How do the words "NASA's regulations" lead you to think that "NASA requirements" are irrelevant?
Possibly because the words "NASA's regulations" don't appear anywhere in the article cited?
The article states that propulsive landing was deleted from human transport missions because "it would have taken a tremendous amount of effort to qualify that for safety for crew transport." But it was deleted from robotic Mars missions because "'I'm pretty confident that is not the right way" and SpaceX has "a far better approach". (Those are Musk's words, not mine.)
You're mixing up two different things. This article was about Red Dragon, which was a proposed unmanned Mars mission. Commercial crew is a different thing-- doesn't go to Mars, doesn't land on Mars, does carry humans.
SpaceX isn't really such a good counter to that. They benefitted from around a century of government research into rocketry, aerospace and space flight, as well as lots of government subsidies. Their biggest customer is one of the biggest governments in the world. And although they're doing it in very innovative ways, they're serving a pretty well-established market.
And, most particularly, they leveraged NASA funding to build the Falcon-9.
To his credit, Musk doesn't ever try to hide that-- he clearly and directly acknowledges NASA's help. In interviews, he points out that after Space-X failed on their first three launches, NASA was the only one willing to invest in them, and they would have gone bankrupt without it.
In fact, SpaceX may have found the right middle ground -- working with NASA changed them from a company with a record of a string of failures to a company with a record of a sting of successes, but they are separated from NASA enough that they can try cool stuff without too long a string of regulations and reviews. Good for them.
They're still working with NASA. Let's hope they can keep that middle ground, distant enough to be innovative, close enough to be rigorous.
Space exploration was one of the favourite things for liberals to point fingers at and scream "let's see free market tackle THAT". Now it is, they're in panic.
Yeah, I'll challenge that statement. I don't think I've ever heard a liberal say that.
It just hasn't been a big issue on the liberal agenda, frankly.
NASA doesn't have a design that has been successfully used to land a human on Mars. NASA doesn't have a successful design for reusable rocket either.
Right on the first, wrong on the second.
The space shuttle was a reusable launch vehicle that flew in 1981-- before half of you slashdotters were even born. More reusable than Falcon-9, in fact, since the Falcon 9 throws away the second stage (which tends to be the more expensive part).
(The problem with the space shuttle is that the technology got frozen in 1981. It should have been retired in favor of some better next-generation launcher by the 1990s. Instead, the demonstrated problems got patches, but the design itself never evolved, never changed. )
Very un-American of you to tell someone how to spend their money. May be you should grow up and contribution something to society.
I don't know if what's "American," but I love the idea of private individuals trying stuff with their own funding and succeeding or failing on their own merits.
It's like you didn't even read the article or pay attention to what he said. So I guess someone has to repeat it for you.
To the contrary, it's like you read the article but weren't really familar with the mission.
NASA's regulations for propulsive landing of a Dragon 2 capsule are too difficult to reasonably meet.
Red Dragon was a proposed private mission to Mars. It is not a NASA mission, and NASA requirements are irrelevant.
I like Musk. I like the approach of trying stuff, and if it doesn't work, try something else. They worked on this idea and, when they got down into the details, decided the propulsive landing technique wouldn't work, so they gave up on it. Good for them.
But don't blame NASA. It wasn't a NASA mission in the first place.
What Makes a Contract Invalid? :
...A contract can be void for the following reasons:...The contract restricts the rights of a party
You can absolutely contract rights away, just not those that would cause the court to deem the contract substantively unconscionable.
in this particular instance, the contract restricts the rights of an entity that was not part of the contract (the company that would otherwise hire you).
When a contract is void, it is not valid. It can never be enforced under state or federal laws. A void contract is null from the moment it was created and neither party is bound by the terms. Think of it as one that a court would never recognize or enforce because there are missing elements.
A contract can be void for the following reasons:
The terms of the agreement are illegal or against public policy (unlawful consideration or object)
A party was not of sound mind while signing the agreement
A party was under the age of consent
The terms are impossible
The contract restricts the rights of a party
Because the alternative source is not the primary source. It is a mirror of the primary source.
(Also, often the mirrors are on less-robust servers, so it's nice to have the primary source available for when the mirror inexplicably gets slashdotted.)
If blacks and whites are equally likely to default on a loan...
But... They AREN'T.
