Seen the Nature Comms paper that caused outrage? Did you also think: “this is not physiognomy, just an algorithm mimicking human biases; surely there is no intent to use that algorithm for other purposes"?
This came out yesterday (h/t @zeyneparsel ) 1/n https://www.thesun.co.uk/news/12796321/most-trustworthy-celeb-faces-kim-kardashian-holly-willoughby/
This came out yesterday (h/t @zeyneparsel ) 1/n https://www.thesun.co.uk/news/12796321/most-trustworthy-celeb-faces-kim-kardashian-holly-willoughby/
Quotes from the piece (with some comments):
"Experts from PSL Research University, in France, have created an algorithm which scans the faces in painted portraits and photographs to discover how trustworthy people are based on their facial muscle contractions." 2/n
"Experts from PSL Research University, in France, have created an algorithm which scans the faces in painted portraits and photographs to discover how trustworthy people are based on their facial muscle contractions." 2/n
(This statement does not specify that the algorithm is a (even quite poor; see original paper) model of human *biased* and *culturally shaped* impressions of "trustworthiness" with no basis in truth). 3/n
(The following sentence is:)
"The algorithm doesn't assess trustworthiness in itself - rather the qualities and traits the sitters of a photograph wanted to portray in their image." 4/n
(Two observations w.r.t. to this qualification, below:)
"The algorithm doesn't assess trustworthiness in itself - rather the qualities and traits the sitters of a photograph wanted to portray in their image." 4/n
(Two observations w.r.t. to this qualification, below:)
1) As Todorov explains in this clip, ppl form "TW" impressions in a blink of an eye. AFAIK there is no claim here that ppl can control their faces to look "trustworthy", other than smiling (but for that we do not need this fancy algorithm, do we?). 5/n https://www.youtube.com/watch?time_continue=26&v=c_vbCzWnpEk&feature=emb_logo
So the claim that the algorithm detects, e.g., the "trustworthiness" a person *wants* to display seems to be a speculation by the authors & *not* based on the psych literature on human impressions of "trustworthiness". Any experts in this area that can elucidate this point? 6/n
2) For most readers, the qualification is undone by the rest of the text. E.g., the algorithm is referred to as "facial profiling technology", and authors are not consistently careful to prevent the impression that, indirectly, this technology is a form of physiognomy. 7/n
For instance:
“However, from studying we know an untrustworthy or dishonest person will have sunken cheeks, furrowed brows and eyes set close together."
“And a trustworthy, honest person usually has a smiling face with prominent cheekbones and high eyebrows.” 8/n
“However, from studying we know an untrustworthy or dishonest person will have sunken cheeks, furrowed brows and eyes set close together."
“And a trustworthy, honest person usually has a smiling face with prominent cheekbones and high eyebrows.” 8/n
Here, authors mention some features under a person's control (smiling, high eyebrows) but also features that are not (eye set close together, prominent cheekbones). Attributing intentions to display "trustworthiness" based on features one can't control seems a slippery slope. 9/n
You can tell that also the text writer is unclear on the limitations of the algorithm, and makes the occasional statement that reduces to physiognomy.
E.g., under a picture of Joe Biden:
"His smile and cheekbones prove that he can be trusted, according to science." 10/n
E.g., under a picture of Joe Biden:
"His smile and cheekbones prove that he can be trusted, according to science." 10/n
This is just the 1st popular press piece that came out (or we saw) about the Nature Comms paper. There'll be more. And the more there are, the more the limitations of the algorithm will be lost in communication and the more it becomes ... physiognomy. 11/n https://twitter.com/IrisVanRooij/status/1310690623092142090?s=20
For a piece detailing problems with the original @NatureComms paper, incl. what defences of the work were tried and how they fail, see this great summary and analysis by @rory_spanton : https://twitter.com/rory_spanton/status/1310866047264870400?s=20 12/n