Chinese insurer’s facial-recognition software offers a cautionary tale(bloomberg.com)
bloomberg.com
Chinese insurer’s facial-recognition software offers a cautionary tale
https://www.bloomberg.com/opinion/articles/2019-06-14/china-knows-how-to-take-away-your-health-insurance
26 comments
Not every dynamic system is self correcting, in fact if the hypothesis is that a system is ‘stable’ in that sense the next step in the discussion should bear on:
- the forces in action and their relative directions
- the regime within which we could consider dynamic changes to be stable
- the time range we should expect the system to converge to the hypothesized equilibrium state
Most importantly in my opinion, is not assuming that a stable regime exists within a time/space envelope that corresponds with a physically achievable state in the real system we are trying to approximate.
I would put that sort if assumption into the category of ‘common falacies.’
- the forces in action and their relative directions
- the regime within which we could consider dynamic changes to be stable
- the time range we should expect the system to converge to the hypothesized equilibrium state
Most importantly in my opinion, is not assuming that a stable regime exists within a time/space envelope that corresponds with a physically achievable state in the real system we are trying to approximate.
I would put that sort if assumption into the category of ‘common falacies.’
I should not have used the term "self correcting" as it is suggestive of fitting neatly into a dynamic system model - I was envisioning more of a social change, where the collective views of society reach a tipping point (though may be gradual) as the driving principles of such organizations work to devalue it's own product, it's hard to model the views of society, but there have been many such changes throughout history.
It won't, because the insurance doesn't become worthless. It only means that, if you were able to get it, you're less likely to need a payout.
I'm offering an extreme dichotomy to highlight the effect... reality will not be so extreme but the value difference will be great enough to cause competition (self correction). Another aspect of a company focused on reducing payout by selection is that they will probably be less prepared to support customers who actually need help, it all feeds into an opportunity for competition as the product of the gambling insurer has less and less value.
It boils down to two quite opposed principles:
1. I want to try to prevent payout and focus on taking home as much of the sales as possible.
2. I want to try to spread the cost of health care and provide an honest and valuable service.
It's hard do both well. I'd rather be in boat #2 in case I actually need help i'd trust them to be to be more capable of providing it since they are expecting it rather than trying to avoid it. If enough people realise this, they will all head for option 2, and option 1s will have to change their principles or pivot into some sort of health vetting service.
It boils down to two quite opposed principles:
1. I want to try to prevent payout and focus on taking home as much of the sales as possible.
2. I want to try to spread the cost of health care and provide an honest and valuable service.
It's hard do both well. I'd rather be in boat #2 in case I actually need help i'd trust them to be to be more capable of providing it since they are expecting it rather than trying to avoid it. If enough people realise this, they will all head for option 2, and option 1s will have to change their principles or pivot into some sort of health vetting service.
Looks to me like an insurer that manages to discriminate who they insure, but pays out as well or better than non-discriminating insurers (say, because they have more money, because they need to pay less often), will come out on top.
I don't think the principles of trying to maximize profit, and paying out insurance, are so acutely opposed they could not coexist in the same company. It's just the conflict between cutting costs and maintaining quality - something almost all companies face.
I don't think the principles of trying to maximize profit, and paying out insurance, are so acutely opposed they could not coexist in the same company. It's just the conflict between cutting costs and maintaining quality - something almost all companies face.
China already has universal healthcare, it’s just not very good yet.
Universal healthcare? What does that mean?
In China, we only have one type of healthcare, the one that needs to be paid. There is no healthcare for free here.
(Unless of course, you're a high ranking CCP party member, that way you can enjoy all the goodies from all over the country for free, including free blood plasma from young people* to keep you feel good when you're old. *Caution: HIV maybe included due to technical difficulties).
In China, we only have one type of healthcare, the one that needs to be paid. There is no healthcare for free here.
(Unless of course, you're a high ranking CCP party member, that way you can enjoy all the goodies from all over the country for free, including free blood plasma from young people* to keep you feel good when you're old. *Caution: HIV maybe included due to technical difficulties).
