AI SUCKS

[bestkeptsecret678]

The bigger, more sophisticated banks will have newer technology to make this less of a problem for you. Outsmarting the fraudsters is hard, expensive, and unfortunately, honest customers like you can be inconvenienced. If your financial institution is smaller than the big guys, then they may not have access to modern fraud prevention systems to deter the Crooks that won’t inconvenience you.

Sounds clinical, I realize, but the universe of criminals defrauding banks and their card holders is a serious issue. Its going to get tougher, not easier.

 

[pink_silk_glove]

Well at the risk of turning this into the politics of economics, banks are by far the biggest perpetrators of fraud in the first place. The irony that they would be protecting us from fraud is galactic. It's like a pimp beating up a rival pimp for beating up his ho and saying, "only I can beat up this ho!"

 

[Erozetta]

You can typically set a range for how far away in person charges can be made. I have mine set to be allowed in MA, NH, and ME.

And it doesn't require calling the bank, just logging into your account and going under manage cards then find the option for your regions. Mine is under controls and alerts then location.

You can also typically set up what type of merchants you want to always allow and which ones you don't. (So you could mark restaurants and gas stations as clear to use, but not drugstores or the like.)

 

[Robin_Starveling]

This has nothing to do with AI. Back in the ’90s, I got a call from my card company asking if I was in Thailand. When I said no, they told me someone had used my card there but assured me they were handling it. So, the practice of location verification existed well before AI was a thing.

The strange part of your story isn’t the security check, it’s that your card company froze everything without even reaching out to you first.

If real AI had been involved, it probably would’ve used your phone location, security cams, and other data points to figure out where you actually were, saving you the hassle.

 

[Bramblethorn]

This has nothing to do with AI. Back in the ’90s, I got a call from my credit card company asking if I was in Thailand. When I said no, they told me someone had used my card there but assured me they were handling it. So, the practice of location verification existed well before AI was a thing.

"AI" is a nebulous term that means whatever a marketer wants it to mean, but machine learning tech was around and being used for fraud detection by the 1990s.

SAS (the software company) claims "SAS has been a pioneer in machine learning since the 1980s, when neural networks were first used to combat credit card fraud". I can't find any further info on the 1980s, but here's a paper from 1994 discussing Mellon Bank's use of artificial neural networks for CC fraud detection.

 

[Lifestyle66]

Nothing has changed in my job circumstances; I've been doing this for years and never had a problem.

Its not like im buying a fucking TV. I'm trying to get lunch.

And its not like im thousands of miles away. I travel within NJ and PA, usually no more than a 100 miles give or take.

No, this is very new.

The bank/credit card industry is trying to decrease the fraud which they cover at a loss.

Last March, we were traveling to Florida for about the seventh time in 5 years. Shortly after we arrived at our destination, I received a text alert that my Shell fuel card had been declined, ... at a Lowes. It's just like any other general purchase credit card and I SHOULD be able to use it at Lowes. But I didn't, and for whatever reason, they declined that purchase.

We had stopped the previous day for gas at a small, rural gas station, and I think there was a credit card skimmer there, which caught the card number, etc for the scammer to try using. If the card company had allowed that transaction to go through and I later challenged it, I'm only responsible for about $50 and the company is stuck losing the rest of whatever it cost.

 

[TheLobster]

"AI" is a nebulous term that means whatever a marketer wants it to mean, but machine learning tech was around and being used for fraud detection by the 1990s.

SAS (the software company) claims "SAS has been a pioneer in machine learning since the 1980s, when neural networks were first used to combat credit card fraud". I can't find any further info on the 1980s, but here's a paper from 1994 discussing Mellon Bank's use of artificial neural networks for CC fraud detection.

At heart, expert systems like these (which is another term from the same bucket) are just weighted sum of some predetermined signals, along with a filter which eventually turns those sums into binary decisions based on various thresholds. It can be a neural network, or it can be a loop with one variable and a condition after it.

 

[FrancesScott]

"AI" is a nebulous term that means whatever a marketer wants it to mean, but machine learning tech was around and being used for fraud detection by the 1990s.

This area is rife with misinformation and marketing spin (pretty much the same thing). Nothing about machine learning and / or LLMs merits the I tag in AI, let alone AGI.

For as long as technology has been created it has been anthropomorphized to sell it to humans who know no better. ChatGPT is not your friend or confidant. It’s a statistical inference engine lashed to natural language processing. It’s the very definition of the Wizard of Oz behind the curtain.

LLMs are great for getting a quick orientation about an area, as a precursor to proper research. That is as long as you don’t care if the initial orientation may be wildly inaccurate or incomplete.

Anyone who thinks LLMs are experts in anything is going to get a big surprise. They have a glossy veneer which can suggest expertise to those who know little about an area and lack the skills to find out themselves. But to someone who actually has knowledge, their output is often laughably wrong.

