Hacker News Reader

Comment by wiremine | Hacker News Reader

This story mostly regurgitates this Politico story:

https://www.politico.com/news/2022/09/02/amazons-ftc-problem...

The key paragraph:

"Among the concerns the FTC is investigating is whether the data generated about a consumer’s home by iRobot’s Roomba vacuum will give it an unfair advantage over a wide variety of other retailers. That concern was voiced by multiple critics of the deal when it was first announced early this month. For example, Amazon could have an advantage with a consumer looking to buy a couch, by using detailed home maps generated by iRobot to suggested particular items."

How detailed are the maps? I'm assuming they're using SLAM. [1] If so, the maps aren't that detailed, and I struggle to see how the maps will give them an unfair _retail_ advantage.

I do get the need for the review, it's just this point that doesn't make sense to me.

[1] https://en.wikipedia.org/wiki/Simultaneous_localization_and_...

View on HN

This is Amazon, the company which still offers to sell me more vacuum cleaners after I buy one.

The company which will completely ignore keywords like "6A" in a desperate attempt to show you anything but what you're looking for.

I'm not so worried!

Added: this is also the company which sells a popular line of home-security cameras, some of which are meant to be installed indoors. Worrying about your vacuum snitching on you makes good sci-fi but maybe this product isn't where regulators should be focusing.

vacuums need friends!

I have 4 and should probably buy that wet/dry shop vac I've been meaning to.

The reason after you make a purchase you still see offers for the same product is because you are still statistically more likely to purchase a vacuum than average. What if the vacuum doesn’t work well, it’s too heavy, the package doesn’t arrive? You’ll be turning around and buying another.

Would mainly make sense if I returned the original order, otherwise there must be much more efficient use of the recommendation space.

Well, no there isn't, and this company has the best data available on the planet to prove it.

There should be a corollary to Hanlon's Razor. Something like, "Don't attribute to genius that which is adequately explained by laziness."

Frankly, calling it Bezos's Razor would be very appropriate, because that's basically how the whole Amazon storefront works.

Not quite. ML algorithms that continuously crunch billions of data points don't need to be called "genius", they're simply way more exhaustive in their search of optimums than "a vague idea from a regular person on the Internet that the best reaction to such situation is to do x" (with absolutely zero evidence to show for it of course).

AIUI, recommendations can come from multiple providers. If very detailed recommendations that take into account the fact that you’ve already bought a vacuum timeout, then your recommendations come from a dumber provider that is based on recent searches.

By “provider”, do you mean internal Amazon recommendation services?

Just because they are inept now doesn't mean we should allow them something that could turn into an unfair advantage once they figure out how to actually do it.

So your argument is Amazon is too dumb to be a threat? Even if we assume this is true for the present, seems very short-sighted to me.

Yes but.... Amazon would be happy to sell your data to a company who knows how to abuse it!

How detailed are the maps?

Pretty detailed. I used to own one of the high-end Roombas that would show you a map of your house, and the wifi signal strength in all the nooks and crannies (really quite useful when you move to a new home).

It doesn't show what kind of chair you have or what you have on the walls, or anything like that. But it's not hard at all to determine the size of the rooms, the layout, the placement of key pieces of furniture, and similar things. From the way a home is arranged and the amount of furniture, I can easily think of ways that the map could be used to infer information about a home. (Larger couch and entertainment center may indicate someone more interested in consuming media. Individual chairs around a smaller coffee table — perhaps they like entertaining.)

The maps are far from accurate, since there are a number of problems with a Roomba building a full map: Closed doors; areas cordoned off by lighthouses; unless you put blue painter's tape over the sensors, Roomba thinks rugs with black in them are cliffs.

But the maps don't have to be prefect to be very revealing. Especially since the higher-level Roombas have cameras on them. I don't know how detailed the images they capture are, but even if they are only vague monochrome blurs, that's enough to determine patterns of activity in the house. (Family regularly comes home at 5pm? Start advertising dinner food on partner web sites and apps.)

Even if the maps are basic, it would allow Amazon to jack up the prices for amazon.com queries from the same IP address if the place, say, is really big. iRobot app also encourages naming all areas in the house, so that too can be pretty telling.

In fact, iRobot requires the access to the location services to initialize their iOS apps, so even without a map this would allow Amazon to link an IP address to the exact property. Creative abuse of this combo can go a very long way.

Although I agree with the general concern

> it would allow Amazon to jack up the prices for amazon.com queries from the same IP address if the place, say, is really big.

Afaik this sort of airline-style price discrimination is not something amazon engages in ever. Maybe I’m misinformed

I think rather than misinformed, you just left out a word in your claim. If you added the word "yet", then it would seem more appropriate to me.

If the advertising world believes that targeted ads are the holy grail, then why wouldn't retailers believe targeted pricing to be also be the holy grail?

Targeted pricing is the holy grail, it’s just normally done through market segmentation and targeted advertisement.

My personal belief is that if Amazon tried what you’re suggesting, they’d be shooting themselves in the foot (lose any remaining customer trust, regulators would be all over them, and competitors like Walmart would use it to gain market share overnight). In fact I was really hoping you’d share an article describing Amazon doing this so I have justification to cancel prime, ha!

