4 ms·

The argument seems to be a little different than what you’ve taken from it. My interpretation was something like and efficient frontier model between multiple

The argument seems to be a little different than what you’ve taken from it. My interpretation was something like and efficient frontier model between multiple variables where “zero fraud” isn’t actually a position on that frontier. So, if you find a place with zero fraud, you can possibly increase the total utility of the system by aiming for slightly less than perfect but being back on the efficient frontier. Arguably zero of anything is a great ideal but not maximally efficient. Morally, the situation is more in favour of getting to zero.

regularfry 3y ago
I don't think you can even make a moral argument in favour of zero fraud, because that implies choosing to harm genuine users. There's always a trade-off between fraud prevention and genuine use. This is extremely important in social benefit systems, where people can literally die if the system incorrectly thinks they are trying to defraud it and cuts them off. Non-zero fraud is useful for political point-scoring, but zero fraud is a terrible goal on its own. You also need to be measuring false positive cost, and drive that to zero too. The moral argument has to be for the efficiency frontier itself, I can't see any other way this works.
- eviks 3y ago
  Why is 0 terrible if it's also on the efficiency frontier? And what's you moral argument for the level of fraud at the frontier? Is the same efficiency better with lower levels of fraud or higher or indeterminate?
  - regularfry 3y ago
    > Why is 0 terrible if it's also on the efficiency frontier? When would this be the case? I'm having trouble picturing it. > And what's you moral argument for the level of fraud at the frontier? The frontier is exactly the point at which you cannot reduce the amount of fraud without it costing genuine users (and you) more than you gain. So if you're at the frontier and you want to reduce the amount of fraud further, you are acknowledging that harms to users are less important to you than losses from fraud, and doing so would be immoral. The efficient frontier should be the point of least moral harm, almost by definition - at least, as long as all the relevant externalities are priced in. > Is the same efficiency better with lower levels of fraud or higher or indeterminate? That's a trolley problem question. Higher levels of fraud at the same level of efficiency would mean there's a corresponding rise in value delivered from genuine usage of the service, so cost/benefit is the same and preference boils down to personal choice. But it sidesteps the more difficult question, which is how taking action to reduce the level of fraud actually plays out: there'll be diminishing returns as you take more and more extreme steps. Each step will likely have an incrementally more harmful effect on the genuine users, so efficiency won't remain the same.
SideburnsOfDoom 3y ago
You can find a place with zero fraud by denying all transactions. This is of course, maximally inefficient and totally useless. It's less about "allowing fraud" and more about managing false positives in the fraud detection subsystem: it should ideally detect fraud, and only fraud. The false positives need to be low enough for the system to work, without also having so much fraud that the system stops working. Ideally there are no false positives at all, but at some point the effort to reduce this further exceeds the returns. Like with fraud.
eviks 3y ago
Efficiency frontiers are a game of fantasy (reality is never at the frontier), so it's not relevant whether there is a "zero fraud" point, but also for the same reason a tough argument to make that there are no situations where (even in theory!) such a frontier exists But that's also not connected to the main critique - I get how you'd welcome more efficiency (at higher costs), but do you get to welcoming costs outside of trying to do counterintuitive rhetoric?

regularfry 3y ago

I don't think you can even make a moral argument in favour of zero fraud, because that implies choosing to harm genuine users. There's always a trade-off between fraud prevention and genuine use. This is extremely important in social benefit systems, where people can literally die if the system incorrectly thinks they are trying to defraud it and cuts them off. Non-zero fraud is useful for political point-scoring, but zero fraud is a terrible goal on its own. You also need to be measuring false positive cost, and drive that to zero too. The moral argument has to be for the efficiency frontier itself, I can't see any other way this works.

eviks 3y ago
Why is 0 terrible if it's also on the efficiency frontier? And what's you moral argument for the level of fraud at the frontier? Is the same efficiency better with lower levels of fraud or higher or indeterminate?
- regularfry 3y ago
  > Why is 0 terrible if it's also on the efficiency frontier? When would this be the case? I'm having trouble picturing it. > And what's you moral argument for the level of fraud at the frontier? The frontier is exactly the point at which you cannot reduce the amount of fraud without it costing genuine users (and you) more than you gain. So if you're at the frontier and you want to reduce the amount of fraud further, you are acknowledging that harms to users are less important to you than losses from fraud, and doing so would be immoral. The efficient frontier should be the point of least moral harm, almost by definition - at least, as long as all the relevant externalities are priced in. > Is the same efficiency better with lower levels of fraud or higher or indeterminate? That's a trolley problem question. Higher levels of fraud at the same level of efficiency would mean there's a corresponding rise in value delivered from genuine usage of the service, so cost/benefit is the same and preference boils down to personal choice. But it sidesteps the more difficult question, which is how taking action to reduce the level of fraud actually plays out: there'll be diminishing returns as you take more and more extreme steps. Each step will likely have an incrementally more harmful effect on the genuine users, so efficiency won't remain the same.

SideburnsOfDoom 3y ago

You can find a place with zero fraud by denying all transactions. This is of course, maximally inefficient and totally useless. It's less about "allowing fraud" and more about managing false positives in the fraud detection subsystem: it should ideally detect fraud, and only fraud. The false positives need to be low enough for the system to work, without also having so much fraud that the system stops working. Ideally there are no false positives at all, but at some point the effort to reduce this further exceeds the returns. Like with fraud.

eviks 3y ago

Efficiency frontiers are a game of fantasy (reality is never at the frontier), so it's not relevant whether there is a "zero fraud" point, but also for the same reason a tough argument to make that there are no situations where (even in theory!) such a frontier exists But that's also not connected to the main critique - I get how you'd welcome more efficiency (at higher costs), but do you get to welcoming costs outside of trying to do counterintuitive rhetoric?