3 ms·

After GDPR our next move needs to be a advertising category whitelist. I say the EU makes a list of categories that you are allowed to categorize users in: -

After GDPR our next move needs to be a advertising category whitelist. I say the EU makes a list of categories that you are allowed to categorize users in: - Age (only year, no month or day this is not relevant). - Sex - City or geographic area with at least 500,000 people. - 5 items selected from a list of hobbies / preferences.. etc. Companies should simply not be _allowed_ to target people based on categories like being ill, being poor, being terminal, being depressed etc... And here's the kicker: if every company is held to these same standards no one company has a disadvantage. Well, except the entire ads industry scene which is now set in anarchy.

FlyingLawnmower 8y ago
I like the sentiment, but unfortunately predefining a set of categories doesn't prevent the issues. Even without access to a specific attribute like "wealth" or "race", the learning algorithms used here can still be used to target poor people, minorities, etc. by leveraging correlations within the dataset. Without these features, the bias just gets dispersed through other correlated variables, and makes it harder to detect that an algorithm is disproportionately targeting poor people. Additionally, even a modest list of attributes can be sufficient to uniquely identify someone. I think limiting collection is a good thing in general, but to meet your stated goals, we unfortunately have to do more.
dwighttk 8y ago
"Companies should simply not be _allowed_ to target people based on categories like..." But what if they just use good enough proxies for those categories? e.g. "X% of people who chose these 3 hobbies have a household income of..." you are just making it so they can't leave a paper trail that points to them saying some words along those lines. They can still come up with the theory in their head and test it and end up saying "we don't target people based on household income, we just advertise to people who chose these three hobbies."

FlyingLawnmower 8y ago

I like the sentiment, but unfortunately predefining a set of categories doesn't prevent the issues. Even without access to a specific attribute like "wealth" or "race", the learning algorithms used here can still be used to target poor people, minorities, etc. by leveraging correlations within the dataset. Without these features, the bias just gets dispersed through other correlated variables, and makes it harder to detect that an algorithm is disproportionately targeting poor people. Additionally, even a modest list of attributes can be sufficient to uniquely identify someone. I think limiting collection is a good thing in general, but to meet your stated goals, we unfortunately have to do more.

dwighttk 8y ago

"Companies should simply not be _allowed_ to target people based on categories like..." But what if they just use good enough proxies for those categories? e.g. "X% of people who chose these 3 hobbies have a household income of..." you are just making it so they can't leave a paper trail that points to them saying some words along those lines. They can still come up with the theory in their head and test it and end up saying "we don't target people based on household income, we just advertise to people who chose these three hobbies."