Thousands in London Face Incorrect Benefit Cuts From Automated Fraud Detector (sky.com) 95
Thousands of people could soon be receiving letters threatening to cut off vital housing benefits as they face being incorrectly targeted by a new automated fraud detector. From a report: The government-backed London Counter Fraud Hub, developed by BAE, has been hailed a success after being trialled in four boroughs - Camden, Ealing, Croydon and Islington. Using vast quantities of data from millions of households, it is designed to target potential fraud cases involving the single person council tax discount, subletting in local authority housing and business rate relief and rating.
Ealing, the lead council for the project, found the automated elements of the system targeting single person discount fraud was 80% effective -- which is seen as an acceptable benchmark. With just over one million claimants of council tax single person discount in London, the London Counter Fraud Hub estimates it will detect around 40,000 fraudulent cases in the first year. Critics say the 20% error rate is unacceptable as around 8,000 people will receive letters wrongly accusing them of fraud.
Ealing, the lead council for the project, found the automated elements of the system targeting single person discount fraud was 80% effective -- which is seen as an acceptable benchmark. With just over one million claimants of council tax single person discount in London, the London Counter Fraud Hub estimates it will detect around 40,000 fraudulent cases in the first year. Critics say the 20% error rate is unacceptable as around 8,000 people will receive letters wrongly accusing them of fraud.
automation (Score:2, Informative)
Nothing wrong with using an automated system that assumes conclusions. As long as you have
1) A viable system in place for remedying the failures. If you're unsure about the soft word in there, peg it at "better than youtube's"
2) The claimed conclusions are asserted tentatively. (The data in our system suggests) YOU LIED ABOUT X AND Z WILL HAPPEN (if not addressed).
Unfortunately we fuck up both. Privatize the savings, socialize the fuck up, charge forward trusting the data without hesitation, why stop at fla
Re: (Score:1)
The number that abuse the system is miniscule. The bigger problem is at the other end with tax dodging companies and individuals with more money than they or their descendants would even be able to spend for centuries.
Re: (Score:2)
His what?
We don't need $30 aspirin and $250/ 15min doctor (Score:2)
We don't need $30 aspirin and $250/ 15min doctor bills.
I've got all me own teeth. In a jar. (Score:2)
It's all those Pollacks, Ruminants and Albino Huns. They'll all be sent packing soon thanks to that lovely Mr Farage.
He's so dapper, isn't he? Never see him with without a proper shirt and tie.
You couldn't get bananas. They all went to the brave boys in the trenches you know. Unless you dropped your knickers for any Yank that walked by, that is. So I heard.
Re: (Score:3, Interesting)
This is just a high-level screening, false positives are not a big deal.
FTFA:
"We will not cancel anyone's council tax discount without giving them a fair say, which is why we are writing to them first. The council has experience of dealing with vulnerable people and others who need help communicating and will ensure we use the best channels of communication and provide support where necessary, as we do already," the spokesperson said.
Re:80% False Positive !?!? (Score:4, Funny)
Translation: There will be a premium rate phone number that goes to India. It will never be answered, not that it would be any use if it was.
Re: (Score:2)
If you were suspected of tax fraud you probably earned enough that the Tories consider you to be human.
Only 4% fraud on 'government tit'? Bullshit! (Score:2, Insightful)
They do need to work on their error rate, but they should be working on the false negatives. The false positives can be handled with a human investigator.
Re: (Score:1)
For example, council tax is paid annually and is based (crudely) on property value. However, discounts and uplifts apply in certain circumstances. If 1 adult only lives at an address, they receive a 25% discount. If the property is unoccupied, it receives no discount, and may attract a supplement of up to 100%. If the property is occupied only by vulnerable people or students, then they receive a 100%
Re: (Score:2)
Don't let that strawman crush you.
It's obviously about being cost effective. Not even looking at 90% of scammers while only finding 20% in error is clearly not investigating enough leeches.
