Chinese insurer’s facial-recognition software offers a cautionary tale
bloomberg.com
bloomberg.com
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.
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).
Never donate blood in China, unless you are forced to by your school/work unit/whatever.
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.
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.
- 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.’
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.
Elephants have huge brains yet they don’t rule the world.
Brain size correlates strongly with body size.
Also: Phrenology is bunkum.
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.
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...
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.
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.
1. How can you expect the AI to determine liars?
2. Risk pooling versus segmentation is bound to be controversial, especially in health.
The lie detector is just bunk though. If something is really a factor then measure it or test for it objectively.
It's only reasonable insofar as you view insurance as a profit-maker and deprive it of its social function of financial relief.