61–70 of 71 posts
Earlier quoted context omitted.
Accuracy is not the top objective in these systems, fairness is.
In this example, what is fairness, if not the most accurate prediction possible?
1. Not punishing someone for the sins of their family.
2. Not punishing someone for the unfair treatment that their family suffered in the past.
Earlier quoted context omitted.
If you take as a given that I’m going to see ads (as I do), I’d much prefer to see ads that are relevant and interesting or amusing to me.
So you're preferring the form of manipulation that works best on you? I'd prefer not to be manipulated at all. In fact I loathe the feeling of being manipulated so much that it extends to all but the least obnoxious brands that try to advertise to me. I can't be the only one who is allergic to advertising that tries to appropriate things from my social groups or background to sell me their shitty mobile data plan, in…
Earlier quoted context omitted.
This would entail no magazine/blog should also be targeting to a social group. i.e. no men's or women's magazine
What a silly argument, the magazine is not the advertiser. It's fine for content to target groups but not ads! You can have a men's magazine and the ads on there can target men, you can't have a men's magazine that sends out prints with ads targeting specific subsets of men. Content targeting is fine, user targeting is not.
"you can't have a men's magazine that sends out prints with ads targeting specific subsets of men."
Is anyone else trying to understand the distinction that's being made here? Does the "sends out prints" here differ in some vital way?
Earlier quoted context omitted.
The real problem is that there is an actual racial disparity in recidivism rates, so an algorithm that makes accurate predictions will predict the racial disparity that actually exists. There is no way to solve that without significantly impairing the accuracy of the predictions -- which is to say releasing convicts who we know have an unreasonably high probability of recidivism merely because there were too many oth…
most of the standard metrics of fairness for machine learning don't just just try to equalize proportions of positive/negative labels. they look at error rates. under these measures of fairness, a perfectly accurate predictor is regarded as perfectly fair, regardless of a disparity in base rates in the two populations. some of the predictive policing models still fail under these metrics -- they are more prone to mak…
Unless your predictor is perfectly accurate, the errors will be proportional to the base rate. If you're predicting that more X will do Y then you have more chances to be wrong.
Improving accuracy is the only real way to reduce the error rate. If you can't do that then you're left with malicious nonsense like fudging the base rate, which is just trading false positives for false negatives and not actually making anything better.
Earlier quoted context omitted.
Those regulatory concerns should fall onto the purchasers list of responsibilities. The EEOC has the authority to censure the parties misusing their advertising options. For example as a marketer who does work in all of the above industries , technology should not limit my use of my dollar. If i have 10,000 for an apartment campaign, i should be able to spend 5k on ads showcasing my apartment gym focused on men and 5…
Author of the underlying study here. Facebook has been insisting that non-discrimination should be the responsibility of the purchasers, but we've shown over [0] and over [1] again [2] that even when the advertiser targets all groups proportionally (no misuse of advertising options), Facebook subselects who to show their ads to in a skewed way, leaving the advertiser and the users no recourse. [0] https://arxiv.org/p…
Advertising industry has known this since ancient times. Military signals intelligence has been essentially built with that tidbit as its core. Even weak proxies are formidable if you have bucketloads of them to choose and combine from.
Now, let's make one thing clear: I am not a FB apologist. In fact, I find the modern advertising systems abhorrent, immoral and outright vile. But even then, this article felt like the authors chose to miss the point. Hiding information is incredibly difficult - and conflating "information theory is damn hard" with "FB[ß] are evil and/or immoral" feels intellectually dishonest.
For what it's worth, I would actually love to see research _and_ well sourced articles about the practical net effects of information theoretical attacks, intentional or not, on the human populations as observed through the various e-stalking platforms.
ß: I'm using "FB" here as a shortcut for FANG+MS+others.
Earlier quoted context omitted.
What a silly argument, the magazine is not the advertiser. It's fine for content to target groups but not ads! You can have a men's magazine and the ads on there can target men, you can't have a men's magazine that sends out prints with ads targeting specific subsets of men. Content targeting is fine, user targeting is not.
"You can have a men's magazine and the ads on there can target men" "you can't have a men's magazine that sends out prints with ads targeting specific subsets of men." Is anyone else trying to understand the distinction that's being made here? Does the "sends out prints" here differ in some vital way?
Earlier quoted context omitted.
What a silly argument, the magazine is not the advertiser. It's fine for content to target groups but not ads! You can have a men's magazine and the ads on there can target men, you can't have a men's magazine that sends out prints with ads targeting specific subsets of men. Content targeting is fine, user targeting is not.
So you are saying articles about best jobs targeted by gender will not create social bubbles, but an ad will do so?
I did not say anything about the content of articles, my comment was entirely about ads. If the article contains some topic and the ad targets that topic, the ad is not creating any new bubble. If the ad targets the user specifically because the user belongs to some group then ads are influencing specific groups in specific ways and forming new bubbles advantafeous to the highest bidder.
An ad for something men would like on a men's magazine or website targets men because they are already visiting a men specific content. An ad about the same thing on a random news site because the user is believed to be a man influences the user and other similar users to be influenced in the same way, even when they're not interacting with content specific to their membership of some group. This seems harmless but in reality the groups targeted are much more specific. So you may have those ads targeting "white men between 25-30,with college degree and living in an affuent neighborhood" , so in reality you now have that specific micro group being influenced differently, forming new bubbles. Men who don't also fit that criteria are not targeted, and are not influenced to buy nice undies. Replace undies with other more nefarious things (e.g.: food, medication,housing,job opportunities,etc...) And you can appreciate how ads are essentially micro-segregating people into bubbles based on statistical presumptions. Now if a usee visits a content,ads relevant to that content make a lot of sense. The problem is the user being targeted when they don't interact with relevant content and users interacting with relevant content not receiving relevant ads (e.g.: spouse reading about a gift for her husband won't see the nice undie ads).
Ads affect much more than what we buy, they affect out associations,peferences and views on subject matters. Non-consensual surveillance (stalking) should not be used to influence very specific groups of people. Now, I don't get why you have a problem with that?
Earlier quoted context omitted.
"You can have a men's magazine and the ads on there can target men" "you can't have a men's magazine that sends out prints with ads targeting specific subsets of men." Is anyone else trying to understand the distinction that's being made here? Does the "sends out prints" here differ in some vital way?
What part doesn't make sense. Sends out different prints containing targeted ads to specific subscribers as opposed to the advertiser targeting the content. Are you arguing to prove some point?
Earlier quoted context omitted.
What part doesn't make sense. Sends out different prints containing targeted ads to specific subscribers as opposed to the advertiser targeting the content. Are you arguing to prove some point?
The added "different" makes more sense. So a magazine wouldn't be allowed to send out different magazines for subscribers in New York versus LA?
Earlier quoted context omitted.
Accuracy is not the top objective in these systems, fairness is.
In this example, what is fairness, if not the most accurate prediction possible?
I should be judged based on the interpretation of my situation, not because someone who lives in a similar neighborhood was previously a bad bet.