So, apparently we are talking about different things.
The article I was discussing was one that analyzed data showing an AI algorithm added bias that wasn't there-- it was making predictions that were selectively wrong: overpredicting crime for blacks, underpredicting crime for whites.
You are talking about something that I never brought up, "but what about some hypothetical AI that actually put out accurate predictions? If those predictions happened to match the bias that society has, would that be racism?"
You weren't paying attention when I explained that if you break out for ANY variable, there will be variance. Of course there will be a difference between those measurements. There'd likewise be a difference if you broke it out between left-handedness and right-handedness. It doesn't mean a damn thing, but statistically there would be a measurable difference. So YES, the algorithm WILL MOST CERTAINLY come back with a different rate between rightys and leftys on their ability to repay loans. As we would expect it to. Or race, or whatever.
And I was talking about the article in question, in which an algorithm predicted a difference between blacks and whites that was not there in the real world results.
And while you're saying we want a more accurate predictions, which I wholly agree with, it sounds like you're also saying you also want the prediction to be equal when broken out by race and applied to society.
No. I'm saying although it is likely that any algorithm will make errors, the algorithm can be called biased if it just "happens" to selectively make errors that are harmful to blacks and helpful to whites.
Yeah, data like how likely they are to default on a loan. If you break out that statistic by race, you're going to see different numbers. That's just... how statistics work. . break out anything and you'll see variation. But that's the exact data you're looking for when making a loan. That's the goal. But you're saying that if there's any racial variation in that statistic it's a "invisibly encoded race" field.
No, not exactly. I'm saying that if blacks and whites are equally likely to default on loans, but the algorithm comes out with a result that blacks are more likely to default on loans, then there must be something internal to the algorithm which discriminates based on race.
If that factor is not explicit, it must be invisible.
And therefore approving or denying loans based on how likely they are to default on a loan is racism and illegal.
No, you weren't paying attention. If blacks and whites are equally likely to default on a loan, then using an algorithm which (inaccurately) predicts blacks are more likely to default is bias ("bias" was the term I used; you're the one who altered to "racism and illegal.")
This is what you're saying is desirable, and I'm saying it's not.
In the case analyzed in the article, the algorithm predicts that blacks are much more likely to commit future crimes than they actually are, and whites are less likely. The results are inaccurate, and they are inaccurate in a way that reflects the bias of society.
I am saying that is is desirable that the results not be inaccurate in a way that is biased against blacks and in favor of white.
suppose that whites living in black-majority neighborhoods are, for some reason, likely to not repay loans (possibly because if they weren't financially distressed they'd move to the lily-white suburbs); but blacks living in black-majority neighborhoods have no problem (because there's nothing exceptional about blacks living in black-majority neighborhoods, that's just how "majority" is defined.) So, an algorithm tags "living in black majority neighborhood" as correlating with defaulting on loans. The net result is that blacks are denied loans even though they do not have a higher probability of default. The results of the loan algorithm are not race neutral-
...hmmmm Just like the other guy, you've given a pretty good argument that the algorithm SHOULD be told the race of the target. Because the way you make the program stop unfairly dinging the black neighborhood is to tell it the races of the people therein, so it'd see that the white trash is bringing down the hood. And spotting those sort of trends would make the AI hella racist. But of course, in your example, the biggest factor IS race.
That was an example case-- a Gedankenexperiment-- of how race could be encoded into ostensibly non-racial data, showing why the problem is non-trivial. You are the one concluding that this means race should be considered by the algorithm. That is indeed one possible solution. It is not clear that it is the only possible solution, or the most desirable solution.
But the results are all that matters
Correct, anything that more accurately predict loan default (or recidivism per the article) helps make a better tool and save money and lets good people out on parole and keeps bad people in prison.
"Accurately predict" is the key phrase here. The whole point of the article in question is that not only do the results do not accurately predict recidivism rates, but that the inaccuracy is biased in favor of whites and against blacks.
I have no doubt that the algorithm accurately represents the data it was trained with.
Upon what data are you making that confident statement "the algorithm accurately represents the data it was trained with"? How in the world do you know that? Did you examine the algorithm? Or the data it was trained with? Or are you just saying "trust the computer, the computer is always right" (or, possibly, "Programmers never make mistakes.")
