> "we were able to form a model to predict the personality of every single adult in the United States of America."
Edit: also this article is a good read about how they used Facebook likes to build up profiles. Also contains a summary of the video: https://motherboard.vice.com/en_us/article/mg9vvn/how-our-li...
The thing that they seem to bring to the table however is the massive amount of personalized data. That is the issue here, because that data can be abused by nefarious actors if it were to fall in the wrong hands. What usually takes a much more personal touch (like going out to voters, talking to them, or if you can't do that, have teams on the ground that talk to voters and tell you before your speech what they care about and why) can be done in mass due to essentially aggregating private data that users "agree to" obliviously or even don't consent to. That is the issue at hand, not like Cambridge somehow warped the people in Wisconsin's brains to mush and made them go to the polls.
Machine Learning models are tools for computing specific predictions. They are not thoughts, have no concepts, etc.
The model isnt what it is used for. A model isnt what it describes. The ability of an ML system predict a person's action isn't a reasoning process. A rock rolling down a hill predicts the path of water rolling down a hill. The rock isnt thinking about it.
Describing ML models as if they were cognitive gives people the impression that computers possess the relevant concepts and understanding which is why its seems scary. It's just current through a write to a fancier dial.
That would be dystopian if it weren't so ridiculous. They can do nothing of the sort- predict the personality of any adult in the USA. At best they might be able to make predictions on a population basis, but anything more precise than that is out of the question. Behaviour prediction on an individual scale is pure, unadulterated fantasy.
This is just typical overselling of a service, by people who can use maths to obfuscate the fact that they 're making it up as they go along, targeted to people who wouldn't understand the maths if a five-year old explained it to them anyway.
The access they have to online data sources would absolutely be enough to start correlating things to build an individual profile. Say you can link someone's FB, Twitter, (HN?!) account posts, likes, shares, to credit card data, maybe even search keywords from shady advertisers, to vehicle registration records, tax records, census data, etc.
Do you really think this profiling is that far-fetched that they need to BS everyone? It doesn't have to be perfect to be many times more effective than untargeted ads...
Impressive.
How do I know they weren't the reason Cruz lost? How do I know that they didn't lose the popular vote for Trump? Take their word for it?
No thanks.
If they even exist in the future, I guess...
Incidentally, the psychologist mentioned in the FB statement appears to have changed his surname to Spectre.
See: http://cpwlab.azurewebsites.net/CV/Aleksandr%20Kogan%20CV%20...
I can do a model of personality using ML/DL anytime as well, it's a piece of cake. How good the model is is another question; I'd be surprised if they were significantly better than random noise given how pathetic "science" of psychology is.
https://www.theguardian.com/technology/2017/oct/26/cambridge...
> Cambridge Analytica used its own database and voter information collected from Facebook and news publishers in its effort to help elect Donald Trump, despite a claim by a top campaign official who has downplayed the company’s role in the election. ... In another case, in the late stages of the November election, Schweickert said the company acquired data on voters who voted early – data it collected from local counties and states – and linked the information to individual Facebook profiles.