First gonna list the two papers of Kosinski in 2013 that generated lots of talking.
Private traits and attributes are predictable from digital records of human behavior (2013), by Michal Kosinski, David Stillwell, and Thore Graepel.
... Facebook Likes... predict a range of highly sensitive personal attributes including: sexual orientation, ethnicity, religious and political views, personality traits, intelligence, happiness, use of addictive substances, parental separation, age, and gender... dataset of over 58,000 volunteers...Manifestations of user personality in website choice and behaviour on online social networks (2013), Michal Kosinski et al.
... a sample of over 1/3 million users... their website choices and Facebook profile features, relates to their ... standard Five Factor Model personality questionnaire.Wait... DEVIANTART???
It also found that the visitors to the following websites have similar personalities... Table 4:
- deviantART.com
- Tumblr.com
- Etsy.com
- Gaiaonline.com
- Fanboy.com
- RainyMood.com
19:40 shows two "average selfies". The extroverted selfie has no nostrils, because it's taken from a high angle (common trick in selfie).
Page 48 shows faces that are "dreamed" by the neural network. Basically the idea is to "reverse" the network so that instead of taking a face and giving a judgment ", it takes a judgment and returns a face that best fits the judgment.
I'm amused by how there's shades of glasses on the introverted faces.
Notice how all the dreamed up faces are beautiful, which is explained by "koinophilia", or "average is beautiful". They are dreamed up, so they are pretty much what you'd get by averaging the examples.
The dreamed-up gay man face is notably feminine. It seems there's quite some truth in the stereotype that gay man are feminine.
Page 63: Accuracy is lower for females. It means female sexual orientation is less shown in their faces, or perhaps there are more females with fluid orientation that doesn't fit either sexual orientation closely enough to be predicted?
Automatic Personality Assessment Through Social Media Language (2015), Park G et al has nice pictures about what words are predictive of personality:
Answer: Left is 外向的, right is 内向的.
In 2016, Cambridge Analytica got famous for manipulating the US election. A presentation was given by the company here: Cambridge Analytica - The Power of Big Data and Psychographics
- The three methods of election manipulation are Behavioral messaging, Data analytics, and Addressable ad tech.
- Election campaign ads should be catered according to psychometics (OCEAN in particular), not demographics (gender, age, etc).
- Behavior messaging: sending messages according to their personality.
- Data analytics: using data made by people to predict their psychology.
- Demographics and psychographics predicts personality.
- Addressable ad tech: personalize ads for everyone, no mass ads.
- Ted Cruz used the methods to get approval from 5% to 35% in a year. Donald Trump used it to win the election.
What's going to be like in the future, with these algorithms with a very good understanding of human emotions and personalities?
Something like a Pinkie Pie, watching us forever, trying to be privy of every little personal detail and mood of everypony...
No comments:
Post a Comment