In any case, though, the purpose of the algorithm is not to "accurately represent the data it was trained with." The purpose of the algorithm is to make accurate predictive decisions which are used in the real world and affect peoples lives. If these predictions are inaccurate, the phrase "accurately represent the data it was trained with" is simply a euphemism for "wrong".
It sounds like they need to add additional factors to the risk scoring, so that it can have greater forward-predictive value, not just backtesting. My comment is addressed more to the concept that "it's biased, therefore it must be bad". If the data is biased, then the algorithm should be, too.
If the data is biased, the algorithm should correct out that bias. That is what data analysis does.
If the algorithm gives incorrect results because it cannot correct out the bias implicit, the proper answer is to stop using that algorithm. Not to say "oh, well the data is biased so we can expect results to be biased; live with it."
You can't have AI that learns on its own and have AI that isn't racially biased unless you artificially code blocks to it reaching certain logical conclusions.
What? No? You don't block their inputs, not their outcome. You just hide the race of the targets it's looking at. As in, the sql library which has all that data it's digesting? You just exclude the 'race' field. DONE. The algorithm isn't judging based on race.
Nope, excluding the "race" fields does not mean the algorithm isn't judging on race. If the other fields have race invisibly encoded into them, it can still be judging based on race... but now it's doing it in a way that you can't see any more.
That won't stop it from looking at... say... the location where someone lives for approving or disapproving a loan. And lo-and-behold that's pretty similar to looking at someone's race.
Exactly. Race can be coded into other data.
But that's not their race and an unbiased look at something that DOES indicate loan-worthiness. If that's the sort of thing the ACLU disapproves of, they've got a very difficult fight on their hands, because where does that end?
It is a difficult problem. But just because it is difficult to exclude invisible bias due to race does not mean that it is not desirable to do so.
In the example you give, suppose that whites living in black-majority neighborhoods are, for some reason, likely to not repay loans (possibly because if they weren't financially distressed they'd move to the lily-white suburbs); but blacks living in black-majority neighborhoods have no problem (because there's nothing exceptional about blacks living in black-majority neighborhoods, that's just how "majority" is defined.) So, an algorithm tags "living in black majority neighborhood" as correlating with defaulting on loans. The net result is that blacks are denied loans even though they do not have a higher probability of default. The results of the loan algorithm are not race neutral-- even though the data appears to be both objective and not explicitly including race. But the results are all that matters. How do you make loans race neutral in this situation?
It's hard. But, again: just because a problem is hard, doesn't mean that it should be ignored.
Indeed, I would consider racial bias to be a subset of "faulty programming."
Far from it. A system that lacked the racial bias reflected in reality would by it's very nature be flawed, and racially discriminatory.
Stop right there. We're talking about different things.
You are talking about "racial bias reflected in reality", but the article I am referring to is talking about racial bias that is in the output of the AI but is not reflected in reality. The article talks about the comparison of the AI output with actual results that show that the AI overpredicts blacks will commit crimes, and underpredicts that whites will commit crimes. The AI is not "reflecting reality".
The whole point is that the AI is inserting racial bias that does not exist in the actual world. (Or, more to the point, that the data being input to the AI already has racial bias in it, and the AI output is reflecting that bias.)
Perhaps you should read the links you post.
Spring thaw of 'snow' and 'ice' in mountains or far south of the arctic circle are completely irrelevant for what is going north of the arctic circle.
I'm not sure what you're referring to. The first link I posted, for example was not about "snow and ice in mountains," it was about "boreal forests and tundra". ("Boreal" means "northern", but in context it's almost always used to mean the far north, Arctic and subarctic. By the way, your post says "north of the arctic circle," but I was talking about Russia. Much of Russia is subarctic, but very little is "north of the Arctic circle".)
Here is a quote from that first link, saying specifically what it was covering:
"Research scientists have been studying freeze/thaw dynamics in North America and Eurasia's boreal forests and tundra to decipher effects on the timing and length of the growing season. These regions encompass almost 30 percent of global land area.... Large expanses of boreal forest and tundra are underlain by permafrost, a layer of permanently frozen soil found underneath the active, seasonally thawed soil.
Are you really such an idiot?
None of you links has on the first glance ANYTHING to do with what we where talking about before. Who cars that land that already is farmland is thawing a week or two more early? You claimed that climate change would create new farmland in Russia, which it won't.