Honestly, I was just going all 1984 on the concept.

However, if Amazon was really wanting to do this, they already know the avg income on everyone of their customers that have an Amazon credit card. That would be the ultimate in targeted pricing. Not just Amazon, any store branded card.

They tried it before, it was quickly noticed, and they apologized. But since then, they've started doing it with Prime membership, offering some deals only to Prime members while everyone else pays full price. That's not quite the same thing but certainly is hanging out in the same neighborhood.

That's not the same thing at all. It's a whole service, advertising these benefits explicitely, for which you pay explicitely.

Because it needs Bluetooth and/or WiFi to connect to the vacuum. Those both require location privileges

Wait, what? An app can't get access to Bluetooth without also getting my location? That seems like a bit of an oversight

Other way around. An app with Bluetooth/Wifi access can determine your approximate location by seeing what devices are nearby. So the OS asks you to grant location permissions since, in effect, that is what you're granting the app.

Do you have neighbors? The Roomba may be able to connect to the internet even if you never give it your wifi password or pair it with your phone.

https://en.wikipedia.org/wiki/Amazon_Sidewalk

What does that have to do with requiring location services to configure the app?

It means they could eliminate that requirement but it wouldn't be a meaningful improvement to privacy, because the roomba would still know where it is.

Amazon has your shipping address. Given an address, I can get the size of your house, the number of floors, room distribution and price from a few API calls, no ML required.

I have a feeling that I lot of people don't appreciate the degree of information about them that is public as a matter of policy and law. It's cheaper and easier to obtain today than a few decades ago but much of it has long been public.

Of course, availability through an API can greatly expand how the information is used but, for the most part, it's tricky to say that data is public but you have to make it painful to access.

Price discrimination based on home dimensions is illegal in Europe, so most people here are fine with some of their info landing on a publicly accessible cadastre database.

The issue probably isn't the final map but the raw data used to generate it. Which in the case of the newer iRobots is cameras. So the question is, how much is extensive continuously updated photos of everything in your house worth to Amazon?

For example, you search for decorative pillows and they filter based on those that match the style and color scheme of your living room. That would be an advantage over other retailers, and is just one simple example. OTOH implementation is everything and that wouldn't work well if you just ordered new furniture from somewhere else and are looking for pillows to match that.

If they could actually provide filters like that people would give them pictures of their couch and living room.

Yeah but this sounds like an awesome service, actually

> How detailed are the maps? I'm assuming they're using SLAM. [1] If so, the maps aren't that detailed

A few years ago I would have probably agreed. But modern machine learning can draw spooky inferences from seemingly sparse data, on a scale that makes it practical. As the roomba bumps around a room it will learn where obstacles are but not what those obstacles are. But the size and shape of the footprint of those obstacles probably contains enough information to make guesses at what those might be. Approximate identification of some obstacles as couches is probably possible, just from the size and shape, and likely the position of those obstacles in the room.

That won't tell you how old the couch is, or whether it's worn out or not, but if the roomba can't find any probable couches in that person's house, advertising couches to that person may pay off.

Also, did the roomba start finding a new room recently, but the other rooms have remained the same? That person may have build an addition to their house, or started furnishing their basement. Did all the rooms change but the roomba is still paired to the same smartphone? They may have moved into a new home. In either case, if the new room[s] don't have couch-shaped obstacles, advertise couches.

And this is just with the bumpers. If Amazon adds a camera to roombas, the snooping potential is obvious.

Yep, pretty much exactly this. I've spent some time with SLAM, and in my experience, it's usually just a part of a larger ML pipeline. Not only can SLAM be paired with external data sources to generate labeled 3D objects, but inferences can be drawn directly from a generated point cloud without any ML involved.

My best guess for this is that floor plans and furniture can be used to _fingerprint_ a user, rather than knowing that the user owns a couch so that The Company can recommend a particular CPG. Knowing who their target demographics are with greater accuracy certainly feeds back into their recommender systems, but not so directly, which makes me concerned for this investigation's success.

Even occupancy grids can be used to detect if you some kind of furniture or what kind of windows you have. Also, iRobot added visual slam some time ago, those maps can give huge amount of detail even if they don't contain images, just feature based bag of words models.

If you're talking about SLAM you're overanalyzing the situation. Essentially, unfettered decades of rapacious corporate abuse has created a group of people who are essentially skeptical of anything that big business does. FTC Commissioner Lina Khan is one of those people and is now coming at Big Tech with a broadsword. Because Big Tech is so abusive, most of this makes sense, but much of it does not (eg. blocking a small VR app acquisition by Facebook)

Huh? As long as they have pictures, they can just run a classifier on it to figure out if you have couch etc. SLAM or not doesn't matter.

My map is very detailed. Down to a few inches. It also has me label rooms, so it knows my Living Room vs Bedroom vs Bathroom.

Just because the maps aren’t detailed enough for this use today doesn’t mean Amazon won’t add the ability going forward

Reply on news.ycombinator.com