Hacker News Reader

Comment by pasbesoin | Hacker News Reader

In my opinion, in the U.S. such practice represents a clear commercial intrusion upon the First Amendment right of free association. And I hope that any and all companies that practice it are sued into oblivion. I have zero tolerance for such behaviour.

Our society does better when we communicate with each other. As in, "united we stand, divided we fall".

If we are not to turn entirely into a perpetual exercise in "pecking order", there have to be some limits upon what can be "punished" -- commercially or otherwise.

So, take your "social network credit risk" and shove it.

View on HN

The Constitution and its amendments are designed to protect citizens from the government, not from private corporations. There may be other laws preventing this sort of credit reporting, but the Constitution does not.

I don't think that's totally correct.

"...in Pembina Consolidated Silver Mining Co. v. Pennsylvania - 125 U.S. 181 (1888), the Court clearly affirmed the doctrine, holding, "Under the designation of 'person' there is no doubt that a private corporation is included [in the Fourteenth Amendment]"

http://en.wikipedia.org/wiki/Corporate_personhood

huh? All that says is that corporations enjoy the same protections from government that people do. It doesn't say anything about protecting people from corporations.

What is wrong with using data to make more accurate predictions? Why does it matter where the data came from? That's their job, to make accurate predictions about how likely people are to pay back a loan. This benefits everyone. Interest rates will be lower and more accurately reflect the risk.

If I had a time machine and could go to the future and see who would pay back their loans and who wouldn't, that would be ideal. But we don't have that, we can only make predictions by looking at data that happened to correlate with that in the past.

Because you hose people who are not actually high-risk but, for whatever the reason, chose to hang out with people who are high-risk. This is not really a socially acceptable thing to do, because you are effectively punishing those individuals for their choice in friends. While the First Amendment may not apply here, I'd say most people value the spirit of "freedom of association" and thus I expect it to have been applied elsewhere.

I wouldn't equate not giving someone a loan because they are at high risk as "punishment". But anyways it's just a data point. Why does it matter what the data point is, if it helps make more accurate predictions?

No it's not socially acceptable to use certain data points to make decisions, but this has good justification behind it. People will irrationally weigh stuff like race or gender to make decisions even when it doesn't matter. But this is being done by a completely unbiased computer program. And they wouldn't use it if it made their predictions worse. So that reasoning of the predictor being irrational doesn't apply in this case.

this is being done by a completely unbiased computer program

And I'm sure the program was written by a completely unbiased computer, too!

Probably you have never seen Bayesian modeling in action. It doesn't care what data you throw at it, it will establish models out of it. The theory IS neutral.

I highly doubt they would intentionally make their models more inaccurate to reflect their own biases. Machine learning algorithms are pretty unbiased as far as weighing the correlation of specific features goes.

I wonder how Obama would have been classified in his "Community Organizer" days.

    if job == 'Lawyer':
        score + 100
    if poor_people_friends > 150:  
        score - 100

That's not how machine learning. I mean no one just programs "if poor_people_friends" or keeps a running score. They just feed the machine the data.

And the machine just guesses a score?

Are you saying the machine doesn't actually rank you according to the data you give it?

So are you saying that if being black is correlated with a lower rate of loan payback, that is a fair indicator for raising their interest rate? It's not clear to me where you think the line should be drawn, if at all, given a strong correlation.

Yes. It's a data point. It's no different than taking into account their job, past history, age, gender, etc. We aren't talking about humans being irrational like "I hate black people, so I am going to charge them more or refuse to do business with them", but rather an unbiased computer algorithm that is simply improving it's ability to make predictions. What data it uses to do that doesn't matter.

That clearly makes sense economically, and I know this already exists (car insurance, for example), but I don't think many people would agree with this on a social level. A security camera with great AI might only follow the black people in a crowded store because it makes sense statistically. But putting people at an unfair disadvantage for what they can't control is a slippery slope.

No I don't think people are comfortable with it, but it's silly. It's not putting people at an unfair disadvantage any more than predicting using stuff like age or past history. If you are going to be do predictions at all it's pointless to exclude data that could make them more accurate.

> This benefits everyone.

Everyone?

LOL

True, there is a downside for people who are at higher risk of not paying back the loan. But that's the entire purpose. If everyone has to pay the same interest rate, less people can afford loans and those that do have to pay to make up for all the others that don't pay their loans back. The economy can suffer indirectly as well, so it does affect everyone.

