4 ms·

Just to elaborate a bit because this is a pet-peeve. What's going on here is that people are misdirected. They've seen a bunch of sci-fi where physical robots

by mlsu 4y ago

Just to elaborate a bit because this is a pet-peeve. What's going on here is that people are misdirected. They've seen a bunch of sci-fi where physical robots moving through space can sense the world around them and use the information they gather to make the world legible to them. The assumption (fair, because moving robots look like us!) is that that is the easiest way to gather information is to physically See, Touch, and Hear it. Sophisticated robot system that looks scifi -> this must be the leading edge of the cyberpunk dystopia. The reality is completely different. Your life is already legible to the computer. For the purposes of whatever cyberpunk dystopia you imagine we are in, your 1kb credit card statement is like an 8K Imax camera to the Ministry for Coercion. The leading edge of the cyberpunk dystopia is the apparatus that tries to get you to be more online. Write more posts and emails, sign up for smart banking, use an internet system to keep track of your employees, digitize paper files and make them legible. Why would any intelligent malignant thing use the extremely power hungry and inefficient raw input from a physical robot for its purposes when it could use the extraordinarily efficient raw text format of bank statements, insurance claims, credit reports, browser history... Every piece of information that you imagine could be used against you in the hypothetical dystopia hinted at in this piece would fit inside of 100 megabytes of plain text. Plain text that you, or an entity working for your insurance company, typed into a computer. They already have it. These things -- scifi-looking robots -- really are just trying to clean your house better and sell you products more cheaply.

godelski 4y ago
I kinda agree with you, but also not. I think a lot of people here are over assuming what the robots can sense and yes, this is that Sci-Fi like reality. But this doesn't mean we should throw the baby out with the bath water. Anyone that has seriously worked with data will tell you how subtle things can reveal a lot of information. Let's look at these robots. Knowing that there are pets and/or children in a house is useful data to sell kids/pet toys and other items. You might think "Amazon already has that info from purchasing history" but a second source builds higher confidence and can differentiate if something was bought for yourself or for a friend. There's also information like how much the roomba fills up. Are you a messy person or very organized? Do you take off your shoes at the door? When you sit down at the couch? At all? Do you make frequent spills or messes in certain rooms? This obviously has selling potential. You don't need microphones or super sensors. Basic information can do a lot. There's also a lot to be said about redundancy and verification. I think you're over-correcting in your rebuttal. This is the reason data is scary, because it is simple and things you wouldn't think are useful. But they are useful when you have thousands or millions of data points.
- mlsu 4y ago
  This is a great point, and you are absolutely correct about extracting signal from noise. But not this kind of data! There might be some useful signal in the points you describe. But it will be extremely difficult to extract, because first you have to sense it ($$$$ -- what is a pet hair to a robot?), then you have to turn the sensor data into something parsable ($$$$ -- how do you determine if this thing the vacuum picked up is a pet hair?), then you have to do the hard data work to make a link between these physical world things and a signal Amazon actually cares about ($$$$ -- OK, they have a Shih Tzu. How can we improve product recommendations with that?). The time of a sensor fusion/robotics person/statistics person at Amazon will not be spent on any of this, not for many years. The data that Amazon does not need these robots to collect, such as zip code, other devices on the LAN, purchase history and login time, and so on -- there is enough low-hanging fruit there to occupy every newly minted PhD for years. The opportunity cost of a statistician working on this robot stuff for an hour outweighs the benefit by an order of magnitude or more. The opportunity cost of a robotics expert, this is even more true. Robots run into stuff and break in the warehouse every single day; for a company like Amazon, that is millions of dollars. I do concede: they will definitely collect and store it. But it will be a very long time before any of this data is used in the name of evil. On the other hand, if they make the robot good at cleaning your house, they might sell a $400 robot to a hundred million people. I totally agree with your point, though, and will offer an Amazon-flavored extension of it: Look at how Amazon gets YOU to freely give them data to digitize your life. They want you to sign up for digital statements on the ChaseBank Silver Plus Amazon Card. Why? They lobby against data-protection legislation. Why? The ebook app wants location data from your phone. Why? They want you to purchase PS4 games on Amazon, instead of on the publisher's website. Why? The robots thing is easy to understand for us because we are morphologically similar to Amazon robots. The rest of this stuff, though, is the real low hanging fruit. And there is enough of it to occupy every single $200,000/yr data science PhD being graduated -- for now, and for a long time coming.
  - godelski 4y ago
    So I still think your dog example is too complicated. All you need to know is if there's a 4 legged creature walking around. You can probably differentiate a cat from a dog here. You can also probably differentiate a small, medium, or large dog. You don't need to know the exact type for this data to be incredibly useful. Just knowing the size will get you a lot of useful data. It is really easy to over complicate the situation. A little data goes a long way.
tootie 4y ago
It's true. Why is Amazon/Google/Facebook considered the evil triumverate of the panopticon and not JP Morgan Chase/Capital One/AmEx? These three know more about the US economy than the BLS or Dept of Commerce.
- godelski 4y ago
  Why are you framing it as an "or" situation? I'm pretty confident most people dislike both these sectors. It is an "and" situation. Turning "ands" into "ors" is really whataboutism.

