I normally wouldn't even comment on such a poorly reasoned out article, but given the climate, I don't think it does anyone good to hold back when trying to explain things. If you're intimately a part of tech implementation, I encourage you to speak up often, and politely to defuse these sort of inflammatory pieces.
To be clear, practically no one sits down and goes "How can we discriminate today?" outside possibly a few rumored actors I've heard of. This is critically important to understand if you actually want to get to the core of how we unintentionally end up with seemingly laughably biased systems. It is almost always through second or third order effects through which thus type of thing slips in.
Case and point:
Facial recognition would go something like this:
"Hey, we can use this neural network thing to recognize faces. What can we do with that?"
"Well, let's see if it can find surprising trends by feeding in our datasets, plus some control images to revolutionize the way we market to our customers! We have their Facebook or email, let's just grab a picture associated with that and see what happens."
<...something marginally useful happens...>
<...rumors spread of something marginally useful happening when applying this technique in this case...>
<...multiple companies try to eke out some competitive edge by experimenting with their own variations...>
<...a data scientist gets interested in all this hubub, and does a meta-analysis, finding some flaw, or how the entire technique basically degenerated to some form of wealth concentration or population based heatmap in a really difficult for regular people to understand way...>
<...media reports that technologists are evil racist masterminds due to the coincidental segmentation of disadvantaged populations...>
All of it coming out of a "Huh, I wonder if that would work here?"
Or you get another one, where you end up marketing to a particular demographic, but without realizing it, the thing you market is in and of itself part of the reason the demographic is what is, and doesn't foster much customer migration out of. An example being businesses operating on the basis of "poor people fees" as I call them.
Businesses set out to make tools that help people not be poor, but the income stream is structured around collecting these types of fees from their target market segment. They are looking a it business performance metrics and doing things good for business. That something good for business done for the best of reasons could in the end be bad is not a realization that a lot of people end up putting in the work to arrive at.
Technology very rarely solves problems they aren't directly designed to solve. What it is really good at though is amplifying what problems already exist within the axiomatic space in which they exist. In the words of Einstein "We can not solve our problems with the same level of thinking that created them."
Most tech of the last few decades has actually been designed to:
-Enable transactions that were previously untenable to facilitate given the reliance on physical means of exchange
-surveillance/remote monitoring
-accountability laundering
-penny-pinching and fiscal optimization
Once you understand that finance has firmly decoupled from even the most remote semblance of correlation to actual utility created in anything else other than wealth movement; you'll see why it seems like nothing ever gets solved.
The problem isn't racism. It's your money is in your pocket, and we want it in our coffers with the largest ratio of value gained vs our value list that the market will allow. That simple.
3 comments
[ 3.2 ms ] story [ 20.2 ms ] threadApparently "Technology Review" also wants to cash in on the current "Black Lives Matter" hype-train with this pay-walled article.
Has marthastewart.com already taken a political stance?
To be clear, practically no one sits down and goes "How can we discriminate today?" outside possibly a few rumored actors I've heard of. This is critically important to understand if you actually want to get to the core of how we unintentionally end up with seemingly laughably biased systems. It is almost always through second or third order effects through which thus type of thing slips in.
Case and point:
Facial recognition would go something like this:
"Hey, we can use this neural network thing to recognize faces. What can we do with that?"
"Well, let's see if it can find surprising trends by feeding in our datasets, plus some control images to revolutionize the way we market to our customers! We have their Facebook or email, let's just grab a picture associated with that and see what happens."
<...something marginally useful happens...>
<...rumors spread of something marginally useful happening when applying this technique in this case...>
<...multiple companies try to eke out some competitive edge by experimenting with their own variations...>
<...a data scientist gets interested in all this hubub, and does a meta-analysis, finding some flaw, or how the entire technique basically degenerated to some form of wealth concentration or population based heatmap in a really difficult for regular people to understand way...>
<...media reports that technologists are evil racist masterminds due to the coincidental segmentation of disadvantaged populations...>
All of it coming out of a "Huh, I wonder if that would work here?"
Or you get another one, where you end up marketing to a particular demographic, but without realizing it, the thing you market is in and of itself part of the reason the demographic is what is, and doesn't foster much customer migration out of. An example being businesses operating on the basis of "poor people fees" as I call them.
Businesses set out to make tools that help people not be poor, but the income stream is structured around collecting these types of fees from their target market segment. They are looking a it business performance metrics and doing things good for business. That something good for business done for the best of reasons could in the end be bad is not a realization that a lot of people end up putting in the work to arrive at.
Technology very rarely solves problems they aren't directly designed to solve. What it is really good at though is amplifying what problems already exist within the axiomatic space in which they exist. In the words of Einstein "We can not solve our problems with the same level of thinking that created them."
Most tech of the last few decades has actually been designed to:
-Enable transactions that were previously untenable to facilitate given the reliance on physical means of exchange -surveillance/remote monitoring -accountability laundering -penny-pinching and fiscal optimization
Once you understand that finance has firmly decoupled from even the most remote semblance of correlation to actual utility created in anything else other than wealth movement; you'll see why it seems like nothing ever gets solved.
The problem isn't racism. It's your money is in your pocket, and we want it in our coffers with the largest ratio of value gained vs our value list that the market will allow. That simple.