This seems unlikely. I figure it's far more likely that the AI is simply solving the wrong problem.
No, the problem is that the input data it used had invisible bias. There is an old saying in the computer industry "Garbage in, garbage out.". If the input is biased, the output will be biased.
If the AI's job is to assess the odds of recidivism, taking into account all available data, then it's neither going to go out of its way to be racist, nor go out of its way not to be racist.
What the heck is wrong with computer engineers? You guys think "oh, the problem can't be bad programming, the computer is never wrong. It has to be the user. Somehow."
No, of course it didn't "go out of its way" to be racist. It just happened that the results were racist. One easy explanation for this is that the racism was inherent in the input data.
If it's showing a bias against black convicts, presumably that's because black convicts really do have worse recidivism rates for whatever reason.
So, what you're saying here is that you didn't read the article. (here)
The point of the article was that data showed that the predictions of the AI did not match the data, and the way in which they did not match the data was that they overpredicted recidivism for blacks, and underpredicted recidivism for whites.
(Of course, that's 'recidivism rate' according to the data. It doesn't disprove, say, the existence of a racist police force with a racially-biased arrest pattern.)
True. That's another, different reason that an AI might give output results that are racist.
I'd be willing to bet that if you did a backtesting study, pitting the AI against human judges, the AI would beat them.
The data showed that the AI was slightly better than a coin flip.
Slightly.
It might well also be far more racist, as the judges are likely to want to discount race on moral grounds. They don't just want an AI that predicts recidivism rates, they want one that does so whilst incorporating our senses of morality and fairness. They're obviously not the same thing.
So, basically you're repeating here that you didn't read the article.
Indeed, I would consider racial bias to be a subset of "faulty programming."
But..in the article, it said they were NOT using race as an AI training factor....so, it wasn't racial bias being programmed in.
Racial bias very clearly was programmed in. It turned out to have been programmed in by using weighting factors that were themselves dependent on race.
The problem is not that the data set reflects the reality. The problem is not that the AI makes mistakes, but that the particular mistakes the AI makes reflect the bias of the society that programmed it.
Why do you use the term 'mistakes'? If the AI accurately reflects the culture that feeds the data to it, it is not a mistake.
It wasn't supposed to "reflect the culture that feeds data to it." It was supposed to predict the probability that a given person on parole would commit a future crime. It did that wrong. And it did that wrong in a specific way, that preferentialy assumed that blacks were more likely to commit crimes than the data shows that they actually did, and that whites were less likely.
"Further, the fact that more people of a particular race are prosecuted is not a reflection of bias in the data, rather a bias in the prosecution."
In this case, "persecuted" was more accurate.
Data is Data. It cannot exhibit a bias.
I can only surmise that you're not an experimental scientist. Data has bias all the time.
In physics (my field) the bias usually has no social consequence-- astronomical statistics, for example, are biased toward bright stars (since they're much easier to see than faint ones, and hence overrepresented in the data set). In social "sciences," however, the bias very often does have social consequences. SAT scores from children whose parents spend tens of thousands of dollars on SAT Prep courses, for example-- surprise!-- score better on SAT exams than ones who don't. The data shows a correlation of SAT score with parental income. Is this real? Better correct for the SAT-prep course effect before making a conclusion.
Data is biased. All the time. Be ready for it.
...Plus, being from the Guardian, I am skeptical that they didn't twist the data some to obtain their desired outcome, which ironically touches on the subject of this story.
Slashdot Mirror
How do the words "NASA's regulations" lead you to think that "NASA requirements" are irrelevant?
Possibly because the words "NASA's regulations" don't appear anywhere in the article cited?
The article states that propulsive landing was deleted from human transport missions because "it would have taken a tremendous amount of effort to qualify that for safety for crew transport." But it was deleted from robotic Mars missions because "'I'm pretty confident that is not the right way" and SpaceX has "a far better approach". (Those are Musk's words, not mine.)
You're mixing up two different things. This article was about Red Dragon, which was a proposed unmanned Mars mission. Commercial crew is a different thing-- doesn't go to Mars, doesn't land on Mars, does carry humans.
One thing I know is that Space X has renewed hope for space travel, where NASA could not
SpaceX renewed hope for space travel by working with NASA.
