Read on Twitter
I don't think this paper is half as bad as a number of Twitter users have made it out to be—I certainly don't think it's flatly racist physiognomy. Here's why (brief thread): https://twitter.com/baumard_nicolas/status/1308715606196342784
The model is based on work by Todorov et al. on perceptions of trustworthiness from facial features. This work was designed to measure bias, not reify bias. Predictably—because of racism—the participants judged, e.g., dark skin to appear less trustworthy
The authors understood their participants were biased. In fact, one of Todorov's papers on this topic begins "People reliably and automatically make personality inferences from facial appearance despite little evidence for their accuracy." https://www.sciencedirect.com/science/article/pii/S1364661308002350
They were trying to figure out why and how people make judgments based on facial features even when this is a very bad strategy! It does _not_ suggest that these facial features are linked in any way to actual trustworthiness.
So the art historical question Baumard is asking is something like, "People today make judgments of trustworthiness based on configurations of facial features. How were these facial features used in English portraiture over time?"
They then ask a set of broader questions, something like "What are the relationships between trustworthiness displays, survey measures of trust, and other cultural/sociopolitical factors, including GDP?"
You might not find these questions interesting, you might think English portraiture too narrow, or question whether it's the paintings that are changing or the way facial features are interpreted, or raise criticisms about the strength of the effects...
... and you might also find the narrative about GDP pollyannaish in the manner of Pinker, and think it diverts attention in the wrong direction. All of these criticisms are fair enough!
But it's clear to me that this work is not an attempt to revive physiognomy. It's almost the opposite—it unpacks how physiognomic intuitions have shaped the tradition of English portraiture, and what their prevalence is in broader socioeconomic context
Arguing that this work should be retracted strikes me as uncomfortably close to saying "We should never make tools that can detect bias, because any tool that detects bias could also be used to enforce it."
The authors are quite clear that what they are doing measures bias, not objective trustworthiness. They do not claim these biases are good or desirable.
Because this thread has gotten quite a bit of attention, I want to make it completely, explicitly clear that the argument I'm making here is meant to be strictly limited to the question of reifying physiognomy—not these other possible problems https://twitter.com/beausievers/status/1309486096351473668
Specifically, I think there is a reasonable argument to be made along this line from @stephenniem that the paper's pollyannaish narrative about GDP and violence misses the mark. This parallels arguments against Pinker's claims about declining violence. https://twitter.com/stephenniem/status/1309543545217445888
Along these lines, here's a sharp criticism that hits the mark. Even if we accept the results, we shouldn't take an optimistic or positive view of the historical or present-day role of trust displays: https://twitter.com/criticalneuro/status/1309513704128557057?s=20
At the risk of digging my own grave, and because I am being held to task by respected friends, I also want to say that failing to account for how trustworthiness displays fit into the colonial context is not morally equivalent to endorsing physiognomy.
This is certainly not a paper I would have written, and it has serious flaws worthy of critical commentary, but I don’t think it should be retracted.
This is another important criticism of the paper: that, given the risks, the caveats about bias need to be center stage.
The quoted thread also refers to an extremely bad paper on "ML gaydar" (by Kosinski) which does clearly endorse and reify physiognomy https://twitter.com/bmwiernik/status/1309836655185080321
The quoted thread also refers to an extremely bad paper on "ML gaydar" (by Kosinski) which does clearly endorse and reify physiognomy https://twitter.com/bmwiernik/status/1309836655185080321
And I think this is a strong criticism of my thread, which I (perhaps foolishly) wrote with only psychologists in mind.
For most, the details of where this research came from simply don't matter. They just don't want a boot on their face. https://twitter.com/CreativOriginal/status/1309843784864346112?s=20
For most, the details of where this research came from simply don't matter. They just don't want a boot on their face. https://twitter.com/CreativOriginal/status/1309843784864346112?s=20
Anyway, I'm logging off. Donate to organizations that support criminal justice reform in the US.
I just donated $150 to the Texas Organizing Project https://organizetexas.org/about-top/
I just donated $150 to the Texas Organizing Project https://organizetexas.org/about-top/
