Why Banks Are Suddenly Closing Down Customer Accounts

Re:Why Banks Are Suddenly Closing Down Customer Accounts by nine_k on Sunday November 05, 2023 @11:29PM (#38158870)

Listing the specific problem that led to a service denial could be helpful, but could be not. I bet that it's very rarely something like "You did <this one forbidden thing>, so we ban you according to the rule 3.6.4.2 of ToS". It could be something like: "The pattern of your activity is very similar to patterns of activity of 9823 known scam attacks". Or even more detailed: "8 out of 11 statistical models analyzing 982 parameters of your activity during last 48 days showed the probability of your account being used for illicit fund transfers above 0.92, and the remaining 3, above 0.57. In particular, our 100M-paremeter AI model (link to a PDF with technical details) has classified your activity thusly: (a 20-row table follows). According to the law, we use a publicly available model, widely accepted as a standard in the industry. You can inspect the model's weights at huggingface (link follows)".

Interpreting such results is a job. Disputing such results in a court, or with an arbiter, would be really hard, unless you're a practitioner of the pattern-recognition craft yourself.

[ Parent | on HN ]

Re:Why Banks Are Suddenly Closing Down Customer Accounts by oblio on Sunday November 05, 2023 @11:53PM (#38158998)

Doesn't matter. Just forcing them to disclose that info would force them to clean up their act.

[ Parent | on HN ]

Re:Why Banks Are Suddenly Closing Down Customer Accounts by lamontcg on Monday November 06, 2023 @01:20AM (#38159461)

There's no need for the model weights and parameters. The probabilities output from a multilabel classification model would be entirely sufficient to tell what the model thought the user had done. It was trained on a category and it is claiming that the user is part of that category. That category should have a human description of what the corporate entity was looking for and that should be enough for some kind of arbiter to determine if the 'accused' was actually doing that or not. The question isn't what the model is doing, the question is if the person is a false positive or not, which should be able to be determined without any reference to the model at all. If the model is doing too many false positives then determining why the model is broken becomes the job of the corporation (ideally with some punitive financial damages as incentive).

Although I do think that the model weights and parameters should in some way be subpoenable so that the model can be 'compelled to testify' if someone has enough legal, technical and fiscal resources to sue the company.

[ Parent | on HN ]