I enjoy the analogy of the small town gas stations. But I think it falls short in describing the problem of price segmentation, where a company offers different prices to different customers. Some countries have gone far in regulating sales as a response to this, as firms would set prices extremely high and offer a perpetual sale (furniture and cosmetics are really bad with this).
What worried me is when marketing signals are used to raise prices for me above what they should be. Airlines do this when you are searching for flights, based on browser, OS, and other signals.
However, if we ban price segmentation altogether, then we risk preventing businesses from offering sales, which serve some business and social utility (to borrow the term from the speech). But it's worth spending time thinking about what is ok and what isn't. Chances are that some businesses are already doing what isn't ok.
> My guess is that some point there will be an antitrust case featuring the "algorithm defense," which basically says: "Hey, I just set up the smart learning algorithm and let it run. How could I know that it would interact with other smart learning algorithms in a way that led to collusion?" And the antitrust authorities (or other law enforcement) will need to argue that when a guy named Bob sets up and signs off on an algorithm, Bob needs to be personally responsible for what that algorithm does.
How do you guys think this is going to be handled? It seems very complicated, because how are we even going to know which algorithm is at fault, or even IF any can be conclusively determined to be at fault?
Furthermore, is he talking about Bob the implementer or Bob the approver? In many organizations the people actually writing the code are not the same people who determined the specs and asked for it. Bob the implementer may have been given a limited spec that is followed, unknowing there is potential that an outside influence can manipulate the results. At that point who holds the responsibility?
Maybe it's just me, but recently I have noticed increased discussion regarding ethics in tech. It seems like there is growing awareness that all the focus on "how" to do something has eclipsed the fundamental question of "if" we should do it. This isn't to say we should just shy away from learning algorithms and complex systems, but maybe we should be put more of an effort to discuss the outcomes and how we agree to handle them before charging full steam ahead into uncharted territory.
Liability should start with the people deploying the system. They're the ones who have the fullest picture of how it's to be used. Contracts should be able to move that around, to some degree, but must do so loudly so that everyone knows what they're getting into - not buried in the fine print in "the standard boilerplate".
What about a more socially sensitive domain, such as college admissions, hiring, or setting pay? What if you put such an algorithm on said task -- to avoid human bias -- and later observe that the algo, say, does not hire <pick your group>?
Instead of Institutional Review Boards approving experiments we could have Algorithm Validation Boards. Hold out a portion of your dataset for validation, and have the algorithm designers examine the outcomes in communication with ethicists, lawyers, and business strategists.
I don't think the two hands can properly "validate" algorithms without communicating. The algorithm designer can maximize AUC, but what if one <group>'s class is 95% label A; the designer always predicts A for <group>. How bad is ALWAYS missing 5% for label B? If you can put a price on it, then the developer can build it into the algorithm. But if the price is difficult to accurately estimate, or non-monetary qualities are desirable, it may be hard to build them into the classifier ahead of time. On the other hand if the cost of perfect <hard to quantify criterion> reduces AUC significantly, algorithm designers need to communicate that...
That's already happening. Can't remember which article to link to, but machine learning algorithms make decisions that reflect pre-existing biases (e.g. harsher sentences for black people).
Humans produce the data that the ML algorithms train on, so without whitening the data somehow (and there's another contentious issue) we should expect the result to be biased as well.
> Finally, and most critically, there is complete price transparency because everybody can see the prices everyone else charges just by looking at those big signs. If we take away any one of those facts, the whole thing will generally fall apart on its own. For example, if firms could somehow secretly discount and steal market share from their rivals, they have a significant incentive to do that and so on. ...
This is why many big retailers use consumers as a watchdog, typically with offers like "we guarantee you the lowest price, if you find cheaper elsewhere, we'll give you back the difference times 2!". That way consumers think the retailer is having their back and protecting them when actually they prevent other retailers from lowering prices by keeping the market share. Consumers should not tolerate that bullshit and go with the lowest retailer in the first place and realize that they're being used to rat on the cheaper retailer... but there's a good incentive to get the difference * 2.
That only works on goods that have a high margin. Many times a company does that with a product, a competitor will notice and start selling the product at-cost: hence causing their competitor to loose money on the deal. Of course some then take that hit as a loss-leader, but that only works if the products are cheap. That why you'll often see "we guarantee you the lowest price, if you find cheaper elsewhere * excluding apple products * excluding online stores" etc. which I think most consumers notice and end up disregarding the "price guarantee" => however some don't, which I have a hunch might be the most profitable customers anyway....