People aren't "at risk" of defaulting. Either they have, or they haven't. Risk is entirely a calculation of the lender, and any miscalculation thereof leads to a vicious cycle of repeatedly shafting the unlucky.

>People aren't "at risk" of defaulting. Either they have, or they haven't.

Past history correlates with future performance, but not necessarily 100%. People who have always paid back loans sometimes default. People who have really bad credit sometimes start making payments again. And in any case, it's entirely useless if the person has no past history to look at, which is what this is for.

>Risk is entirely a calculation of the lender, and any miscalculation thereof leads to a vicious cycle of repeatedly shafting the unlucky.

Good thing we are using 21rst century computers and machine learning algorithms on vast amounts of data. At no time in history were lenders able to make predictions so accurately. If there are miscalculations, they are less than would be otherwise.

Because you introduce pressure to separate members and different "elements" of a community. And to further segregate e.g. geographical communities. (Not that there isn't already plenty of pressure for the latter.)

Perhaps I'm naive, but I hold to the... philosophy, I guess I'll call it for the moment, that most people want a strong community. And that that is aided by people being in touch with each other, including and especially people of different social and economic strata.

Take this proposed ranking to its extreme: You will have people who are ostracized and who perhaps cannot become "un-ostracized" because connecting with them forms not only perhaps a personal risk but also a commercial risk for other parties.

Some societies have versions of this, for example, the caste system.

How "mobile" are people going to be, when every "up-link" to someone of higher standing is cause for a computer somewhere to drag down the status of the up-linked.

We also have problems of this sort in the U.S. with our insurance industry -- I'm thinking at the moment particularly of health insurance. Insurance is supposed to be a pooling of risk and of resources with which to deal with that risk. If you don't take out as much as you put in, count your blessings! You have above average health.

If you need help, it is there for you. "I am my brother's keeper."

In times past, we knew less about diseases. You knew less about what might "get you" and where it was coming from.

These days, we know (somewhat -- less than some arrogant doctors would like you to think) more about diseases. We think we can control some factors -- diet and exercise being prime examples.

We still have to contend with injuries, including injuries that we do not cause, such as some car accidents.

And... your control of "controllable/willpower" factors may be less than some people would think. Live in a "food desert", and fresh produce may be difficult to acquire. Live in certain places in my metropolitan area, and an evening walk for exercise presents its own not inconsiderable risks.

Nonetheless, many people these days seem included to think of insurance as akin to a savings plane. You should get out what you pay in. If not, you're being ripped off!

That's not how it works. That's not how it's intended to work.

Back to social media and risk evaluation. Do I, should I have to evaluate every potential connection for the economic risk it may bring me?

We are close to having ubiquitous surveillance cameras (including ultra-high resolution airframe mounted cameras that can track a vehicle or even a pedestrian across an entire city). Do I have to worry that every chance encounter and conversation on the street may effect not just my political and security rating but also my economic rating? Talk to that bum asking for change, and your credit score goes down 50 points. Ok, you can talk to him/her for a maximum of 20 seconds; any longer, and we will think you have an "un-natural" "affinity" for such a lifestyle.

Bankers can't "red line" in making mortgage loans. Schools (including private ones) can't deny admission upon the basis of race -- nor, in some states, upon sexual preference.

We constantly place limitations upon the scope and focus of commercial evaluations. Social media is a landscape that will have to be dealt with, in this regard.

>Do I, should I have to evaluate every potential connection for the economic risk it may bring me?

Yes you are right there is an incentive problem created by this. Once you start measuring something, people start trying to game the measurements. That said, no one is forcing you to make such decisions, or even go to a business that uses such evaluations.

>Bankers can't "red line" in making mortgage loans. Schools (including private ones) can't deny admission upon the basis of race -- nor, in some states, upon sexual preference.

>We constantly place limitations upon the scope and focus of commercial evaluations. Social media is a landscape that will have to be dealt with, in this regard.

Yes but the intention of those laws are to stop irrational bigotry. Like someone who hates some group and refusing to do business with them because they want to punish them or whatever. We are talking about an unbiased computer algorithm that just weighs the data and uses it to make more accurate predictions. That's the goal, to make accurate predictions. The data used doesn't matter.

What happens if Bill Gates friends me?

Reply on news.ycombinator.com