 

[TheLobster]

We had stopped the previous day for gas at a small, rural gas station, and I think there was a credit card skimmer there, which caught the card number, etc for the scammer to try using. If the card company had allowed that transaction to go through and I later challenged it, I'm only responsible for about $50 and the company is stuck losing the rest of whatever it cost.

This is what I call a "US-inflicted wound." The whole problem is the prevalence of credit cards used for fast and mindless payments, and the questionable security measures exercised when a payment is made. It's all designed to maximum throughput rather than minimize fraud.

Here in Europe, until the advent of RFID pay-by-touch cards, the smallest purchase would require you to input the PIN. Credit cards are also generally much less prevalent, and mostly everyone simply uses their debit cards. Merely knowing the numbers of the card is not enough, including CVC/CVV, to make a purchase in a physical PoS.

And if you're wondering how restaurants work... Well, we don't just give away the card to the waiter; rather, when it's time to pay the bill he comes around with a mobile terminal that's functionally equivalent to those cashier's have at the registers. Your card never leaves your hands, let alone sight.

 

[Lifestyle66]

This is what I call a "US-inflicted wound." The whole problem is the prevalence of credit cards used for fast and mindless payments, and the questionable security measures exercised when a payment is made. It's all designed to maximum throughput rather than minimize fraud.

Here in Europe, until the advent of RFID pay-by-touch cards, the smallest purchase would require you to input the PIN. Credit cards are also generally much less prevalent, and mostly everyone simply uses their debit cards. Merely knowing the numbers of the card is not enough, including CVC/CVV, to make a purchase in a physical PoS.

And if you're wondering how restaurants work... Well, we don't just give away the card to the waiter; rather, when it's time to pay the bill he comes around with a mobile terminal that's functionally equivalent to those cashier's have at the registers. Your card never leaves your hands, let alone sight.

It's not just the card number.

When you must enter a pin, zipcode, or CVC number, all of that can be captured and re-used for purchases elsewhere. And I've never had any waiter in any restaurant copy my card (in forty years!) This incident was when I used the card at a gas pump at a gas station. I had never handed that card to anyone, ever, because I only used it for fuel purchases at the pump. And yet, a scammer still caught it all.

Some restaurants in the US are going to that same method of bringing around the card payment device to the customer's table, and the waiters are told to never touch the customer's card. But, as in Europe, YOU the customer are paying for those additional devices.

EDIT: There was onetime decades ago when I found a fraud charge on my credit card statement, back in the days of paper slips we had to sign. The charge was on my statement THREE YEARS after I moved away from that area to the other side of the country! So, it was a case of someone finding one of those old slips of paper before they required those CVC numbers. I called the credit card company, and they deleted the entire charge from my bill.

 

[Bramblethorn]

This area is rife with misinformation and marketing spin (pretty much the same thing). Nothing about machine learning and / or LLMs merits the I tag in AI, let alone AGI.

I agree. My point was that if we're going to call today's algorithmic CC fraud detection "AI" (which I agree is a misnomer) then that kind of "AI" has been around for quite a while.

Of course the technology has developed since the 1980s-90s, and the hardware has advanced tremendously. But it's still ANNs and similar ML tech trying to answer the question "what is the probability of this transaction being fraudulent?"

 

[FrancesScott]

I agree. My point was that if we're going to call today's algorithmic CC fraud detection "AI" (which I agree is a misnomer) then that kind of "AI" has been around for quite a while.

Of course the technology has developed since the 1980s-90s, and the hardware has advanced tremendously. But it's still ANNs and similar ML tech trying to answer the question "what is the probability of this transaction being fraudulent?"

I agree entirely.

 

[Bramblethorn]

If the card company had allowed that transaction to go through and I later challenged it, I'm only responsible for about $50 and the company is stuck losing the rest of whatever it cost.

And in the long run, some of those losses come back to merchants and card users in the form of fees/etc. Or they decide that they're unwilling to cover certain kinds of high-risk business at all, and that becomes a nuisance to us.

 

[iwatchus]

Of course the technology has developed since the 1980s-90s, and the hardware has advanced tremendously. But it's still ANNs and similar ML tech trying to answer the question "what is the probability of this transaction being fraudulent?"

The commercial successes of the late 80's and early 90's (Amex card purchase approvals was the landmark one) were all logic programming based and knowledge based, but not related to ANN's at all. There are so many other AI and ML technologies that had and still have value, but the ANN's captured the limelight and refuse to give it up even though their failures are becoming more and more obvious.

 

[Bramblethorn]

The commercial successes of the late 80's and early 90's (Amex card purchase approvals was the landmark one) were all logic programming based and knowledge based, but not related to ANN's at all.

Acknowledging the importance of expert systems, but I did provide a link specifically about ANN fraud detection tech; that paper was published in 1994 which probably means the tech was in use at least a year before that. I'd call that early 90s?

There are so many other AI and ML technologies that had and still have value, but the ANN's captured the limelight and refuse to give it up even though their failures are becoming more and more obvious.

Uh huh. People love black-box algos.