lorry Posted March 5, 2018 Posted March 5, 2018 I was just thinking of project ideas and this popped into my head. It would certainly be interesting to do. It would rustle so many fucking jimmies! I don't think the problem would be particularly difficult, that is the fun bit, but creating a data-set would be a nightmare! But just imagine it lads! Any ideas on how one might be made? (data-set, not a classification model)
lorry Posted March 5, 2018 Author Posted March 5, 2018 p.s. Obviously started googling around, here is a (possibly) interesting paper (FDR interesting, not OP interesting). Machine learning methods to predict child posttraumatic stress: a proof of concept study
barn Posted March 5, 2018 Posted March 5, 2018 Hi @lorry You might find this interesting (Johann Kaspar Lavater - criminality & facial features, Physiognomy tab) , regarding facial features and their correlation to predictive qualities. I'm highly suspicious that it (the research you've linked) has any merit as it seems to me 'physiognomy' but I don't know for sure.
lorry Posted March 5, 2018 Author Posted March 5, 2018 I dunno, barn. If a little bully can successfully pick a victim (the victim being a neglected/abused child, thus, ptsd-esque) or a pedophile can successfully pick a victim, then someone with a huge head and thick rimmed glasses (psysiogmony banter) can probably train a machine to do the same (if given sufficient data). http://www.clinical-innovation.com/topics/artificial-intelligence/ai-facial-analysis-system-improve-diagnosis-rare-diseases
lorry Posted March 5, 2018 Author Posted March 5, 2018 Ok, making that labeled data set is probably not going to work. I was thinking something like... write a spider to pull information from facebook adoption pages. But then we get instantly stumped by the fact the adopted children are.... adopted! Not related to either parent. So then I suppose we could find sperm donor mothers with partners. They would be good, but a lot of work. I had a better idea. So we take a whole bunch of large families with pictures of lots of labeled relatives. Then we train a regression model on the genetic distance (based on relation, ie, parent, sibling, grandparent, cousin, etc) against facial similarity. Then we use this regression model to predict the genetic distance between father and child. Pick a suitable threshold, and boom! We have our classification model! (We can use the model to predict genetic distance between all available family members, if a father is non-paternal, the mother still should be, huge signal! And siblings/cousins/grand-parents will/will not be if we know which side of the family they are from)
barn Posted March 5, 2018 Posted March 5, 2018 On 3/5/2018 at 2:04 AM, lorry said: I dunno, barn. If a little bully can successfully pick a victim (the victim being a neglected/abused child, thus, ptsd-esque) or a pedophile can successfully pick a victim, then someone with a huge head and thick rimmed glasses (psysiogmony banter) can probably train a machine to do the same (if given sufficient data). Expand Don't get me wrong, I like the idea of simplifying identification; akin to i.e. how the shape of the nail can be an additional cue to for example of hypotension...etc. It's just that since I have known about physiognomy, I haven't seen any conclusive evidence to an existing correlation between personality traits and facial features. It's a many thousands years old question, frequently recurring and gaining popularity time to time. If I remember correctly, the Chinese were amongst the first to try to implement it for screening/picking applicants for the right positions in the empire. Regarding bullies, pedophiles... etc. If I understand it correctly, they identify their prey based on behaviour, level of social integration, some arbitrary preferences related to their own personal history. Similar to how predators in nature pick out the old, weak, ill, solitary ones due to the relatively easier scenario they represent. Would you agree? On 3/5/2018 at 2:15 AM, lorry said: Ok, making that labeled data set is probably not going to work. I was thinking something like... write a spider to pull information from facebook adoption pages. But then we get instantly stumped by the fact the adopted children are.... adopted! N Expand Hahaha... indeed. On 3/5/2018 at 2:15 AM, lorry said: (We can use the model to predict genetic distance between all available family members, if a father is non-paternal, the mother still should be, huge signal! And siblings/cousins/grand-parents will/will not be if we know which side of the family they are from) Expand I wouldn't be surprised if a similar model had already existed, correlating data on prominent genes and their likelihood of expression when comparing members of a family tree. Although it must be very limited and again, only applicable to medical conditions, not personality. By the by, I can think of a dozen institutions eager to get their hands on such a way to screen people for their personality traits... the Police, Military, Universities...