SpaceX isn't really such a good counter to that. They benefitted from around a century of government research into rocketry, aerospace and space flight, as well as lots of government subsidies. Their biggest customer is one of the biggest governments in the world. And although they're doing it in very innovative ways, they're serving a pretty well-established market.
And, most particularly, they leveraged NASA funding to build the Falcon-9.
To his credit, Musk doesn't ever try to hide that-- he clearly and directly acknowledges NASA's help. In interviews, he points out that after Space-X failed on their first three launches, NASA was the only one willing to invest in them, and they would have gone bankrupt without it.
In fact, SpaceX may have found the right middle ground -- working with NASA changed them from a company with a record of a string of failures to a company with a record of a sting of successes, but they are separated from NASA enough that they can try cool stuff without too long a string of regulations and reviews. Good for them.
They're still working with NASA. Let's hope they can keep that middle ground, distant enough to be innovative, close enough to be rigorous.
Space exploration was one of the favourite things for liberals to point fingers at and scream "let's see free market tackle THAT". Now it is, they're in panic.
Yeah, I'll challenge that statement. I don't think I've ever heard a liberal say that.
It just hasn't been a big issue on the liberal agenda, frankly.
> previous NASA designs used successfully.
NASA doesn't have a design that has been successfully used to land a human on Mars. NASA doesn't have a successful design for reusable rocket either.
Right on the first, wrong on the second.
The space shuttle was a reusable launch vehicle that flew in 1981-- before half of you slashdotters were even born. More reusable than Falcon-9, in fact, since the Falcon 9 throws away the second stage (which tends to be the more expensive part).
(The problem with the space shuttle is that the technology got frozen in 1981. It should have been retired in favor of some better next-generation launcher by the 1990s. Instead, the demonstrated problems got patches, but the design itself never evolved, never changed. )
Very un-American of you to tell someone how to spend their money. May be you should grow up and contribution something to society.
I don't know if what's "American," but I love the idea of private individuals trying stuff with their own funding and succeeding or failing on their own merits.
It's like you didn't even read the article or pay attention to what he said. So I guess someone has to repeat it for you.
To the contrary, it's like you read the article but weren't really familar with the mission.
NASA's regulations for propulsive landing of a Dragon 2 capsule are too difficult to reasonably meet.
Red Dragon was a proposed private mission to Mars. It is not a NASA mission, and NASA requirements are irrelevant.
I like Musk. I like the approach of trying stuff, and if it doesn't work, try something else. They worked on this idea and, when they got down into the details, decided the propulsive landing technique wouldn't work, so they gave up on it. Good for them.
But don't blame NASA. It wasn't a NASA mission in the first place.
What Makes a Contract Invalid? : ...The contract restricts the rights of a party
...A contract can be void for the following reasons:
You can absolutely contract rights away, just not those that would cause the court to deem the contract substantively unconscionable.
in this particular instance, the contract restricts the rights of an entity that was not part of the contract (the company that would otherwise hire you).
Why does CA think it can unilateral terminate a voluntary contract between an employee and employer?
CA is not "terminating" a contract. It is declaring that the contract was not valid in the first place.
What Makes a Contract Invalid? :
When a contract is void, it is not valid. It can never be enforced under state or federal laws. A void contract is null from the moment it was created and neither party is bound by the terms. Think of it as one that a court would never recognize or enforce because there are missing elements.
A contract can be void for the following reasons:
The terms of the agreement are illegal or against public policy (unlawful consideration or object)
A party was not of sound mind while signing the agreement
A party was under the age of consent
The terms are impossible
The contract restricts the rights of a party
(Also, often the mirrors are on less-robust servers, so it's nice to have the primary source available for when the mirror inexplicably gets slashdotted.)
Take it a step further, and use steganography.
You mean encryption.
steganography would be hiding bits within the image-- encryption would be hiding the image.
And some other sources reporting the story:
the hill http://thehill.com/policy/nati...
zdnet http://www.zdnet.com/article/u...
Yahoo https://finance.yahoo.com/news...
The link seems to go to an article on net neutrality. Correct link is here: http://www.reuters.com/article...
Soon he was starring as master of disguise Rollin Hand on Mission: Impossible -- which ran from 1966 to 1973 -- and on Space: 1999...
Wow, I had no idea he played master of disguise Rolin Hand on Space:1999.
