“Phrenology” has an old-fashioned sound. It sounds like it belongs in a history book that is filed somewhere between bloodshed and velocipedes. We’d like to believe that judging the worth of people by the size and shape of their skull is a practice well behind us. However, phrenology is raising its lumpy head again.
In recent years, machine learning algorithms have promised governments and private companies the ability to extract all sorts of information from people’s appearance. Several startups are now claiming to be able to use artificial intelligence (AI) to help employers identify the personality traits of applicants based on their facial expressions. In China, the government has pioneered the use of surveillance cameras to identify and track ethnic minorities. Meanwhile, it has been reported that schools are installing camera systems that automatically sanction children for non-compliance based on facial movements and micro-expressions such as eyebrow twitches.
Perhaps best known is that a few years ago, AI researchers Xiaolin Wu and Xi Zhang claimed to have trained an algorithm to identify criminals based on the shape of their faces with an accuracy of 89.5%. They did not go so far as to endorse some of the ideas about physiognomy and character that were popularized in the 19th century, particularly from the work of Italian criminologist Cesare Lombroso: criminals are underdeveloped, subhuman beasts, recognizable by their sloping foreheads and hawk-like Noses. However, the seemingly high-tech attempt of the most recent study to find crime-related facial features is based directly on the “composite photographic method” developed by the Victorian all-rounder Francis Galton, in which the faces of several people were superimposed in a certain category to find the features that were on To indicate properties such as health, disease, beauty, and crime.
Face recognition and phrenology
Technology commentators have referred to these facial recognition technologies as “literal phrenology”. They have also linked it to eugenics, the pseudoscience of improving humanity by encouraging people who are believed to be best suited to reproduce. (Galton himself coined the term “eugenics” and described it in 1883 as “all influences which, to such a small extent, tend to give the more suitable races or blood strains a better chance than they usually do to prevail quickly against the less suitable would. “had. ‘)
In some cases, the explicit goal of these technologies is to deny opportunities to those deemed unsuitable. In other cases, it may not be the goal, but it is a predictable outcome. What exactly is the problem we want to point out when we discard algorithms by calling them phrenology? Are we saying that these methods are scientifically flawed and don’t really work – or are we saying that it is morally wrong to use them anyway?
T.Here’s a long and tangled story of how “phrenology” was used as a withering insult. Philosophical and scientific criticism of the company has always been intertwined, although their interrelationship has changed over time. In the 19th century, phrenology critics objected to the fact that phrenology attempted to locate the location of various mental functions in different parts of the brain – a move that was considered heretical as it challenged Christian ideas about the unity of the soul posed. Interestingly, however, trying to discover a person’s character and intellect from the size and shape of their head was not seen as a serious moral issue. In contrast, the idea of locating mental functions is pretty much undisputed today. Scientists may no longer believe that destructiveness lies above the right ear, but the notion that cognitive functions can be located in specific 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. There was debate about what functions were where and whether skull measurements were a reliable way to determine what was going on in the brain. However, the most influential empirical criticism of ancient phrenology came from the studies by French doctor Jean Pierre Flourens that damaged the brains of rabbits and pigeons. From this he concluded that the mental functions are distributed rather than localized. (These results were later discredited.) The fact that phrenology was rejected for reasons most contemporary observers would no longer accept only makes it harder to figure out what we are aiming at when we use phrenology as a bow today.
The statistical bias
Both the “old” and the “new” phrenology have been criticized for their sloppy methods. In the recent AI crime study, the data was taken from two very different sources: mug photos of convicts versus images of work websites for non-convicts. This fact alone could explain the algorithm’s ability to distinguish between groups. In a new foreword to the paper, the researchers also admitted that it was a “serious oversight” to regard judicial convictions as synonymous with crime. However, equating convictions with crime seems to register itself mainly as an empirical error: the use of mug shots of convicted criminals, but not of those who escaped, leads to a statistical bias. They said they were “deeply stunned” by the public outrage in response to a paper that was “intended for purely academic discussion”.
By Wu and Zhang (2016)
In particular, the researchers do not comment on the fact that the conviction itself depends on the impressions police, judges and juries make of the suspect – which makes a person’s “criminal” demeanor a confusing variable. Nor do they mention how the intense surveillance of certain communities and the inequality of access to legal representation skew the data set. In their response to criticism, the authors do not start from the assumption that “a criminal requires a large number of abnormal (outlier) personal characteristics”. Indeed, their wording suggests that crime is an innate trait rather than a response to social conditions such as poverty or abuse. Part of what empirically makes their dataset questionable is that whoever is classified as “criminal” is barely value neutral.
One of the strongest moral objections to using facial recognition to detect crime is stigmatizing people who are already being treated beyond the police force. The authors say that their tool should not be used for law enforcement purposes, but should only provide statistical arguments for why it should not be used. They find that the false positive rate (50%) would be very high but fail to realize what this means for humans. These false positives would be people whose faces resemble people who have been convicted in the past. Given the racist and other prejudices that exist in the criminal justice system, such algorithms would overestimate crime among marginalized communities.
The most controversial question seems to be whether reinventing physiognomy is a fair game for the purposes of “purely academic discussion”. One could raise objections on empirical grounds: eugenicists of the past like Galton and Lombroso ultimately found no facial features that predisposed a person to crime. This is because such connections cannot be found. Likewise, psychologists like Cyril Burt and Philippe Rushton, who studied the hereditary nature of intelligence, had to play quickly and easily with their data to establish correlations between skull size, race, and IQ. If there was something to discover, the many people who have tried over the years probably wouldn’t have dried up.
The problem with reinventing physiognomy is not just that it has previously been tried unsuccessfully. Researchers who continue to look for cold fusion after the scientific consensus has advanced are also criticized for hunting unicorns – but the disapproval of cold fusion lags far behind opprobrium. In the worst case, they are seen as a waste of time. The difference is that the potential harm from cold fusion research is much less. In contrast, some commentators argue that facial recognition should be as tightly regulated as plutonium because it has so few non-harmful uses. If the cul-de-sac project you’re trying to revive was invented to prop up colonial and class structures – and when the only thing that can measure it is the racism inherent in those structures – it’s hard to justify going over it again to attempt. only from curiosity.
However, calling facial recognition research “phrenology” without explaining what it is about is probably not the most effective strategy for communicating the power of complaint. In order for scientists to take their moral responsibilities seriously, they must be aware of the harm that can result from their research. Hopefully, if you make it clearer what is wrong with working called “phrenology” it will have more implications than simply throwing the name around as an insult.
This article was originally published on Aeon by Catherine Stinson and republished under Creative Commons.
Published on March 13, 2021 – 14:00 UTC
Comments are closed.