Old and new phrenology: Face recognition by the size and shape of the skull

2024 Author: Seth Attwood | [email protected]. Last modified: 2023-12-16 15:55

Phrenology is an old-fashioned lady. This concept is probably familiar to you from history books, where it is located somewhere between bloodletting and cycling. We used to think that evaluating a person by the size and shape of a skull is a practice that has remained deep in the past. However, phrenology rears its lumpy head here and again.

In recent years, machine learning algorithms have enabled governments and private companies to collect all kinds of information about the appearance of people. Several startups today claim they can use artificial intelligence (AI) to help determine the personality traits of job candidates based on their faces. In China, the government was the first to use surveillance cameras to detect and track the movements of ethnic minorities. Meanwhile, some schools use cameras that track children's attention during lessons, detecting facial and eyebrow movements.

And a few years ago, researchers Xiaolin Wu and Xi Zhang said they had developed an algorithm for identifying criminals by face shape, providing an accuracy of 89.5%. Quite reminiscent of the ideas of the 19th century, in particular, the work of the Italian criminologist Cesare Lombroso, who argued that criminals can be recognized by their sloping, “animal” foreheads and hawk noses. Obviously, the attempts of modern researchers to isolate facial features associated with crime are based directly on the "photographic composite method" developed by the master of the Victorian era, Francis Galton, who studied people's faces in order to identify signs that indicate such qualities as health, illness, attractiveness. and crime.

Many observers consider these face recognition technologies to be "literal phrenology" and associate them with eugenics, a pseudoscience that aims to identify the people most adapted to reproduction.

In some cases, the explicit purpose of these technologies is to de-empower those deemed "unusable." But when we criticize such algorithms, calling them phrenology, what problem are we trying to point out? Are we talking about the imperfection of methods from a scientific point of view - or are we speculating about the moral side of the issue?

Phrenology has a long and convoluted history. The moral and scientific sides of her criticism have always been intertwined, although their complexity has changed over time. In the 19th century, critics of phrenology objected to the fact that science was trying to pinpoint the location of various mental functions in different parts of the brain - a movement that was viewed as heretical because it challenged Christian ideas about the unity of the soul. Interestingly, trying to uncover a person's character and intelligence from the size and shape of their head was not perceived as a serious moral dilemma. Today, on the contrary, the idea of localizing mental functions causes fierce controversy over the moral side of the issue.

Phrenology had its share of empirical criticism in the 19th century. There has been controversy about what functions are located and where, and whether measurements of the skull are a reliable way to determine what is happening in the brain. The most influential empirical criticism of the old phrenology, however, came from the research of the French physician Jean Pierre Flourens, who based his arguments on the study of the damaged brain of rabbits and pigeons, from which he concluded that mental functions are distributed, not localized (these conclusions were later refuted). The fact that phrenology has been rejected for reasons that most modern observers no longer accept makes it difficult to determine where we are aiming when we criticize the science today.

Both "old" and "new" phrenology are criticized primarily for methodology. In a recent computer-assisted crime study, data came from two very different sources: photographs of inmates and photographs of people looking for work. This fact alone can explain the features of the resulting algorithm. In a new foreword to the article, the researchers also acknowledged that accepting court sentences as synonymous with crime propensity was a "serious oversight." Nevertheless, the sign of equality between convicts and those prone to crimes, apparently, is considered by the authors mainly as an empirical flaw: after all, the study studied persons only who were brought before the court, but not those who escaped punishment. The authors noted that they were “deeply bewildered” by public outrage in response to material intended “for purely academic discussion”.

It is noteworthy that the researchers do not comment on the fact that the conviction itself may depend on the perception of the suspect's appearance by the police, judges, and jury. They also did not take into account the limited access of various groups to legal knowledge, assistance and representation. In their response to criticism, the authors do not depart from the assumption that “many abnormal (external) personality traits are required to be considered a criminal”. In fact, there is an unspoken assumption that crime is an innate characteristic and not a reaction to social conditions such as poverty or abuse. Part of what makes the dataset empirically dubious is that whoever gets labeled “criminal” is unlikely to be neutral towards social values.

One of the strongest moral objections to using facial recognition to detect crime is that it stigmatizes people who are already embittered enough. The authors say their tool should not be used in law enforcement, but only provide statistical arguments as to why it should not be used. They note that the rate of false positives (50 percent) will be very high, but are oblivious to what that means from a human point of view. Behind these "mistakes" people will be hiding, whose faces simply look like those convicted of the past. Given racial, national and other biases in the criminal justice system, such algorithms end up overestimating crime among marginalized communities.

The most controversial question seems to be whether the rethinking of physiognomy serves as a "purely academic discussion." One might argue on an empirical basis: the eugenicists of the past, such as Galton and Lombroso, ultimately failed to identify the facial features that predisposed a person to crime. This is because there are no such connections. Likewise, psychologists who study the inheritance of intelligence, such as Cyril Burt and Philip Rushton, have failed to establish a correlation between skull size, race, and IQ. No one has succeeded in this for many years.

The problem with rethinking physiognomy lies not only in its failure. Researchers who continue to seek cold fusion are also facing criticism. At worst, they are just wasting their time. The difference is that the potential harm of cold fusion research is much more limited. On the contrary, some commentators argue that facial recognition should be regulated as strictly as plutonium trafficking, because the harm from both technologies is comparable. The dead-end eugenic project that is being resurrected today was launched with the aim of supporting colonial and class structures. And the only thing he is able to measure is the racism inherent in these structures. Therefore, one should not justify such attempts by curiosity.

However, calling facial recognition research "phrenology" without explaining what is at stake is probably not the most effective strategy to criticize. For scientists to take their moral duties seriously, they need to be aware of the harm that can arise from their research. Hopefully a clearer statement of what is wrong with this work will have a greater impact than unfounded criticism.