Currently listening to a data science podcast that's like "okay there are legal restrictions on how targeted ads can be, and even stricter moral restrictions (here's a reminder why that's good), but,, if those weren't there how far could we go what's the limit" and it's really fun in a perverse way lmao. There's some really cool science being done on that question!! And the professor involved, the host, and me are all really grateful that it is entirely unusable lmao.
There's a cool digression there of like, okay we can actually get to the point where the question becomes sociological/psychological. Maybe even philosophical. Because,,, how many interests/traits can a person even have really
If one googles watches and watch reviews and subscribed to the watch subreddit then we can very safely assume they would be interested in ads about watches, and there's tons of work in finding the level of interest and how many ads to show based on that degree
But there's not really anything out there about how long after the last search it takes before we can safely assume that they no longer care about watches
(We can't see ones bank balance btw so this is why you'll see ads for things you already bought. We know you were interested, but don't know you made the purchase)
So you'll get profiles with hundreds even thousands of interests which,,, can a person even have half that many? What is the cognitive load? If someone can only have five or ten at a time then our models are over targeting to the point where we might as well not be targeting at all. It's a really interesting problem!
Also running along that is the question of, how do you tell when an interest is one someone is pained to have. How do you tell pride from shame - someone searching LGBT matters because they are out and proud, vs someone who is stumbling into it still holding shame. Who would love to hear about that upcoming gay party down the block this weekend vs who would be pained and even anxious to see that ad. There's no way currently! Tied to that is like, what information is dangerous to hold onto. When does one's search history become private health information protected by hippa? We theoretically have the tools for a snake oil salesmen to target people who would die from their products, that should be regulated but isn't. Or when ones illness is terminal, should we really have systems that remind them of this while scrolling. How do you tell people looking for treatments and wanting ads about that vs those for whom treatment isn't an option who would be hurt to see false hope?
Yeah no once you ask the question of what is the practical limit instead of what is the legal/moral limit you get into some really interesting territory. I'm really enjoying hearing this conversation
so we currently have microtargeting and the subject here is called nano-targeting, a degree of targeting that is strictly theoretical
a paper discussed was asking the question of how many interests on a profile do we need before the audience is one person, as in how many "this person loves watches, hates golfing, watches adventure time, ect" do we need before we can build a profile that is entirely distinct and individual. I want you to guess.
the answer is 43 if using whatever interests are available.
if you are able to be strategic and get the optimized interest metrics then it falls to 4.
it is theoretically possible, in an ideal world where you can cross reference every database available to us (which luckily currently do not speak to one another), to cut down an audience from 3 billion to a single user, with just four pieces of browsing data. luckily you would already know enough about them to make the issue moot but it is still distressing
there's a cute story in the paper of a couple who used this in order to send one another love notes through facebook ads lmao
but outside of that there are no use-cases for this that aren't terrifying haha
in a practical sense though it does illuminate the fact that there is diminishing returns on data aggregation, there need to be industry standards on limits for things like this so that you don't end up with only a few thousand users in the audience for a campaign. science is yet to be done on finding out what the optimal number is, where that balance is, but yeah it's needed.
aaaa this stuff is so cool haha