition ability.
The future of super-recognition?
With David Robertson (Strathclyde University), I wrote a 2020 article for the US Department of Homeland Defense in house journal (Journal of the Homeland Defense & Security Information Analysis Centre), in which we discussed the benefits to US police and security of employing super-recognisers (Davis & Robertson, 2020). We suggested that one reason for the lack of US interest in super-recognisers may be the already widespread (and controversial) deployment of Artificial Intelligence (AI) driven computerised face recognition algorithms by US police. There seems to be a common belief that, "Who needs humans when computers can do the job?"
However, this belief may be inappropriate. The need to avoid employing staff with prosopagnosia (sometimes called 'face blindness'), and instead to employ super-recognisers is implied within the ISO/IEC 30137-1 "Information technology — Use of biometrics in video surveillance systems" guides of best practice.
This ISO standard emphasises the need to use human operator validation of all Artificial Intelligence (AI) facial recognition alerts, in order to discard the regular false-positives that such systems generate.
"An operator support environment to aid in making decisions on whether an alert should be followed up (and how) or rejected as a false alert." (6. Operator Decision)
The ISO guidelines focus on ensuring that operators are suitably trained to perform the role. As training has limited impact on improving accuracy at face matching/recognition tasks, most experts in the field of super-recognition would probably agree that it would be far more productive to employ those with superior skills in the first place.
Finally, it is also unlikely that for a case to proceed to a criminal trial, decisions as to the identity of a person depicted in say CCTV images will ever be left to an algorithm. A human being will always be required to make that final decision. And who better than a super-recogniser?
Such easy win-wins are normally far harder to find.