7 ms·

This kind of thinking completely discards the concept of intent. People read things for all kinds of reasons. I have a family member who was an avid reader of t

This kind of thinking completely discards the concept of intent. People read things for all kinds of reasons. I have a family member who was an avid reader of the "Weekly World News", yet he did not believe in aliens, "Bat Boy", or the illuminati. I used to regularly read conservative blogs and listen to Fox News to keep up with what people I disagree with are doing/saying. So the problem is that you can make some statistical predictions about what someone's views are based on their reading habits. But you can't morally judge them on that basis because you don't (and can't) know why they read something and what they thought about it. This is similar to how BMI is commonly misused. BMI is a fine indicator of population health, but is not always indicative of individual health.

thereisnospork 5y ago
You need to define a confidence interval for 'know'. We as a society judge and condemn people without 100% certainty as a matter of course. Both at a personal level and systematically, e.g. the judicial system. To your example of BMI, it would be a perfectly reasonable public health policy[0] to use machine learning to probabilistically identify persons of an excess BMI to send them pamplets and resources for weight loss / exercise. Will the odd fit person get a letter from the surgeon general telling them "being a fatass is bad for their health"[1]? Of course. I don't see why that is a terminal problem. [0]If a bit creepy for the privacy aspects - which are out of scope here. [1]In far more words than is necessary, of course
- throwawayboise 5y ago
  I would not be a perfectly reasonable public health policy to do what you suggest, it would be a waste of time and money. Everybody knows that being fat is unhealthy. The people who are fat and not doing anything about it just don't care, or have other priorities.
  - varelse 5y ago
    Or they have a story far more complicated than what have you have reduced them to being. Just attempting to manage the pandemic in the US has demonstrated how complicated and contradictory people turn out to be. But I do agree that pamphlets about BMI aren't likely to change their outlook. They each need their own personalized moments of clarity about their path and it's not all a sure thing they ever have theirs. But also, a few people are obese because they are on steroids or they have some genetic issue independent of their choices in life. They are not the norm, but they'll get lumped in with the norm and that's offensive and improper. That said, if FB found out their userbase is unusually obese compared to the rest of a country's citizens, that informs them they have a potential moral hazard in their hands. But if you're deeply offended at such telemetry, maybe consider no using Facebook. You'll be fine without it. I personally make sure to post ridiculous and contradictory responses to the insipidly and horribly targeted ads they push into my feed. Some of them have gotten me suspended for violating their "community standards" that IMO would not get me suspended anywhere else.
varelse 5y ago
You're using outliers to invalidate far more accurate predictions of the ensemble. You're right about people like yourself. IMO they are not the norm. Most "Weekly World News" readers are, I suspect, far more gullible than you. And far more Fox News viewers agree with their talking points and agenda than disagree or they wouldn't be watching, also IMO.
autoexec 5y ago
> So the problem is that you can make some statistical predictions about what someone's views are based on their reading habits. But you can't morally judge them on that basis Companies are making these kinds of flawed assumptions about you and every one of us every single day. They often then sell that info to data brokers who happily sell that data to others who will then start with incorrect assumptions about you as an individual and let that influence how they interpret the rest of the data they collect about you. It's a real problem because those flawed data sets you aren't allowed to see, contest, or update are increasingly being used to meaningfully impact your everyday life in ways that you'll never be aware of. The bottom line is that companies don't care. If they will make more money by being right most of the time that's what they are going to do. You might be the fittest, healthiest person on Earth, but if your health insurance company sees that people in your area code have started buying fast food more often they can decide to raise your rates. They won't tell you why they did it. They'll just do it. You could be the most financially responsible person on Earth but if you live in the wrong zip code don't be surprised when you get denied certain services or told that a company's polices are one thing when they would have told you they were something else if you lived on the other side of town. That said, it's probably a whole lot easier to make accurate predictions about people than you think. Sure you'd read fox news, but most of your time is probably not spent on right wing sites, you probably aren't leaving comments espousing right wing talking points, and you probably aren't donating money to right wing causes. With enough data it's not that hard to figure out if you're regularly hanging out at stormfront because you're working for the Anti-Defamation League or because you're a racist. Algorithms can detect (and exploit) mental illnesses like bipolar disorder and Alzheimer's companies can certainly detect "stupid" well enough for their own needs.
Miraste 5y ago
This is a common mistake people make when thinking about marketing data analysis. It's similar to bits of entropy in browser tracking - the resolution of your screen and the fonts you have installed don't identify you by themselves, but when they're correlated their effectiveness scales much better than human intuition estimates. So you read conservative blogs and listen to Fox News. If Facebook is predicting, for instance, your voting habits, their algorithm won't look only at that. It'll look at where you live, what your job is, the political stances of the people you most associate with, which groups you're a part of and how their members typically lean, what stores you shop at, what search terms you use, and far more metrics than I can list here. Can you honestly say that all of those will point to incorrect conclusions? If you somehow live a live entirely contrary to your internal beliefs you're in a minority so small as to be irrelevant.
kelnos 5y ago
Right, but FB can figure out your intent much of the time as well. Simply clicking on a conservative news story might not indicate you're conservative, but if those are the only kind of stories you click on, that's a signal. And if 5 minutes after clicking that story, you reshare it, that's another signal. And if sentiment analysis done on whatever you wrote along side the reshare suggests a positive/agreeable reaction, then that's yet another signal. And the aggregation of these signals, especially if you do this often enough, can often correctly identify your political views.

