That's hilariously bad - as someone with no more than a general knowledge. But I do know that portraits reflected fashions, and tried to make the subject look good in most cases.
jimbob wrote: Fri Sep 25, 2020 2:17 pm
That's hilariously bad - as someone with no more than a general knowledge. But I do know that portraits reflected fashions, and tried to make the subject look good in most cases.
Yes, as the Tweeter says, they were of course almost all commissioned by wealthy people who wanted the art equivalent of airbrushing to look good and show off to their mates. Most portraits of rich people in the past served the same purpose as selfies now.
My avatar was a scientific result that was later found to be 'mistaken' - I rarely claim to be 100% correct
ETA 5/8/20: I've been advised that the result was correct, it was the initial interpretation that needed to be withdrawn
Meta? I'd say so!
Bird on a Fire wrote: Fri Sep 25, 2020 3:47 pm
You can't always see correlations in complex data by eyeballing scatter plots. That's why statistical tests exist.
Yes but an R-squared of -0.12 with no prior reason to suppose one would be either positive or negative, and when one of the variables is "perceived trustworthiness" isn't exactly brilliant.
Bird on a Fire wrote: Fri Sep 25, 2020 3:47 pm
You can't always see correlations in complex data by eyeballing scatter plots. That's why statistical tests exist.
Yes but an R-squared of -0.12 with no prior reason to suppose one would be either positive or negative, and when one of the variables is "perceived trustworthiness" isn't exactly brilliant.
Oh sure, it's explaining a very small percentage of variance and doubtless has a small effect size, even considering the fact that "perceived trustworthiness" is a difficult concept to measure with many many influences and you'd likely never get an R2 over about |0.5|.
None of that means that they haven't found a statistically significant relationship in the data. The problem there is a wider one (that "statistical significance" doesn't mean that an effect is large or interesting).
We have the right to a clean, healthy, sustainable environment.
How did the title of this paper get past peer-review?
Surely it should have been something like???
"Tracking historical changes in PERCEIVED trustworthiness using machine learning analyses of facial cues in paintings"
You don't need to believe in phrenology or anything like that to study the fact that people do make snap judgements about how trustworthy people are based on what they look like? These judgements may not be correct - but people do judge people based on what their faces look like.
Throughout the paper they keep referring to "trustworthiness" (which they didn't study) when they should have been referring to "perceived trustworthiness".
bob sterman wrote: Sat Sep 26, 2020 7:42 pm
How did the title of this paper get past peer-review?
Surely it should have been something like???
"Tracking historical changes in PERCEIVED trustworthiness using machine learning analyses of facial cues in paintings"
You don't need to believe in phrenology or anything like that to study the fact that people do make snap judgements about how trustworthy people are based on what they look like? These judgements may not be correct - but people do judge people based on what their faces look like.
Throughout the paper they keep referring to "trustworthiness" (which they didn't study) when they should have been referring to "perceived trustworthiness".
jimbob wrote: Sat Sep 26, 2020 6:24 pm
For comparison I've plotted the population rank of countries (China=1) against their alphabetical rank (Afghanistan=1) and got this
I'm definitely nicking that. It's much better than those fake correlations of .95 ("Nicolas Cage movies versus earthquakes in China" or whatever), which (as I learned earlier this year) are actually not an illustration that everything is correlated with something, but rather of how easy it is to produce huge correlations with two time series.
Something something hammer something something nail
jimbob wrote: Sat Sep 26, 2020 6:24 pm
For comparison I've plotted the population rank of countries (China=1) against their alphabetical rank (Afghanistan=1) and got this
I'm definitely nicking that. It's much better than those fake correlations of .95 ("Nicolas Cage movies versus earthquakes in China" or whatever), which (as I learned earlier this year) are actually not an illustration that everything is correlated with something, but rather of how easy it is to produce huge correlations with two time series.
Following that tweet thread, an XKCD led me to webplot digutizer, which *is* a nice simple webpage to digitize pictures of graphs with minimal input
bob sterman wrote: Sat Sep 26, 2020 7:42 pm
How did the title of this paper get past peer-review?
Surely it should have been something like???
"Tracking historical changes in PERCEIVED trustworthiness using machine learning analyses of facial cues in paintings"
You don't need to believe in phrenology or anything like that to study the fact that people do make snap judgements about how trustworthy people are based on what they look like? These judgements may not be correct - but people do judge people based on what their faces look like.
Throughout the paper they keep referring to "trustworthiness" (which they didn't study) when they should have been referring to "perceived trustworthiness".
The only cue you need is to know the portraits are of rich white people. Who didn't get rich by being nice/trustworthy.