lorry Posted March 5, 2018 Author Posted March 5, 2018 On 3/5/2018 at 9:23 AM, barn said: It's just that since I have known about physiognomy, I haven't seen any conclusive evidence to an existing correlation between personality traits and facial features. It's a many thousands years old question, frequently recurring and gaining popularity time to time. If I remember correctly, the Chinese were amongst the first to try to implement it for screening/picking applicants for the right positions in the empire. Expand Did you catch Stef talk about (tweet about) the relationship between attractiveness and IQ? And if there is a relationship between IQ and personality traits, then you have your relationship between facial features and personality. On 3/5/2018 at 9:23 AM, barn said: Regarding bullies, pedophiles... etc. If I understand it correctly, they identify their prey based on behaviour, level of social integration, some arbitrary preferences related to their own personal history. Similar to how predators in nature pick out the old, weak, ill, solitary ones due to the relatively easier scenario they represent. Would you agree? Expand Well, they are selecting for vulnerability, right? And vulnerability is something related to parental neglect, so it goes with ACE score, etc. And they can pick it up from things like.... clothing, posture, appearance, etc. And I mean they can pick it up at a glance (this much at least). If you started looking at the relative frequencies of certain behaviors, such as propensity for eye contact, timber of voice, etc.. etc.. etc.. which would be your level of social integration. But there is a lot you can pick up just from an image. On 3/5/2018 at 9:23 AM, barn said: I wouldn't be surprised if a similar model had already existed, correlating data on prominent genes and their likelihood of expression when comparing members of a family tree. Although it must be very limited and again, only applicable to medical conditions, not personality. By the by, I can think of a dozen institutions eager to get their hands on such a way to screen people for their personality traits... the Police, Military, Universities... Expand A similar model probably does exist, sure. But creating a model is not the difficult part of creating a model. Creating a data-set is the hardest bit, then engineering features to train a model on. That said, if anyone reading this did want to do big 5 personality vs whatever, there is a facebook vs big 5 and IQ score dataset already.
barn Posted March 5, 2018 Posted March 5, 2018 On 3/5/2018 at 7:44 PM, lorry said: Did you catch Stef talk about (tweet about) the relationship between attractiveness and IQ? Expand No, I haven't. Would you be able to quote? On 3/5/2018 at 7:44 PM, lorry said: And if there is a relationship between IQ and personality traits, then you have your relationship between facial features and personality. Expand I don't see the connections between intelligence and attractiveness yet, perhaps if I saw the quote and references to the source. Some at least... Can you help me find it if you don't want to share it here? On 3/5/2018 at 7:44 PM, lorry said: Well, they are selecting for vulnerability, right? Expand Mostly, I'd assume that's correct. On 3/5/2018 at 7:44 PM, lorry said: And vulnerability is something related to parental neglect, so it goes with ACE score, etc. Expand (neglect is only just one aspect)... And bad propaganda, and not taking responsibility... ACE scores do help further investigation. On 3/5/2018 at 7:44 PM, lorry said: And I mean they can pick it up at a glance (this much at least). Expand I can see how in certain circumstances it could be immediate but people can have bad days, cross their arms due to cold,... I doubt such predators wouldn't invest into a 'deeper, detailed investigation', it's risky to be revealed, caught. Good. On 3/5/2018 at 7:44 PM, lorry said: But there is a lot you can pick up just from an image. Expand Currently, I don't think that is sufficient to support your claim. I'd have to see evidence to that. Large sample study, falsifyable theorems... etc. Or maybe I'm just not good enough to "pick up a lot from just an image". (changed the word order, hope the meaning stayed the same) On 3/5/2018 at 7:44 PM, lorry said: similar model probably does exist, sure. But creating a model is not the difficult part of creating a model. Creating a data-set is the hardest bit, then engineering features to train a model on. Expand I would create a rock-solid theory before I ventured any further into juggling with data-sets. Currently, the idea that facial features can be used to decipher personality traits from an image seems quite a flimsy idea. Perhaps it's just me. Regards, Barnsley
