Tag Archives: False Positive

Facial recognition tech fail

Police facial recognition trial led to erroneous arrest
Alexander J Martin
Sky News August 31st 2017



Is there any particular reason why prosopagnosics are Australia’s favourite popularizers of science?

Dr Karl Kruszelnicki is a prosopagnosic, and apparently so is Robyn Williams, who has been the hosting The Science Show on Australian public radio since the last ice age with intelligence and grace and a pleasantly smart but mild English accent. They both work for the ABC in both TV and radio. They have both written many popular science books. They both come across as likable and enthusiastic. Is this just coincidence? Looking overseas, other highly successful popularisers of science, such as Oliver Sacks and Jane Goodall have also been identified as prosopagnosics. In his role as host of QI, actor Stephen Fry has done a lot to educate and popularise science and other types of knowledge. He’s one too. Strange coincidence that this particular type of fame seems to go with a very particular inability to recognize or memorise faces more often that it should for a characteristic that affects around 1 in 50 people? Maybe it is just more likely that a person who is very interested in science is more likely to identify their self as a scientific curiosity? I could contrast this group of people with famous people who have identified as synaesthetes. Synaesthesia, like prosopagnosia is a psychological-neurological characteristic that is uncommon but not rare. and quite interesting but definitely not obvious. Unlike celebrity prosopagnosics, it seems as though famous figures who claim synaesthesia tend to be more into the arts than the sciences. So what gives?

I found out about Robyn Williams and prosopagnosia reading part of the transcript of an upcoming episode of the radio show Ockham’s Razor which is hosted by Williams. The guest of the show is scientist Len Fisher, and guess what? Another prosopagnosic. He’s made the claim that apophenia is the opposite of prosopagnosia. I can see the logic behind this claim but “No”. Super-recognition is the opposite of prosopagnosia, because face recognition is a type of memory ability, and it is also highly specific to visual memory of faces. The concept of super-recognition is a mirror-image of the concept of prosopagnosia, and both specifically relate to the visual memory of faces. In contrast, apophenia is a very loose and general concept; the tendency of humans to perceive meaningful patterns within stimuli or data that are actually random. Apophenia is not specific to faces or to visual stimuli, and it is a more general term than pareidolia, which I’ve previously written about at this blog. The concept of apophenia seems to me to be too vague a concept to have any scientific utility or meaning, rather like the concept of autism. That’s my opinion, but I’m open to good arguments against it.

Another objection that I have to the idea of apophenia as the opposite of prosopagnosia is the apparent assumption that nature cannot create a biological system of face recognition that is accurate and doesn’t have a tendency towards either false positives (type I error or identifying unfamiliar faces as familiar) or false negatives (type II error or identifying familiar faces as unfamiliar). The source of this type of erroneous thinking about face recognition is the common (among scientists and non-scientists) miscategorisation of face recognition as a form of sensory perception rather than a form of visual memory. As far as I know there’s not anything necessarily amiss about the way prosopagnosics see or perceive faces. They don’t see faces as blurs or blanks. They just don’t remember them. And there’s no reason to think that supers have anything super about the way we see faces. There’s nothing super-human about my eyesight acuity or my ability to identify facial expressions. There’s also nothing in my face recognition ability that looks like any trend towards false positives. As I’ve explained in the first post in this blog, I’m not prone to incorrectly identifying strangers as familiar people, as has been observed in some stroke patients. Very occasionally I’ve had interaction between synaesthesia and face recognition, but this doesn’t affect accuracy.

There’s no reason for skepticism of the proposition that evolution can design a visual memory system that is amazingly swift and accurate and operates unconsciously and automatically. This is simply how visual perception works, for humans and for animals that are seen as much less cerebral than humans. Apparently there’s evidence that the humble pigeon can recognize human faces, and other bird species appear to have evolved the ability to visually recognize the difference between the speckles of their own eggs and those of similar eggs laid by the parasitic cuckoo bird. Evolution can achieve accuracy in systems, if there is a need for such systems to evolve, but it is also plausible that such abilities might be uneven in levels within populations, as variation within populations is completely normal and necessary in biological systems.


The trouble with police, large photograph databases and face recognition technology

Hodson, Hal Police mass face recognition in the US will net innocent people. New Scientist. October 20th 2016.


United States Government Accountability Office Face Recognition Technology: FBI Should Better Ensure Privacy and Accuracy. May 2016.


Had you assumed that hiring human super-recognizers to perform face recognition tasks would be less effective, less accurate and more open to bias than using technology? Think again.