thereisnospork 5y ago

You need to define a confidence interval for 'know'. We as a society judge and condemn people without 100% certainty as a matter of course. Both at a personal level and systematically, e.g. the judicial system. To your example of BMI, it would be a perfectly reasonable public health policy[0] to use machine learning to probabilistically identify persons of an excess BMI to send them pamplets and resources for weight loss / exercise. Will the odd fit person get a letter from the surgeon general telling them "being a fatass is bad for their health"[1]? Of course. I don't see why that is a terminal problem. [0]If a bit creepy for the privacy aspects - which are out of scope here. [1]In far more words than is necessary, of course

throwawayboise 5y ago
I would not be a perfectly reasonable public health policy to do what you suggest, it would be a waste of time and money. Everybody knows that being fat is unhealthy. The people who are fat and not doing anything about it just don't care, or have other priorities.
- varelse 5y ago
  Or they have a story far more complicated than what have you have reduced them to being. Just attempting to manage the pandemic in the US has demonstrated how complicated and contradictory people turn out to be. But I do agree that pamphlets about BMI aren't likely to change their outlook. They each need their own personalized moments of clarity about their path and it's not all a sure thing they ever have theirs. But also, a few people are obese because they are on steroids or they have some genetic issue independent of their choices in life. They are not the norm, but they'll get lumped in with the norm and that's offensive and improper. That said, if FB found out their userbase is unusually obese compared to the rest of a country's citizens, that informs them they have a potential moral hazard in their hands. But if you're deeply offended at such telemetry, maybe consider no using Facebook. You'll be fine without it. I personally make sure to post ridiculous and contradictory responses to the insipidly and horribly targeted ads they push into my feed. Some of them have gotten me suspended for violating their "community standards" that IMO would not get me suspended anywhere else.

varelse 5y ago

You're using outliers to invalidate far more accurate predictions of the ensemble. You're right about people like yourself. IMO they are not the norm. Most "Weekly World News" readers are, I suspect, far more gullible than you. And far more Fox News viewers agree with their talking points and agenda than disagree or they wouldn't be watching, also IMO.

autoexec 5y ago

> So the problem is that you can make some statistical predictions about what someone's views are based on their reading habits. But you can't morally judge them on that basis Companies are making these kinds of flawed assumptions about you and every one of us every single day. They often then sell that info to data brokers who happily sell that data to others who will then start with incorrect assumptions about you as an individual and let that influence how they interpret the rest of the data they collect about you. It's a real problem because those flawed data sets you aren't allowed to see, contest, or update are increasingly being used to meaningfully impact your everyday life in ways that you'll never be aware of. The bottom line is that companies don't care. If they will make more money by being right most of the time that's what they are going to do. You might be the fittest, healthiest person on Earth, but if your health insurance company sees that people in your area code have started buying fast food more often they can decide to raise your rates. They won't tell you why they did it. They'll just do it. You could be the most financially responsible person on Earth but if you live in the wrong zip code don't be surprised when you get denied certain services or told that a company's polices are one thing when they would have told you they were something else if you lived on the other side of town. That said, it's probably a whole lot easier to make accurate predictions about people than you think. Sure you'd read fox news, but most of your time is probably not spent on right wing sites, you probably aren't leaving comments espousing right wing talking points, and you probably aren't donating money to right wing causes. With enough data it's not that hard to figure out if you're regularly hanging out at stormfront because you're working for the Anti-Defamation League or because you're a racist. Algorithms can detect (and exploit) mental illnesses like bipolar disorder and Alzheimer's companies can certainly detect "stupid" well enough for their own needs.

Miraste 5y ago

This is a common mistake people make when thinking about marketing data analysis. It's similar to bits of entropy in browser tracking - the resolution of your screen and the fonts you have installed don't identify you by themselves, but when they're correlated their effectiveness scales much better than human intuition estimates. So you read conservative blogs and listen to Fox News. If Facebook is predicting, for instance, your voting habits, their algorithm won't look only at that. It'll look at where you live, what your job is, the political stances of the people you most associate with, which groups you're a part of and how their members typically lean, what stores you shop at, what search terms you use, and far more metrics than I can list here. Can you honestly say that all of those will point to incorrect conclusions? If you somehow live a live entirely contrary to your internal beliefs you're in a minority so small as to be irrelevant.

kelnos 5y ago

Right, but FB can figure out your intent much of the time as well. Simply clicking on a conservative news story might not indicate you're conservative, but if those are the only kind of stories you click on, that's a signal. And if 5 minutes after clicking that story, you reshare it, that's another signal. And if sentiment analysis done on whatever you wrote along side the reshare suggests a positive/agreeable reaction, then that's yet another signal. And the aggregation of these signals, especially if you do this often enough, can often correctly identify your political views.