7 ms·

I went to uni with the guy who didn't share his stuff with Cambridge Analytica (Kosinski). He presented the likes -> OCEAN analysis and it was pretty compelling

by kdelok 9y ago

I went to uni with the guy who didn't share his stuff with Cambridge Analytica (Kosinski). He presented the likes -> OCEAN analysis and it was pretty compelling (and frightening) stuff. In particular, with enough likes, it could deduce things such as gender, race and sexual orientation. Obviously in the wrong contexts, this information is extremely sensitive.

AznHisoka 9y ago
Gender and race doesnt seem very farfetched. You can get halfway there with just the name of a person. Add location and you’re almost there.
- EB-Barrington 9y ago
  FYI - I should have added, my name was never given, nor was any other "personal" information. Nothing more was provided other than a paragraph of text I had previously written, the contents of which had nothing to do with me personally.
  - vbo 9y ago
    They might've extracted useful information from your IP address/geolocation or your useragent, ie maybe figuring out you're accessing the site from tech company X's network. Given tech companies ratio of male to female employees, they may have been able to infer you are male. Also, tech companies pay well, so that's another information nugget about you, etc. The more signals they have, the more precise the outcome.
    - EB-Barrington 9y ago
      Yes, agree. So I should add, just for discussion: 1) I'm self employed. 2) I'm always connected via a VPN connected to a continent that isn't my home continent. 3) I haven't connected to the internet via a "company" network in over five years. I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling". Given millions of paragraphs of text linked to millions of known demographic and other data points (provided by Facebook), a paragraph is all a well-tuned system would need to make quite accurate assumptions about the person who typed the paragraph. Again, likes are not required anymore.
      - RyanZAG 9y ago
        > I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling". I think you only really need one or two obvious tells for something like that. The guy you replied to used ", etc.". I think that alone puts him into the bracket you've given, so I'm fairly sure you're correct here. Looking forward to hearing if you are.
      - vbo 9y ago
        You could also do sentiment analysis on your dataset and link it to likes. Basically the more signals you have the more you can narrow down your classifications or correlations. Btw, your inference is correct. Course, one could argue that it is informed by 25-35 males being one of the most common demographics on HN.
EB-Barrington 9y ago
I remember posting a random paragraph of text I had previously created into an "analyser" on the Cambridge Analytica website - as an alternative to giving Cambridge Analytica access to my FB profile. It then accurately deduced my gender, race, sexual orientation, age group, and other key data - all from a single paragraph. No "likes" required anymore.
- laretluval 9y ago
  Would it have been inaccurate for you if it was just guessing based on base rates?
pacala 9y ago
OCEAN stands for Openness, Conscientiousness, Extroversion, Agreeableness and Neuroticism. Also known as https://en.wikipedia.org/wiki/Big_Five_personality_traits https://en.wikipedia.org/wiki/Big_Five_personality_traits.
- narrowingorbits 9y ago
  *Conscientiousness
  - pacala 9y ago
    Fixed, thanks!
mc32 9y ago
Maybe the "genius", if it can be called that was FB only allowing likes, rather than dislikes. You have to really agree with something to like it, as the default is to not react, and when you do react, most energy is to disapprove of something (for example, negative feedback, online reviews, etc.), so if someone gives something positive feedback it must have nonmarginal significance. That said, the conclusions are very much expected. If i know what you like, i can know what kind of person you might be.

by kdelok 9y ago

AznHisoka 9y ago

Gender and race doesnt seem very farfetched. You can get halfway there with just the name of a person. Add location and you’re almost there.

EB-Barrington 9y ago
FYI - I should have added, my name was never given, nor was any other "personal" information. Nothing more was provided other than a paragraph of text I had previously written, the contents of which had nothing to do with me personally.
- vbo 9y ago
  They might've extracted useful information from your IP address/geolocation or your useragent, ie maybe figuring out you're accessing the site from tech company X's network. Given tech companies ratio of male to female employees, they may have been able to infer you are male. Also, tech companies pay well, so that's another information nugget about you, etc. The more signals they have, the more precise the outcome.
  - EB-Barrington 9y ago
    Yes, agree. So I should add, just for discussion: 1) I'm self employed. 2) I'm always connected via a VPN connected to a continent that isn't my home continent. 3) I haven't connected to the internet via a "company" network in over five years. I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling". Given millions of paragraphs of text linked to millions of known demographic and other data points (provided by Facebook), a paragraph is all a well-tuned system would need to make quite accurate assumptions about the person who typed the paragraph. Again, likes are not required anymore.
    - RyanZAG 9y ago
      > I could infer that you are male from the way you type, and I guess you're somewhere between 25 and 35. I haven't looked up your username or anything else, it's just a "feeling". I think you only really need one or two obvious tells for something like that. The guy you replied to used ", etc.". I think that alone puts him into the bracket you've given, so I'm fairly sure you're correct here. Looking forward to hearing if you are.
    - vbo 9y ago
      You could also do sentiment analysis on your dataset and link it to likes. Basically the more signals you have the more you can narrow down your classifications or correlations. Btw, your inference is correct. Course, one could argue that it is informed by 25-35 males being one of the most common demographics on HN.

EB-Barrington 9y ago

I remember posting a random paragraph of text I had previously created into an "analyser" on the Cambridge Analytica website - as an alternative to giving Cambridge Analytica access to my FB profile. It then accurately deduced my gender, race, sexual orientation, age group, and other key data - all from a single paragraph. No "likes" required anymore.

laretluval 9y ago
Would it have been inaccurate for you if it was just guessing based on base rates?

pacala 9y ago

OCEAN stands for Openness, Conscientiousness, Extroversion, Agreeableness and Neuroticism. Also known as https://en.wikipedia.org/wiki/Big_Five_personality_traits https://en.wikipedia.org/wiki/Big_Five_personality_traits.

narrowingorbits 9y ago
*Conscientiousness
- pacala 9y ago
  Fixed, thanks!

mc32 9y ago

Maybe the "genius", if it can be called that was FB only allowing likes, rather than dislikes. You have to really agree with something to like it, as the default is to not react, and when you do react, most energy is to disapprove of something (for example, negative feedback, online reviews, etc.), so if someone gives something positive feedback it must have nonmarginal significance. That said, the conclusions are very much expected. If i know what you like, i can know what kind of person you might be.