3 ms·

Even if a piece of information isn't informative on an individual basis, it might provide improve the performance of a model that combines it with other inputs.

Even if a piece of information isn't informative on an individual basis, it might provide improve the performance of a model that combines it with other inputs. You don't need to posting about doing crimes for your FB profile to be useful for risk assessment. Your connections or other posts (e.g. about how some emergency means you're suddenly short of cash) could contribute to an elevated score. The proposal doesn't go into much detail about how they imagine it will work, but it sounds like they're just saying that a statistical model (based on a larger set of data) would be more useful than a set of rules based on a tiny set of inputs. This seems entirely reasonable. Here is the relevant paragraph: Input data for MSA platforms should incorporate not only transactional data, customer static data and internal reference lists, but also other dynamic behavioural customer information where proportionate to the risk (e.g. device ID, IP addresses). Using an RBA, input data may also include data from reputable external, publicly available sources, including information on company structures, Ultimate Beneficial Owners (UBO), and watch lists, as well as complementary sources such as market data and verified customer social media accounts. Finally, FIs should establish robust data governance and quality control frameworks.

AnthonyMouse 2y ago
> Your connections or other posts (e.g. about how some emergency means you're suddenly short of cash) could contribute to an elevated score. This sounds more like a way to send desperate people into a death spiral by raising interest rates or denying credit availability right at the worst possible moment than any kind of system that would prevent crime. > The proposal doesn't go into much detail about how they imagine it will work, but it sounds like they're just saying that a statistical model (based on a larger set of data) would be more useful than a set of rules based on a tiny set of inputs. This seems entirely reasonable. It's implying they want to invade everyone's privacy to set up a hallucinatory black box that destroys lives at random under the guise of... I'm not even sure what this is supposed to do. It's not going to yield anything that would be useful in a prosecution so it's more like opaque government-facilitated denial of service without due process. Far from reasonable, it's total crap.
- rrnechmech 2y ago
  This is really awful. Guilty by algo until proven that you can -pay- your way out
- rahimnathwani 2y ago
  > opaque government-facilitated denial of service without due process That's not what the report is about. Banks and financial institutions are required to file Suspicious Activity Reports (SARs) and Suspicious Transaction Report (STRs). Filing such a report doesn't stop a transaction from happening. The report suggests that banks think they could improve the usefulness of these reports and reduce the number of false positives, if they had more freedom to choose the criteria.
  - AnthonyMouse 2y ago
    > Banks and financial institutions are required to file Suspicious Activity Reports (SARs) and Suspicious Transaction Report (STRs). Filing such a report doesn't stop a transaction from happening. It kind of does though, because banks are also not allowed to do business with criminals, so if they suspect you of being engaged in criminal activity -- even if they have no real proof -- guess what they do. Worse than that, they're prohibited from telling you about the SARs, so you have no way to dispute them if they're inaccurate. It's a Kafkaesque absurdity that should simply be abolished rather than attempting to polish the turd with AI pixie dust. > The report suggests that banks think they could improve the usefulness of these reports and reduce the number of false positives, if they had more freedom to choose the criteria. Which is bollocks, but the government keeps pressuring them to do something because the system doesn't work (because banks shouldn't be doing it to begin with), so they're proposing to invade everyone's privacy because they want to do it anyway and this is a convenient excuse.
Aspos 2y ago
Fighting crime should not be a bank problem. Putting this burden on banks is a thing pretty unique to the US. Not sure why anyone would be surprised to find out this does not work.
- cromulent 2y ago
  It's not at all unique to the US. The EU has AML/KYC legislation, which pushes the burden down through the banks to the customer.
  - Aspos 2y ago
    Kind of true, yes, in theory, because FATF is largely controlled by the US and it enforced US-style rules to the rest of the world. However, on the practical level AML/KYC works differently outside of the US. Regulator would create well-defined criteria for banks to execute: To onboard a customer bank must do this and this, check that field against that record. In the US regulators just washed their hands and placed the burden on banks: banks must make all the "reasonable" efforts to "know" their customer and what is is deemed "reasonable" varies across banks, branches and even individual clerks.

AnthonyMouse 2y ago

> Your connections or other posts (e.g. about how some emergency means you're suddenly short of cash) could contribute to an elevated score. This sounds more like a way to send desperate people into a death spiral by raising interest rates or denying credit availability right at the worst possible moment than any kind of system that would prevent crime. > The proposal doesn't go into much detail about how they imagine it will work, but it sounds like they're just saying that a statistical model (based on a larger set of data) would be more useful than a set of rules based on a tiny set of inputs. This seems entirely reasonable. It's implying they want to invade everyone's privacy to set up a hallucinatory black box that destroys lives at random under the guise of... I'm not even sure what this is supposed to do. It's not going to yield anything that would be useful in a prosecution so it's more like opaque government-facilitated denial of service without due process. Far from reasonable, it's total crap.

rrnechmech 2y ago
This is really awful. Guilty by algo until proven that you can -pay- your way out
rahimnathwani 2y ago
> opaque government-facilitated denial of service without due process That's not what the report is about. Banks and financial institutions are required to file Suspicious Activity Reports (SARs) and Suspicious Transaction Report (STRs). Filing such a report doesn't stop a transaction from happening. The report suggests that banks think they could improve the usefulness of these reports and reduce the number of false positives, if they had more freedom to choose the criteria.
- AnthonyMouse 2y ago
  > Banks and financial institutions are required to file Suspicious Activity Reports (SARs) and Suspicious Transaction Report (STRs). Filing such a report doesn't stop a transaction from happening. It kind of does though, because banks are also not allowed to do business with criminals, so if they suspect you of being engaged in criminal activity -- even if they have no real proof -- guess what they do. Worse than that, they're prohibited from telling you about the SARs, so you have no way to dispute them if they're inaccurate. It's a Kafkaesque absurdity that should simply be abolished rather than attempting to polish the turd with AI pixie dust. > The report suggests that banks think they could improve the usefulness of these reports and reduce the number of false positives, if they had more freedom to choose the criteria. Which is bollocks, but the government keeps pressuring them to do something because the system doesn't work (because banks shouldn't be doing it to begin with), so they're proposing to invade everyone's privacy because they want to do it anyway and this is a convenient excuse.

Aspos 2y ago

Fighting crime should not be a bank problem. Putting this burden on banks is a thing pretty unique to the US. Not sure why anyone would be surprised to find out this does not work.

cromulent 2y ago
It's not at all unique to the US. The EU has AML/KYC legislation, which pushes the burden down through the banks to the customer.
- Aspos 2y ago
  Kind of true, yes, in theory, because FATF is largely controlled by the US and it enforced US-style rules to the rest of the world. However, on the practical level AML/KYC works differently outside of the US. Regulator would create well-defined criteria for banks to execute: To onboard a customer bank must do this and this, check that field against that record. In the US regulators just washed their hands and placed the burden on banks: banks must make all the "reasonable" efforts to "know" their customer and what is is deemed "reasonable" varies across banks, branches and even individual clerks.