Hacker News Reader

Comment by whack | Hacker News Reader

People are assuming that some evil mustache-twirling PM/engineer decided to charge users more if they have credits in their account. The truth may be far more banal. They may be using a ML model to determine pricing. The ML model takes as input all known information about that user. One of those inputs, which simply came bundled together with everything else, is the user's credit balance. The ML model, after much historical analysis, figured out that it can charge users more if they have credits available. The relevant engineers treated the ML model as a black box, ran an A/B test, saw a 1.3% increase in revenue, and decided to roll it out.

If this isn't already happening, I expect this will be the future of our industry.

View on HN

You might be right, but I'm failing to see how this is less evil mustache-twirling. A PM/engineer decided "I bet we can create an ML model that will find suckers we can charge more," is basically the same as "I bet people with credits will pay more." In either case, they're making a decision to find people they can screw over. In the latter case, it's based on intuition. In the former, it's based on statistics. However, the motivation remains the same.

If people are thinking this is evil, it's the motivation and the result that matter, not the means through which they achieve their goal.

And if they used ML as a black box, it could act in problematic ways. Maybe the ML figured out that it could charge Black riders more. They didn't look into it and just rolled it out. Things like that can happen when you don't look into stuff.

And I'd say that it's questionably legal. If a store charged you more money because you were using a gift card, I think we'd all agree that the store had unjustly taken value from you. Here, Uber is allegedly charging people with credits a higher amount - essentially stealing gift card value.

I'm not sure "oh, it was just our ML deciding to steal your money," is a great defense. Ok, but what was the ML supposed to do. Well, it was supposed to figure out who we could take more money from for the same product.

> A PM/engineer decided "I bet we can create an ML model that will find suckers we can charge more," is basically the same as "I bet people with credits will pay more."

The difference is the former has an element of being one step removed, and we use that tiny accountability gap to sleep better at night.

The comparison to modern army drone pilots comes to mind... we're now exploding Enemies from the comfort of a playstation controller in a base somewhere in Utah. The PTSD comes in when the operators realize that tiny accountability gap has just been mental gymnastics to suppress what they've really been doing.

_generally speaking_ most data science is "we tried this and it improved this metric" the how and why is often poorly understood, and even when it is, it's often later once the system has been working for awhile.

But also I'm not convinced that getting someone to pay more is the same as screwing them over. For example - countries where fines are based on your income, or needs based programs that charge you less when you make less are mostly accepted as social good

Those things are not inherently as arbitrary and are also much more transparent which makes them hardly comparable in any meaningful way..

This is like physical shops offering different prices to clients based on how they are dressed and other indirect indicators.

> And I'd say that it's questionably legal. If a store charged you more money because you were using a gift card, I think we'd all agree that the store had unjustly taken value from you. Here, Uber is allegedly charging people with credits a higher amount - essentially stealing gift card value.

Lots of things work this way though.

1) Airlines charge more to fly direct from A -> B while charging less for someone to fly from A -> B -> C. Some people try skiplagging to game the system but you run the risk of getting banned by the airline if you do that.

2) Another hilarious experience I have had is when I was paying things for my wedding. Some vendors (florists, cake makers, etc) will charge outrageous rates if you tell them it's for a wedding. But they will charge very reasonable rates if you claim it is just for a non-wedding family event (for the same amount of flowers, same size/layers of cake, etc).

3) A similar thing I have been warned about is electricians charging more if you claim you want an EVSE installed for your new Tesla as opposed to having just a regular NEMA 14-50 outlet installed in your garage. For the former, they will quote much higher while the quotes are cheaper for the latter (this is despite the exact amount of work and materials - you can even get them to confirm they are going to use the same gauge wire, etc).

I think they can do all of these tricks but once it becomes known that they are engaging in these shenanigans, it will dilute the value of their credits. So if Costco tries to sell me $100 Uber gift card for 20% off, then I will instantly value that 20% discount to be worth 0% and not purchase that gift card ever.

The act of feeding the model with data about users' credits necessarily means that they know it's using that as a factor in pricing. What difference does it make if there's an ML model involved?

At some point, the not giving a shit because it maximizes profit is all you need to be a bit judgy.

If they were doing something relatively frictionless that obviously increased efficiency? People wouldn't care.

Mindlessly doing what the model says, even though it's awful for the customer, to marginally increase profit some of the time by charging some customers more? People get pissed off...

That isn't that different than choosing to charge more for accounts with credits. Make "having credits" available to an ML model and then let the ML model "ethics wash" the decision.

> People are assuming that some evil mustache-twirling PM/engineer decided to charge users more if they have credits in their account.

> The ML model, after much historical analysis, figured out that it can charge users more if they have credits available. The relevant engineers treated the ML model as a black box, ran an A/B test, saw a 1.3% increase in revenue, and decided to roll it out.

This is the same thing. The fact that they might be turning a blind eye to why the ML model optimised the way it did doesn't absolve them from responsibility.

Thats almost the same thing isnt it? The motivation is to charge as much as possible to each user, how it's implemented isn't really relevant

Is it less evil if an unsupervised process does it?

It already is.

Yes, ML models are for accountability laundering. More news at 11.

This is exactly what it is. They allow you to be racist, exploitative, and discriminatory and then shrug when people start asking questions. When money is to be made people are geniuses, but when that money is questioned suddenly everyone is stupid. They don't know why it does what it does, and in fact nobody knows.

Reply on news.ycombinator.com