[SkyBelow]

I could see an AI trying to hunt out a person's bottom line. I could offer this job for $10 to everyone, but maybe I'll subtract 0 to 4 dollars when I offer it and see who does or doesn't bite. If someone bites on lower pay, I then record that information and offer them further lower pay in the future.

This isn't really abnormal. Every job does this by setting a wage they are willing to pay and seeing who signs up, knowing that person will now need to only be paid that wage. What is different is the scale and the frequency this is being done. Instead of doing this in a way that impacts a person once every job change, it now impacts them multiple times a day, and the data recorded is more detailed and can be acted on more directly.

None of this is discrimination against a protected class, but if there are any reasons one demographic might, on average, accept lower pay than another, it will lead to large scale discrimination.

The problem is that our common discussion on these topics is lacking the rigor, nuance, or depth to handle questions about this, and thus ends up with two large camps. One that looks at the methods, sees no obvious discrimination in the methods, and say it doesn't count as inequality. The other that looks at the outcomes, notices the clear difference in outcome this leads to, and calls it inequality. Both are, by their own metrics, correct.

[mananaysiempre]

“Price discrimination” (or in this case “wage discrimination”) as described in microeconomics is exactly this—the same seller/buyer demanding/offering different prices for the same goods depending on their idea of how much the buyer/seller will bear. The term has nothing to do except etymology with what sociologists, lawyers, or politicians mean by the word “discrimination” (not that those three groups mean the same thing by it).

[SkyBelow]

The issue is that many small scale price discriminations on individually reasonable criteria might present itself as a large scale discrimination of the type that lawmakers and others do care about. The way terms are overloaded does no favors, but even if we updated the terminology to resolve this, I think the underlying issue will remain.

Pink tax is an example of this happening, though on a scale needing far invasive technology than is currently available. It is presented as (big) discrimination even though it happens as price discrimination.

[vundercind]

It’s more than that, I think: if this paper holds up (or if it doesn’t, but the ideas it covers are valid and the practices it’s concerned with later come into being) then it’s describing a mechanism for pushing down worker wages at the individual level, and within potentially any or all bands of the economy toward the market-clearing rate per worker. A market of many workers becomes many markets of one worker.

This is, um, potentially really bad. It’s several effects that already happen in, if you will, chunkier ways in our economy (especially in the US, with weak or absent unions and poor labor protection laws, compared to many other developed states) becoming applied at a much finer level of resolution (so to speak).

[Kon-Peki]

Stay tuned for my new app: Wildcatr

It is installed on gig worker phones and monitors the offered rates. When one worker is offered abusive rates, all other workers have their future offers filtered from view for some period of time unless it exceeds the typical offer by more than the amount the abused worker missed out on.

[DeliriousDog]

The issue isn't the "hunt for the bottom line" but the fact that simultaneously multiple parties are offered different price points for an unknown reason (to the workers).

You say it's not discrimination, but you cannot definitively make that claim. That's the issue. Red lining isn't immediately discrimination against a protected class, but silently is it. This is not to say that Uber/Lyft are discriminating against a protected class - it's just that because of the lack of transparency we don't know that they are not.

This is a hard thing for people to accept, but we need to take a deep look at how we implement ML to classify things tied to individuals. It's very easy to de-humanize the humans affected by the systems we build, because "it's just an algorithm."

[Aloisius]

> it's just that because of the lack of transparency we don't know that they are not.

Is this not the case regardless of whether an algorithm is used or not?

[marcosdumay]

Setting labor price by exchange-like auction is an abusive practice in any context.

Companies get a pass on job interviews because it's basically impossible to prove. But this doesn't make it ok, it just makes it less damaging than the remedy. (Or at least arguably, a lot of people do argue otherwise, and lots of people are looking for better remedies.)