by mlsu 4y ago

godelski 4y ago

I kinda agree with you, but also not. I think a lot of people here are over assuming what the robots can sense and yes, this is that Sci-Fi like reality. But this doesn't mean we should throw the baby out with the bath water. Anyone that has seriously worked with data will tell you how subtle things can reveal a lot of information. Let's look at these robots. Knowing that there are pets and/or children in a house is useful data to sell kids/pet toys and other items. You might think "Amazon already has that info from purchasing history" but a second source builds higher confidence and can differentiate if something was bought for yourself or for a friend. There's also information like how much the roomba fills up. Are you a messy person or very organized? Do you take off your shoes at the door? When you sit down at the couch? At all? Do you make frequent spills or messes in certain rooms? This obviously has selling potential. You don't need microphones or super sensors. Basic information can do a lot. There's also a lot to be said about redundancy and verification. I think you're over-correcting in your rebuttal. This is the reason data is scary, because it is simple and things you wouldn't think are useful. But they are useful when you have thousands or millions of data points.

mlsu 4y ago
This is a great point, and you are absolutely correct about extracting signal from noise. But not this kind of data! There might be some useful signal in the points you describe. But it will be extremely difficult to extract, because first you have to sense it ($$$$ -- what is a pet hair to a robot?), then you have to turn the sensor data into something parsable ($$$$ -- how do you determine if this thing the vacuum picked up is a pet hair?), then you have to do the hard data work to make a link between these physical world things and a signal Amazon actually cares about ($$$$ -- OK, they have a Shih Tzu. How can we improve product recommendations with that?). The time of a sensor fusion/robotics person/statistics person at Amazon will not be spent on any of this, not for many years. The data that Amazon does not need these robots to collect, such as zip code, other devices on the LAN, purchase history and login time, and so on -- there is enough low-hanging fruit there to occupy every newly minted PhD for years. The opportunity cost of a statistician working on this robot stuff for an hour outweighs the benefit by an order of magnitude or more. The opportunity cost of a robotics expert, this is even more true. Robots run into stuff and break in the warehouse every single day; for a company like Amazon, that is millions of dollars. I do concede: they will definitely collect and store it. But it will be a very long time before any of this data is used in the name of evil. On the other hand, if they make the robot good at cleaning your house, they might sell a $400 robot to a hundred million people. I totally agree with your point, though, and will offer an Amazon-flavored extension of it: Look at how Amazon gets YOU to freely give them data to digitize your life. They want you to sign up for digital statements on the ChaseBank Silver Plus Amazon Card. Why? They lobby against data-protection legislation. Why? The ebook app wants location data from your phone. Why? They want you to purchase PS4 games on Amazon, instead of on the publisher's website. Why? The robots thing is easy to understand for us because we are morphologically similar to Amazon robots. The rest of this stuff, though, is the real low hanging fruit. And there is enough of it to occupy every single $200,000/yr data science PhD being graduated -- for now, and for a long time coming.
- godelski 4y ago
  So I still think your dog example is too complicated. All you need to know is if there's a 4 legged creature walking around. You can probably differentiate a cat from a dog here. You can also probably differentiate a small, medium, or large dog. You don't need to know the exact type for this data to be incredibly useful. Just knowing the size will get you a lot of useful data. It is really easy to over complicate the situation. A little data goes a long way.

tootie 4y ago

It's true. Why is Amazon/Google/Facebook considered the evil triumverate of the panopticon and not JP Morgan Chase/Capital One/AmEx? These three know more about the US economy than the BLS or Dept of Commerce.

godelski 4y ago
Why are you framing it as an "or" situation? I'm pretty confident most people dislike both these sectors. It is an "and" situation. Turning "ands" into "ors" is really whataboutism.