That really was some good disguise!
If blacks and whites are equally likely to default on a loan...
But... They AREN'T.
So, apparently we are talking about different things.
The article I was discussing was one that analyzed data showing an AI algorithm added bias that wasn't there-- it was making predictions that were selectively wrong: overpredicting crime for blacks, underpredicting crime for whites.
You are talking about something that I never brought up, "but what about some hypothetical AI that actually put out accurate predictions? If those predictions happened to match the bias that society has, would that be racism?"
You weren't paying attention when I explained that if you break out for ANY variable, there will be variance. Of course there will be a difference between those measurements. There'd likewise be a difference if you broke it out between left-handedness and right-handedness. It doesn't mean a damn thing, but statistically there would be a measurable difference. So YES, the algorithm WILL MOST CERTAINLY come back with a different rate between rightys and leftys on their ability to repay loans. As we would expect it to. Or race, or whatever.
And I was talking about the article in question, in which an algorithm predicted a difference between blacks and whites that was not there in the real world results.
And while you're saying we want a more accurate predictions, which I wholly agree with, it sounds like you're also saying you also want the prediction to be equal when broken out by race and applied to society.
No. I'm saying although it is likely that any algorithm will make errors, the algorithm can be called biased if it just "happens" to selectively make errors that are harmful to blacks and helpful to whites.
Race can be coded into other data.
Yeah, data like how likely they are to default on a loan. If you break out that statistic by race, you're going to see different numbers. That's just... how statistics work. . break out anything and you'll see variation. But that's the exact data you're looking for when making a loan. That's the goal. But you're saying that if there's any racial variation in that statistic it's a "invisibly encoded race" field.
No, not exactly. I'm saying that if blacks and whites are equally likely to default on loans, but the algorithm comes out with a result that blacks are more likely to default on loans, then there must be something internal to the algorithm which discriminates based on race.
If that factor is not explicit, it must be invisible.
And therefore approving or denying loans based on how likely they are to default on a loan is racism and illegal.
No, you weren't paying attention. If blacks and whites are equally likely to default on a loan, then using an algorithm which (inaccurately) predicts blacks are more likely to default is bias ("bias" was the term I used; you're the one who altered to "racism and illegal.")
This is what you're saying is desirable, and I'm saying it's not.
In the case analyzed in the article, the algorithm predicts that blacks are much more likely to commit future crimes than they actually are, and whites are less likely. The results are inaccurate, and they are inaccurate in a way that reflects the bias of society.
I am saying that is is desirable that the results not be inaccurate in a way that is biased against blacks and in favor of white.
suppose that whites living in black-majority neighborhoods are, for some reason, likely to not repay loans (possibly because if they weren't financially distressed they'd move to the lily-white suburbs); but blacks living in black-majority neighborhoods have no problem (because there's nothing exceptional about blacks living in black-majority neighborhoods, that's just how "majority" is defined.) So, an algorithm tags "living in black majority neighborhood" as correlating with defaulting on loans. The net result is that blacks are denied loans even though they do not have a higher probability of default. The results of the loan algorithm are not race neutral-
...hmmmm Just like the other guy, you've given a pretty good argument that the algorithm SHOULD be told the race of the target. Because the way you make the program stop unfairly dinging the black neighborhood is to tell it the races of the people therein, so it'd see that the white trash is bringing down the hood. And spotting those sort of trends would make the AI hella racist. But of course, in your example, the biggest factor IS race.
That was an example case-- a Gedankenexperiment-- of how race could be encoded into ostensibly non-racial data, showing why the problem is non-trivial. You are the one concluding that this means race should be considered by the algorithm. That is indeed one possible solution. It is not clear that it is the only possible solution, or the most desirable solution.
But the results are all that matters
Correct, anything that more accurately predict loan default (or recidivism per the article) helps make a better tool and save money and lets good people out on parole and keeps bad people in prison.
"Accurately predict" is the key phrase here. The whole point of the article in question is that not only do the results do not accurately predict recidivism rates, but that the inaccuracy is biased in favor of whites and against blacks.
I have no doubt that the algorithm accurately represents the data it was trained with.