The gas station scenario has a lot of similarities with the iterated prisoner's dilemma game; "defecting" here means lowering the price. The rational strategy is for the owners to cooperate (and retaliate if someone defects). In that case, the effects of price collusion can happen even without any actual collusion.
In reality, human beings rarely implement a strategy that's game theoretically optimal, but algorithms might.
With algorithms you can "actually" collude without humans conspiring with one another.
If pricing is public humans (or algorithms) can reverse engineer competitors using the public pricing data, the communication channel for that collusion _is_ the pricing data. Multiple parties doing the equivalent of prisoners' dilemma cooperating independently and secret from each other by reverse engineering each others pricing algorithm is exactly the same as colluding in a smoke filled room.
Personalized pricing, with great details described how it works in retail in era of face recognition, mobiles, web tracking,loyalty cards, data mining etc.etc.
Recommend the read to anyone
The Aisles Have Eyes: How Retailers Track Your Shopping, Strip Your Privacy
I find it interesting that conditions unfavorable to the buyers can arise in various ways without explicit communications between the sellers. For example, this happens if the sellers update prices faster than the buyers can compare them: https://phys.org/news/2012-06-high-gas-prices-self-organized... .
13 comments
[ 3.0 ms ] story [ 47.8 ms ] threadWhat worried me is when marketing signals are used to raise prices for me above what they should be. Airlines do this when you are searching for flights, based on browser, OS, and other signals.
However, if we ban price segmentation altogether, then we risk preventing businesses from offering sales, which serve some business and social utility (to borrow the term from the speech). But it's worth spending time thinking about what is ok and what isn't. Chances are that some businesses are already doing what isn't ok.
How do you guys think this is going to be handled? It seems very complicated, because how are we even going to know which algorithm is at fault, or even IF any can be conclusively determined to be at fault?
Furthermore, is he talking about Bob the implementer or Bob the approver? In many organizations the people actually writing the code are not the same people who determined the specs and asked for it. Bob the implementer may have been given a limited spec that is followed, unknowing there is potential that an outside influence can manipulate the results. At that point who holds the responsibility?
Maybe it's just me, but recently I have noticed increased discussion regarding ethics in tech. It seems like there is growing awareness that all the focus on "how" to do something has eclipsed the fundamental question of "if" we should do it. This isn't to say we should just shy away from learning algorithms and complex systems, but maybe we should be put more of an effort to discuss the outcomes and how we agree to handle them before charging full steam ahead into uncharted territory.
I don't think the two hands can properly "validate" algorithms without communicating. The algorithm designer can maximize AUC, but what if one <group>'s class is 95% label A; the designer always predicts A for <group>. How bad is ALWAYS missing 5% for label B? If you can put a price on it, then the developer can build it into the algorithm. But if the price is difficult to accurately estimate, or non-monetary qualities are desirable, it may be hard to build them into the classifier ahead of time. On the other hand if the cost of perfect <hard to quantify criterion> reduces AUC significantly, algorithm designers need to communicate that...
Humans produce the data that the ML algorithms train on, so without whitening the data somehow (and there's another contentious issue) we should expect the result to be biased as well.
> Finally, and most critically, there is complete price transparency because everybody can see the prices everyone else charges just by looking at those big signs. If we take away any one of those facts, the whole thing will generally fall apart on its own. For example, if firms could somehow secretly discount and steal market share from their rivals, they have a significant incentive to do that and so on. ...
This is why many big retailers use consumers as a watchdog, typically with offers like "we guarantee you the lowest price, if you find cheaper elsewhere, we'll give you back the difference times 2!". That way consumers think the retailer is having their back and protecting them when actually they prevent other retailers from lowering prices by keeping the market share. Consumers should not tolerate that bullshit and go with the lowest retailer in the first place and realize that they're being used to rat on the cheaper retailer... but there's a good incentive to get the difference * 2.
In reality, human beings rarely implement a strategy that's game theoretically optimal, but algorithms might.
If pricing is public humans (or algorithms) can reverse engineer competitors using the public pricing data, the communication channel for that collusion _is_ the pricing data. Multiple parties doing the equivalent of prisoners' dilemma cooperating independently and secret from each other by reverse engineering each others pricing algorithm is exactly the same as colluding in a smoke filled room.
Recommend the read to anyone
The Aisles Have Eyes: How Retailers Track Your Shopping, Strip Your Privacy
[0] https://www.amazon.com/Aisles-Have-Eyes-Retailers-Shopping/d...