Tessa K wrote: Sun Sep 27, 2020 11:34 am
The only cue you need is to know the portraits are of rich white people. Who didn't get rich by being nice/trustworthy.
Yes - but before photographs they could have benefited greatly from having an obsequious portrait artist churning out paintings that made them LOOK way more nice and trustworthy than they actually were!
So on a serious note - given most portraits were commissioned they can shed light on how the person depicted wanted to be seen (rather than what they actually looked like). Bit like modern selfies with filters I guess.
Tessa K wrote: Sun Sep 27, 2020 11:34 am
The only cue you need is to know the portraits are of rich white people. Who didn't get rich by being nice/trustworthy.
Yes - but before photographs they could have benefited greatly from having an obsequious portrait artist churning out paintings that made them LOOK way more nice and trustworthy than they actually were!
So on a serious note - given most portraits were commissioned they can shed light on how the person depicted wanted to be seen (rather than what they actually looked like). Bit like modern selfies with filters I guess.
Surely all of this assumes that
(a) facial features that (allegedly) correspond to trustworthiness don't also correspond to any number of traits which, at any point in time, might be considered desirable for a posh person to communicate
and
(b) the purpose of a painting is to communicate those features to the viewer --- who is probably a rich and powerful person who knows directly how trustworthy you are.
The model that the authors seem to have is that of a modern Western political candidate, trying not to look dodgy to potential voters. This seems unlikely to be what the average subject of a painted portrait from several centuries ago had in mind.
Imagine you're Oliver Cromwell sitting for a picture. You probably don't give a f.ck about being considered trustworthy, and you also know that your portrait isn't going on Instagram. You might just about give instructions to the painter to "make me look really hard", and your judgement of how good a job the painter did will presumably not be based on some algorithm's averaging of obscure parameters of a thousand other faces.
Something something hammer something something nail
May be going off at a bit of a tangent here, but...
I've read that some good poker players are able to read micro-expressions in their opponents' faces, the tiny movements of facial muscles. Is it possible that the algorithm may (or may be alterable to) do something similar?
Just a thought.
Some people call me strange.
I prefer unconventional.
But I'm willing to compromise and accept eccentric.
Aitch wrote: Mon Sep 28, 2020 2:15 pm
May be going off at a bit of a tangent here, but...
I've read that some good poker players are able to read micro-expressions in their opponents' faces, the tiny movements of facial muscles. Is it possible that the algorithm may (or may be alterable to) do something similar?
Aitch wrote: Mon Sep 28, 2020 2:15 pm
May be going off at a bit of a tangent here, but...
I've read that some good poker players are able to read micro-expressions in their opponents' faces, the tiny movements of facial muscles. Is it possible that the algorithm may (or may be alterable to) do something similar?
Just a thought.
The algorithm looks at several-centuries-old paintings. (Also, I would bet money [DYSWIDT?] that the stories about poker players are mostly post hoc justification, taken from interviews with the winners.)
Something something hammer something something nail
Some portraits were also used as an olden days version of Tinder so the main intention was to convey wealth and fertility. Some were notoriously unreliable - Ann of Cleves was a good example.
As someone said above, trustworthiness in images only matters to politicians and used car salesmen. If you were rich enough to have a portrait you wouldn't give a stuff about it.
In later art a lot of models were prostitutes as it wasn't considered ladylike to pose in the buff or to look directly at the viewer - shameless!
Scientist Lou Safra, who worked on the project, said: “We cannot precisely say why these images appear more or less trustworthy.
"However, from studying we know that someone who has sunken cheeks, furrowed brows and eyes set close together is more likely to be perceived as an untrustworthy or dishonest person.
"And that someone with a smiling face, prominent cheekbones and high eyebrows is more likely to be perceived as a trustworthy, honest person.”
Tessa K wrote: Wed Sep 30, 2020 11:39 am
And now the media has picked up on this nonsense which means it's only a matter of time before Trump starts to quote it.
Scientist Lou Safra, who worked on the project, said: “We cannot precisely say why these images appear more or less trustworthy.
"However, from studying we know that someone who has sunken cheeks, furrowed brows and eyes set close together is more likely to be perceived as an untrustworthy or dishonest person.
"And that someone with a smiling face, prominent cheekbones and high eyebrows is more likely to be perceived as a trustworthy, honest person.”
At least there the quoted scientist is making it clear that what is being discussed is perceived trustworthiness - doesn't seem to be suggesting these perceptions relate to actual trustworthiness.
When people make snap judgements about others - e.g. at first glance - trustworthiness and dominance seem to be the 2 most important dimensions that they form judgements about...
Even if these judgements are not accurate surely it's worth studying them? E.g. to understand why some people can get away with all sorts of things while actually being dishonest rogues?