Upon what data are you making that confident statement "the algorithm accurately represents the data it was trained with"? How in the world do you know that? Did you examine the algorithm? Or the data it was trained with? Or are you just saying "trust the computer, the computer is always right" (or, possibly, "Programmers never make mistakes.")
In any case, though, the purpose of the algorithm is not to "accurately represent the data it was trained with." The purpose of the algorithm is to make accurate predictive decisions which are used in the real world and affect peoples lives. If these predictions are inaccurate, the phrase "accurately represent the data it was trained with" is simply a euphemism for "wrong".
It sounds like they need to add additional factors to the risk scoring, so that it can have greater forward-predictive value, not just backtesting. My comment is addressed more to the concept that "it's biased, therefore it must be bad". If the data is biased, then the algorithm should be, too.
If the data is biased, the algorithm should correct out that bias. That is what data analysis does.
If the algorithm gives incorrect results because it cannot correct out the bias implicit, the proper answer is to stop using that algorithm. Not to say "oh, well the data is biased so we can expect results to be biased; live with it."
You can't have AI that learns on its own and have AI that isn't racially biased unless you artificially code blocks to it reaching certain logical conclusions.
What? No? You don't block their inputs, not their outcome. You just hide the race of the targets it's looking at. As in, the sql library which has all that data it's digesting? You just exclude the 'race' field. DONE. The algorithm isn't judging based on race.
Nope, excluding the "race" fields does not mean the algorithm isn't judging on race. If the other fields have race invisibly encoded into them, it can still be judging based on race... but now it's doing it in a way that you can't see any more.
That won't stop it from looking at... say... the location where someone lives for approving or disapproving a loan. And lo-and-behold that's pretty similar to looking at someone's race.
Exactly. Race can be coded into other data.
But that's not their race and an unbiased look at something that DOES indicate loan-worthiness. If that's the sort of thing the ACLU disapproves of, they've got a very difficult fight on their hands, because where does that end?
It is a difficult problem. But just because it is difficult to exclude invisible bias due to race does not mean that it is not desirable to do so.
In the example you give, suppose that whites living in black-majority neighborhoods are, for some reason, likely to not repay loans (possibly because if they weren't financially distressed they'd move to the lily-white suburbs); but blacks living in black-majority neighborhoods have no problem (because there's nothing exceptional about blacks living in black-majority neighborhoods, that's just how "majority" is defined.) So, an algorithm tags "living in black majority neighborhood" as correlating with defaulting on loans. The net result is that blacks are denied loans even though they do not have a higher probability of default. The results of the loan algorithm are not race neutral-- even though the data appears to be both objective and not explicitly including race. But the results are all that matters. How do you make loans race neutral in this situation?
It's hard. But, again: just because a problem is hard, doesn't mean that it should be ignored.
Indeed, I would consider racial bias to be a subset of "faulty programming."
Far from it. A system that lacked the racial bias reflected in reality would by it's very nature be flawed, and racially discriminatory.
Stop right there. We're talking about different things.
You are talking about "racial bias reflected in reality", but the article I am referring to is talking about racial bias that is in the output of the AI but is not reflected in reality. The article talks about the comparison of the AI output with actual results that show that the AI overpredicts blacks will commit crimes, and underpredicts that whites will commit crimes. The AI is not "reflecting reality".
The whole point is that the AI is inserting racial bias that does not exist in the actual world. (Or, more to the point, that the data being input to the AI already has racial bias in it, and the AI output is reflecting that bias.)
The northern hemisphere spring thaw has been advancing by about one day per year since 1988:
- https://www.nasa.gov/home/hqne...
- http://www.foodnutritionscienc...
- http://climatechange.lta.org/c...
- http://flatheadcore.org/featur...
- https://earthobservatory.nasa....
- http://www.sciencedirect.com/s...
- http://onlinelibrary.wiley.com...
Perhaps you should read the links you post. Spring thaw of 'snow' and 'ice' in mountains or far south of the arctic circle are completely irrelevant for what is going north of the arctic circle.
I'm not sure what you're referring to. The first link I posted, for example was not about "snow and ice in mountains," it was about "boreal forests and tundra". ("Boreal" means "northern", but in context it's almost always used to mean the far north, Arctic and subarctic. By the way, your post says "north of the arctic circle," but I was talking about Russia. Much of Russia is subarctic, but very little is "north of the Arctic circle".)