Technically we (and our employers) pay for it via payroll deductions from our paycheck (the one labeled social health insurance? They then gave us a healthcare benefit account (in my case linked to bank of Beijing). It was all very useless, but technically China is listed as a country with universal healthcare.
Never donate blood in China, unless you are forced to by your school/work unit/whatever.
Never donate blood in China, unless you are forced to by your school/work unit/whatever.
I know... i used it. It's nothing like the NHS.
Phrenology makes a hearty comeback this time with the help of AI.
Psychology based on the skull vs psychology based on the face and "micro-expressions"
http://www.victorianweb.org/science/phrenology/intro.html
1.The brain is the organ of the mind.
2. The mind is composed of multiple distinct, innate faculties.
3. Because they are distinct, each faculty must have a separate seat or "organ" in the brain.
4. The size of an organ, other things being equal, is a measure of its power.
5. The shape of the brain is determined by the development of the various organs.
6. As the skull takes its shape from the brain, the surface of the skull can be read as an accurate index of psychological aptitudes and tendencies.
Psychology based on the skull vs psychology based on the face and "micro-expressions"
http://www.victorianweb.org/science/phrenology/intro.html
1.The brain is the organ of the mind.
2. The mind is composed of multiple distinct, innate faculties.
3. Because they are distinct, each faculty must have a separate seat or "organ" in the brain.
4. The size of an organ, other things being equal, is a measure of its power.
5. The shape of the brain is determined by the development of the various organs.
6. As the skull takes its shape from the brain, the surface of the skull can be read as an accurate index of psychological aptitudes and tendencies.
It's quite a bit different in that AIs don't make such chains of reasoning. They just "notice" correlations between inputs and training data with no regard to mechanism. We might construct a chain of causation between a microexpression -> social status -> stress -> blood pressure -> life expectancy, but the AI doesn't and doesn't care. It's about as likely to judge by the lighting of the photo as the presumed character of the subject.
Such AI judgement may be as or more racist than human judgement, given that they could make correlations based on immutable characteristics. In a sense they're only as biased as the training data, but that may also be true of many human racists.
Such AI judgement may be as or more racist than human judgement, given that they could make correlations based on immutable characteristics. In a sense they're only as biased as the training data, but that may also be true of many human racists.
#4, 5, 6 are very dubious.
Elephants have huge brains yet they don’t rule the world.
Elephants have huge brains yet they don’t rule the world.
Other things aren't equal.
Brain size correlates strongly with body size.
Also: Phrenology is bunkum.
Brain size correlates strongly with body size.
Also: Phrenology is bunkum.
BTW -- The author (Cathy O'Neil) is awesome.
This particular problem has always existed in insurance. The best way to make a profit is to not insure sick people. I was hoping the cautionary tale was that it didn't work for the insurer, rather than it being a ripoff for the customer.
Digging a little deeper, it looks like their plan is to implement a telemedicine+drug kiosk system at Chinese scale which is interesting. That should work.
Their PR person claims they hit 80% 'positive' rate on detecting lies using micro-expressions. Seems... unlikely. When the dust settles, it will probably just turn out that that preventing people from lying about identity so you can't escape your past has all the value and the rest of it is worthless to the insurer.
https://www.mobihealthnews.com/content/ping-good-doctor-show...
https://www.forbes.com/sites/shuchingjeanchen/2018/06/06/chi...
This particular problem has always existed in insurance. The best way to make a profit is to not insure sick people. I was hoping the cautionary tale was that it didn't work for the insurer, rather than it being a ripoff for the customer.
Digging a little deeper, it looks like their plan is to implement a telemedicine+drug kiosk system at Chinese scale which is interesting. That should work.
Their PR person claims they hit 80% 'positive' rate on detecting lies using micro-expressions. Seems... unlikely. When the dust settles, it will probably just turn out that that preventing people from lying about identity so you can't escape your past has all the value and the rest of it is worthless to the insurer.
https://www.mobihealthnews.com/content/ping-good-doctor-show...
https://www.forbes.com/sites/shuchingjeanchen/2018/06/06/chi...
The future resembles the past.