lorry Posted March 6, 2018 Author Posted March 6, 2018 On 3/5/2018 at 9:30 PM, barn said: I would create a rock-solid theory before I ventured any further into juggling with data-sets. Currently, the idea that facial features can be used to decipher personality traits from an image seems quite a flimsy idea. Perhaps it's just me. Regards, Barnsley Expand https://www.psychologytoday.com/blog/the-scientific-fundamentalist/201012/beautiful-people-really-are-more-intelligent
barn Posted March 6, 2018 Posted March 6, 2018 On 3/6/2018 at 12:10 AM, lorry said: https://www.psychologytoday.com/blog/the-scientific-fundamentalist/201012/beautiful-people-really-are-more-intelligent Expand Well, I'm starting to get even more suspicious... You, not quoting what according to you Stefan Molyneux has said on twitter, does not help neither. Here are a few quotes from the article you linked, let me know how you see them. 1. (the subtitle, not even the article goes saying: ) "Intelligence is just as strongly correlated with beauty as with education. Posted Dec 12, 2010 2."If you want to estimate someone’s intelligence without giving them an IQ test, you would do just as well to base your estimate on their physical attractiveness as you would to base it on their years of education." 3. "Now, given that it was the children’s teacher who was asked to assess their physical attractiveness, there is a possibility of a halo effect, where teachers believe that better, more intelligent students are physically more attractive." 4." Where does the teachers' belief that more intelligent students are more attractive come from? The notion that more intelligent individuals are physically more attractive is a stereotype, and, just like all other stereotypes, it is empirically true, as both the American and British data show. Teachers (and everyone else in society) believe that more intelligent individuals are physically more attractive because they are." After reading the article, I found it to be partially true. Maybe that's just me.
lorry Posted March 6, 2018 Author Posted March 6, 2018 On 3/6/2018 at 8:36 AM, barn said: Well, I'm starting to get even more suspicious... You, not quoting what according to you Stefan Molyneux has said on twitter, does not help neither. Here are a few quotes from the article you linked, let me know how you see them. 1. (the subtitle, not even the article goes saying: ) "Intelligence is just as strongly correlated with beauty as with education. Posted Dec 12, 2010 2."If you want to estimate someone’s intelligence without giving them an IQ test, you would do just as well to base your estimate on their physical attractiveness as you would to base it on their years of education." 3. "Now, given that it was the children’s teacher who was asked to assess their physical attractiveness, there is a possibility of a halo effect, where teachers believe that better, more intelligent students are physically more attractive." 4." Where does the teachers' belief that more intelligent students are more attractive come from? The notion that more intelligent individuals are physically more attractive is a stereotype, and, just like all other stereotypes, it is empirically true, as both the American and British data show. Teachers (and everyone else in society) believe that more intelligent individuals are physically more attractive because they are." After reading the article, I found it to be partially true. Maybe that's just me. Expand -1. Easy enough to falsify. Go to the facebook vs IQ database, model posting VS IQ. Then model predict IQ against physical attractiveness in facebook photos. 0. Nope, he did it, it is somewhere out there if you want to find it. 1. Yup 2. Yup 3. Supposition, but see -1. 4. Yup, Stereotypes are true. Primitive expression of population averages compared to white norms. Blacks? Criminals. Mexicans? Lazy. Jews? Liars (and plagiarists). Pakistanis? Rapists.
barn Posted March 6, 2018 Posted March 6, 2018 On 3/6/2018 at 2:36 PM, lorry said: -1. Easy enough to falsify. Go to the facebook vs IQ database, model posting VS IQ. Then model predict IQ against physical attractiveness in facebook photos. Expand Sorry, I don't know how this is an answer to one of my question. Which question? On 3/6/2018 at 2:36 PM, lorry said: 0. Nope, he did it, it is somewhere out there if you want to find it. Expand I think it is your responsibility to support your anecdote with evidence, not mine trying to find something that might not exist. Naturally, you don't have to provide evidence if that's how you roll. On 3/6/2018 at 2:36 PM, lorry said: 1. Yup 2. Yup Expand Interesting. Education amounts to less than 10% regarding intelligence, did you know that? Education, years of it, is almost irrelevant regarding IQ. On 3/6/2018 at 2:36 PM, lorry said: 4. Yup, Stereotypes are true. Primitive expression of population averages compared to white norms. Blacks? Criminals. Mexicans? Lazy. Jews? Liars (and plagiarists). Pakistanis? Rapists. Expand My jaw dropped. I think, I don't need to ask more.