Here is a quote from that first link, saying specifically what it was covering:
"Research scientists have been studying freeze/thaw dynamics in North America and Eurasia's boreal forests and tundra to decipher effects on the timing and length of the growing season. These regions encompass almost 30 percent of global land area. ... Large expanses of boreal forest and tundra are underlain by permafrost, a layer of permanently frozen soil found underneath the active, seasonally thawed soil.
Are you really such an idiot? None of you links has on the first glance ANYTHING to do with what we where talking about before. Who cars that land that already is farmland is thawing a week or two more early? You claimed that climate change would create new farmland in Russia, which it won't.
Are we talking about the same thing?
mistakes
This seems unlikely. I figure it's far more likely that the AI is simply solving the wrong problem.
No, the problem is that the input data it used had invisible bias. There is an old saying in the computer industry "Garbage in, garbage out.". If the input is biased, the output will be biased.
If the AI's job is to assess the odds of recidivism, taking into account all available data, then it's neither going to go out of its way to be racist, nor go out of its way not to be racist.
What the heck is wrong with computer engineers? You guys think "oh, the problem can't be bad programming, the computer is never wrong. It has to be the user. Somehow."
No, of course it didn't "go out of its way" to be racist. It just happened that the results were racist. One easy explanation for this is that the racism was inherent in the input data.
If it's showing a bias against black convicts, presumably that's because black convicts really do have worse recidivism rates for whatever reason.
So, what you're saying here is that you didn't read the article. (here)
The point of the article was that data showed that the predictions of the AI did not match the data, and the way in which they did not match the data was that they overpredicted recidivism for blacks, and underpredicted recidivism for whites.
(Of course, that's 'recidivism rate' according to the data. It doesn't disprove, say, the existence of a racist police force with a racially-biased arrest pattern.)
True. That's another, different reason that an AI might give output results that are racist.
I'd be willing to bet that if you did a backtesting study, pitting the AI against human judges, the AI would beat them.
The data showed that the AI was slightly better than a coin flip.
Slightly.
It might well also be far more racist, as the judges are likely to want to discount race on moral grounds. They don't just want an AI that predicts recidivism rates, they want one that does so whilst incorporating our senses of morality and fairness. They're obviously not the same thing.
So, basically you're repeating here that you didn't read the article.
But..in the article, it said they were NOT using race as an AI training factor....so, it wasn't racial bias being programmed in.
Racial bias very clearly was programmed in. It turned out to have been programmed in by using weighting factors that were themselves dependent on race.
The problem is not that the data set reflects the reality. The problem is not that the AI makes mistakes, but that the particular mistakes the AI makes reflect the bias of the society that programmed it.
Why do you use the term 'mistakes'? If the AI accurately reflects the culture that feeds the data to it, it is not a mistake.
It wasn't supposed to "reflect the culture that feeds data to it." It was supposed to predict the probability that a given person on parole would commit a future crime. It did that wrong. And it did that wrong in a specific way, that preferentialy assumed that blacks were more likely to commit crimes than the data shows that they actually did, and that whites were less likely.
Sounds like just faulty programming on that article you referred to...
Indeed, I would consider racial bias to be a subset of "faulty programming."
"Further, the fact that more people of a particular race are prosecuted is not a reflection of bias in the data, rather a bias in the prosecution."
In this case, "persecuted" was more accurate.
Data is Data. It cannot exhibit a bias.
I can only surmise that you're not an experimental scientist. Data has bias all the time.
In physics (my field) the bias usually has no social consequence-- astronomical statistics, for example, are biased toward bright stars (since they're much easier to see than faint ones, and hence overrepresented in the data set). In social "sciences," however, the bias very often does have social consequences. SAT scores from children whose parents spend tens of thousands of dollars on SAT Prep courses, for example-- surprise!-- score better on SAT exams than ones who don't. The data shows a correlation of SAT score with parental income. Is this real? Better correct for the SAT-prep course effect before making a conclusion.
Data is biased. All the time. Be ready for it.
...Plus, being from the Guardian, I am skeptical that they didn't twist the data some to obtain their desired outcome, which ironically touches on the subject of this story.
Huh? MIT Tecnology Review and Propublica were the source. The link in the summary was this: https://www.axios.com/algorith... which linked here: https://www.propublica.org/art... and here MIT Technology Review