Dan Bouk, of Colgate University, is a historian of bureaucracy, including insurance, which has a long history of discrimination:
His work investigates the ways that corporations, states, and the experts they employ have used, abused, made, and re-made the categories that structure our daily experiences of being human. His first book, How Our Days Became Numbered: Risk and the Rise of the Statistical Individual (Chicago, 2015), explored the spread into ordinary Americans' lives of the United States life insurance industry's methods for quantifying people, for discriminating by race, for justifying inequality, and for thinking statistically.
http://www.colgate.edu/facultysearch/facultydirectory/dbouk
Dan Bouk, of Colgate University, is a historian of bureaucracy, including insurance, which has a long history of discrimination:
His work investigates the ways that corporations, states, and the experts they employ have used, abused, made, and re-made the categories that structure our daily experiences of being human. His first book, How Our Days Became Numbered: Risk and the Rise of the Statistical Individual (Chicago, 2015), explored the spread into ordinary Americans' lives of the United States life insurance industry's methods for quantifying people, for discriminating by race, for justifying inequality, and for thinking statistically.
http://www.colgate.edu/facultysearch/facultydirectory/dbouk
These systems can be extremely profitable even when they give massive amount of false positives.
Example:
Without algorithm:
For every 100,000 customers true 1,000 positives. Each positive leads to $1,000 payoff (on average). The cost is $10 per customer.
With algorithm:
For every 100,000 _potential_ customers the algorithm identifies 9,500 false positives and 500 true positives. You reject these 10,000 customers. Now you have 500 true positives for each 90,000 customers. The cost is $5.56 per customer.
The end result of extensive customer profiling is that wrong friends in the social media, wrong words, or any physical trait that has little correlation with something cost incurring can increase the cost of your life.
Example:
Without algorithm:
For every 100,000 customers true 1,000 positives. Each positive leads to $1,000 payoff (on average). The cost is $10 per customer.
With algorithm:
For every 100,000 _potential_ customers the algorithm identifies 9,500 false positives and 500 true positives. You reject these 10,000 customers. Now you have 500 true positives for each 90,000 customers. The cost is $5.56 per customer.
The end result of extensive customer profiling is that wrong friends in the social media, wrong words, or any physical trait that has little correlation with something cost incurring can increase the cost of your life.
The AI Winter is coming. Are you strong enough to survive?
These are all great points.
1. How can you expect the AI to determine liars?
2. Risk pooling versus segmentation is bound to be controversial, especially in health.
1. How can you expect the AI to determine liars?
2. Risk pooling versus segmentation is bound to be controversial, especially in health.
At the end of the day it's a con to bilk people out of more money or deny insurance to those that they don't like or maybe they have some "secret" stolen information that the person is a insurance risk.
It’s reasonable to charge fat people more. It’s something you can measure objectively. It’s also something under your control.
The lie detector is just bunk though. If something is really a factor then measure it or test for it objectively.
The lie detector is just bunk though. If something is really a factor then measure it or test for it objectively.
>It’s reasonable to charge fat people more.
It's only reasonable insofar as you view insurance as a profit-maker and deprive it of its social function of financial relief.
It's only reasonable insofar as you view insurance as a profit-maker and deprive it of its social function of financial relief.
Not allowing insurers to charge people who engage in unhealthy behaviour is effectively forcing people who make good health decisions to subsidize people who make bad health decisions. From an economic perspective this has a negative social function, incentivising bad behaviour and punishing good.
Even if it’s not a profit-maker, someone has to pay, and if someone isn’t paying the appropriate premium for their level of risk, someone else in the pool is bearing the cost.
Well, I'm fat and I don't think other people should have to pay more for my lifestyle choices. Doesn't make any difference if the insurance system is for profit or not.
The further they go down this road, the more their acceptance criteria will converge on insuring people that never need a payout. Once people clock onto that fact, the insurance becomes worthless - At which point other insurance companies will be able to come in and compete on a non-discriminatory basis and operate on the principle of a tax that spreads cost rather than a gamble that attempts to never actually deliver value to the customer.
... or you could just skip straight to an actual tax based health system and cut out the profiteering middle man. Yes I like the NHS.