lorry Posted March 6, 2018 Author Posted March 6, 2018 On 3/6/2018 at 2:55 PM, barn said: Sorry, I don't know how this is an answer to one of my question. Which question? I think it is your responsibility to support your anecdote with evidence, not mine trying to find something that might not exist. Naturally, you don't have to provide evidence if that's how you roll. Interesting. Education amounts to less than 10% regarding intelligence, did you know that? Education, years of it, is almost irrelevant regarding IQ. My jaw dropped. I think, I don't need to ask more. Expand If you want to verify the results, -1 is how you would do it. I didn't think it was more important to find where Stefan tweeted the information, than the information itself. Education amounts to less than 10% of IQ, sure, but education is a significant predictor of IQ. Knowing what you studied, and at what level, predicts IQ pretty well. I didn't realise I was talking with a woman, I do apologise (English, polite).
barn Posted March 6, 2018 Posted March 6, 2018 On 3/6/2018 at 4:36 PM, lorry said: If you want to verify the results, -1 is how you would do it. I didn't think it was more important to find where Stefan tweeted the information, than the information itself. Education amounts to less than 10% of IQ, sure, but education is a significant predictor of IQ. Knowing what you studied, and at what level, predicts IQ pretty well. I didn't realise I was talking with a woman, I do apologise (English, polite). Expand So no evidence(no quote you claim to exist) is your choice of 'rolling', sprinkled with soft insult. Lovely. I found your generalisations negating the existence of individuality and reducing people to averages. The average free-lunch doesn't exist. If I recall correctly, the base argument was that facial features can be used to predict personality traits. Beauty isn't subjective, I know that for sure. Ability, Opportunity, to go into higher education where there's somewhat survived the traces of merit based evaluation process, requires higher IQ, specific personality traits... Definitely. However you can't read how successful/intelligent/wealthy someone is from a photo because that would require biology to negate the effects of peaceful parenting, free-will, the ability to change your mind through well reasoned arguments, therapy for that matter too. You'd have to prove that the world is deterministic, first. If you had said that your idea was to improve predictive capabilities to a certain degree... I could concede to that. In this form, especially how you're avoiding to show evidence... (This isn't an argument) I'm not convinced you've considered the probability that not all your assumptions will hold up to basic scrutiny, then maybe you'll need to reconsider a few things. I'm not saying you should, I'm saying that I don't see much probability currently for it being an option... yet.
lorry Posted March 7, 2018 Author Posted March 7, 2018 On 3/6/2018 at 8:08 PM, barn said: So no evidence(no quote you claim to exist) is your choice of 'rolling', sprinkled with soft insult. Lovely. I found your generalisations negating the existence of individuality and reducing people to averages. The average free-lunch doesn't exist. If I recall correctly, the base argument was that facial features can be used to predict personality traits. Beauty isn't subjective, I know that for sure. Ability, Opportunity, to go into higher education where there's somewhat survived the traces of merit based evaluation process, requires higher IQ, specific personality traits... Definitely. However you can't read how successful/intelligent/wealthy someone is from a photo because that would require biology to negate the effects of peaceful parenting, free-will, the ability to change your mind through well reasoned arguments, therapy for that matter too. You'd have to prove that the world is deterministic, first. If you had said that your idea was to improve predictive capabilities to a certain degree... I could concede to that. In this form, especially how you're avoiding to show evidence... (This isn't an argument) I'm not convinced you've considered the probability that not all your assumptions will hold up to basic scrutiny, then maybe you'll need to reconsider a few things. I'm not saying you should, I'm saying that I don't see much probability currently for it being an option... yet. Expand With all respect... you have no skill in statistics, probability, or machine learning, and I don't care what you think.
barn Posted March 8, 2018 Posted March 8, 2018 On 3/7/2018 at 11:14 PM, lorry said: With all respect... you have no skill in statistics, probability, or machine learning, and I don't care what you think. Expand If this is how you 'roll'... p.s. (false modesty, didn't buy it here... neither... )
Recommended Posts