What facial recognition and the racist pseudoscience of phrenology have in common

“Phrenology” has an old-fashioned sound. Looks like it belongs in a history book, classified somewhere between bloodletting and velocipeds. We would like to think that judging the worth of people based on the size and shape of their skulls is a practice that is well behind us. However, phrenology is once again raising its lumpy head.

In recent years, machine learning algorithms have promised governments and private companies the power to glean all kinds of information about how people look. Several startups now claim they can use artificial intelligence (AI) to help employers detect the personality traits of job applicants based on their facial expressions. In China, the government pioneered the use of surveillance cameras that identify and track ethnic minorities. Meanwhile, there have been reports of schools installing camera systems that automatically sanction children for not paying attention, based on facial movements and micro-expressions such as twitching eyebrows.

Perhaps more notoriously, a few years ago AI researchers Xiaolin Wu and Xi Zhang claimed to have trained an algorithm to identify criminals based on their face shape, with 89.5% accuracy. They did not go so far as to endorse some of the ideas about physiognomy and character that circulated in the 19th century, most notably from the work of Italian criminologist Cesare Lombroso: that criminals are under-resolved, recognizable sub-human beasts. to their sloping foreheads and hawk-shaped noses. However, the recent study’s seemingly high-tech attempt to identify facial features associated with crime borrows directly from the “ photographic composite method ” developed by Victorian jack-of-all-trades Francis Galton – which involved layering the faces of several people in a certain category to find the characteristics indicating qualities such as health, disease, beauty and criminality.

Facial recognition and phrenology

Tech commentators have called these facial recognition technologies “literal phrenology”; they also linked it to eugenics, the pseudoscience of improving the human race by encouraging those deemed most apt to reproduce. (Galton himself coined the term “ eugenics ”, describing it in 1883 as “ all influences which tend to such a distant degree as to give the most suitable races or blood strains a better chance quickly outweigh the less appropriate than they otherwise would have had. ‘)

In some cases, the explicit purpose of these technologies is to deny opportunities to those deemed unfit; in others it may not be the goal, but it is a predictable outcome. Yet when we dismiss algorithms as phrenology, what exactly is the problem we are trying to highlight? Are we saying these methods are scientifically flawed and don’t really work – or are we saying it’s morally wrong to use them anyway?

There is a long and tangled story of how “phrenology” has been used as a fierce insult. The company’s philosophical and scientific critiques have always been closely linked, although their entanglement has changed over time. In the 19th century, critics of phrenology objected to the fact that phrenology attempted to locate different mental functions in different parts of the brain – a movement that was considered heretical, as it challenged Christian ideas about it. unity of soul. . Interestingly, however, trying to figure out a person’s character and intellect based on the size and shape of their head was not seen as a serious moral issue. Today, on the other hand, the idea of ​​localizing mental functions is relatively uncontroversial. Scientists might no longer think of destructiveness to be located above the right ear, but the idea that cognitive functions can be located in particular brain circuits is a standard assumption in mainstream neuroscience.

[Read: How do you build a pet-friendly gadget? We asked experts and animal owners]

Phrenology also had its share of empirical criticism in the 19th century. Debates raged over which functions resided and where and whether skull measurements were a reliable way to determine what is going on in the brain. The most influential empirical critique of ancient phrenology, however, came from studies by French physician Jean Pierre Flourens based on brain damage in rabbits and pigeons – which he concluded that mental functions are distributed rather than localized. . (These results were later discredited.) The fact that phrenology was rejected for reasons that most contemporary observers would no longer accept makes it more difficult to understand what we are aiming for when we use “ phrenology ” as an insult. today.

Statistical biases

Both “old” and “new” phrenology have been criticized for its sloppy methods. In the recent AI Crime Study, the data came from two very different sources: photos of convicts, as opposed to photos of working websites for non-convicts. This fact alone could explain the algorithm’s ability to detect a difference between groups. In a new preface to the article, the researchers also admitted that viewing court convictions as synonymous with criminality was a “ grave oversight. ” Yet equating convictions with criminality seems to register with the authors primarily as an empirical flaw: the use of snapshots of convicted felons, but not of those who escaped, introduces a statistical bias. They said they were “ deeply baffled ” by the public outrage in reaction to an article that was intended “ for pure academic discussion. ”

Notably, the researchers do not comment on the fact that the conviction itself depends on how police, judges and jurors perceive the suspect – which makes a person’s “ criminal ” appearance a variable. of confusion. They also fail to mention how intense policing in particular communities and unequal access to legal representation skews the dataset. In their response to criticism, the authors do not back down on the assumption that “being a criminal requires a host of abnormal (aberrant) personal traits”. Indeed, their framing suggests that crime is an innate characteristic, rather than a response to social conditions such as poverty or abuse. Part of what makes their data set questionable on empirical grounds is that whoever is labeled “ criminal ” is hardly value neutral.

One of the strongest moral objections to using facial recognition to detect crime is that it stigmatizes people who are already overpolished. The authors say their tool shouldn’t be used for law enforcement, but only cite statistical arguments as to why it shouldn’t be deployed. They note that the false positive rate (50%) would be very high, but ignore what that means in human terms. These false positives would be individuals whose faces resemble people who have been convicted in the past. Given the racial and other biases that exist in the criminal justice system, such algorithms would end up overestimating crime in marginalized communities.

The most controversial question seems to be whether reinventing physiognomy is a fair game for the purposes of “pure academic discussion”. One might object on empirical grounds: Past eugenics such as Galton and Lombroso ultimately failed to find the facial features that predisposed a person to crime. This is because there are no such connections. Likewise, psychologists who study the heritability of intelligence, like Cyril Burt and Philippe Rushton, have had to play quickly and freely with their data to fabricate correlations between skull size, race, and IQ. If there was anything to be discovered, the many people who have tried it over the years probably wouldn’t have dried up.

The problem with reinventing physiognomy is not simply that it has been tried unsuccessfully before. Researchers who persist in pursuing cold fusion after the scientific consensus has passed also face criticism for hunting unicorns – but the disapproval of cold fusion is far from shame. At worst, they are seen as wasting their time. The difference is that the potential damage from cold fusion research is much more limited. In contrast, some commentators argue that facial recognition should be regulated as tightly as plutonium, because it has so few non-harmful uses. When the dead end project you want to resurrect was invented for the purpose of supporting colonial and class structures – and when the only thing it is able to measure is the racism inherent in those structures – it’s hard to justify try it one more time, just for curiosity.

However, calling facial recognition research “phrenology” without explaining what is at stake is probably not the most effective strategy for communicating the strength of the complaint. For scientists to take their moral responsibilities seriously, they must be aware of the harm that could result from their research. It is hoped that making it clearer what is wrong with the work titled “phrenology” will have more of an impact than simply throwing the name away as an insult.

This article was originally published at Aeon by Catherine Stinson and has been republished under Creative Commons.

Published March 13, 2021 – 14:00 UTC