The key number would be ratio of cost of an investigation vs. cost of an undetected ongoing scam. As this is rent (expensive bene), they can and should be investigating _all_of_them_, about once/year. Same as SS disability grifters in the USA. But first they should investigate the government 'workers' assigned to inv
Re: (Score:2)
Nothing about this government is cost effective.
Re: (Score:2)
do you have evidence that the number should be more than 4%?
Re: (Score:2)
When free shit is being handed out, it's _always_ more than 4% scam.
Re: (Score:2)
The key number would be ratio of cost of an investigation vs. cost of an undetected ongoing scam. As this is rent (expensive bene), they can and should be investigating _all_of_them_, about once/year. Same as SS disability grifters in the USA. But first they should investigate the government 'workers' assigned to investigate the scumbags. They are always a big part of the problem.
The cost of fraud within Medicaid is much higher than private insurance fraud. However, the administrative cost is much lower. Turns out the total is about the same. It's not just the cost of investigation, it's the ongoing cost of more elaborate paperwork and process in the name of fraud prevention. Much better to have the simpler solution if it ends up costing the same - optimize for the normal people, don't punish them just to punish the grifters too.
Automated fraud detection has great promise to be
Re: (Score:2)
The data fed into the system provides "probable cause", and any follow-up/review will be done by civil servants that never, ever make a mistake.
Re: (Score:2)
I get the joke, but by definition "probable cause" needs to be more than 50%. This is the usual problem with profiling, of course.
Re: (Score:2)
4% sounds about right, or even high. The US has less fraud then that in its welfare programs, in terms of bad actors. Just some of the bad actors are doctors using fraud for millions with SSNs of old people on Medicare they never met.
Re: (Score:2)
Right, no one sells their SNAP benefits for 50 cents on the dollar to buy things not available for SNAP benefits... It's all doctors defrauding people on Medicare. /SMH
Re: (Score:2)
I'm pretty sure the question was "are people getting benefits they're not entitled to." Which is true of the doctors defrauding Medicare (getting money for unperformed procedures). The people selling their SNAP benefits usually are people entitled to SNAP benefits. Therefore, it's not "fraud" in the same sense that we are describing, even thought it is illegal.
Selling them is bad, sure. But in it's not different from buying the food and then reselling it. There's no real way to stop it on an individual
Re: (Score:2)
Re: (Score:2)
You contention just doesn't seem reasonable. In the case of housing single individuals, particularly one who are secually active often let boyfriends, girlfriends, etc. crash at their house for long periods of time. This would indicate they are defrauding the government if they are receiving money because they are living alone. How often would an unauthorized crasher have to stay at a place before the receiver would have to inform the government and stop taking benefits?
This is a typical problem in the U.S.
Old but fitting quote (Score:2)
"To err is human; to really foul things up you need a computer."
Re: (Score:2)
The UK did something similar before, I think it was with disability benefit. They outsourced it to a US company and paid them based on the number of cases processed and the amount of money saved.
But I'm sure a quadriplegic could work on a building site, if he made the effort.
Re: (Score:2)
But I'm sure a quadriplegic could work on a building site, if he made the effort.
Centrelink is way ahead of you...judging by the rate of progress, the entire crew working on Sydney’s new light rail are quadriplegics.
Re: (Score:2)
Good news. The California 'High Speed Rail' build crew has just become available.
They should put your schedule 'right'. You'll have Sydney light rail connecting Alice Springs with the nearest empty crossroads in only 10 years...depending on the breaks.
No manual review? (Score:3)
Automated fraud detection is fine.
But acting on a number which is basically a statistical guess, without some follow-up manual review is plain irresponsible.
The credit industry does this and regulation differs by country. For public services, the justice standards should be different.
I don't think we are hearing the whole story though. It is hard to believe that a 20% error rate is accepted by European standards of fairness.
Re:No manual review? (Score:4, Informative)
Yeah.. I've built a few of these kinds of systems. Fraud detection is a trigger / alert NOT an automatic action unless you have certainty. (Like mathematical 100% accuracy). Can be tiered:
1) Fraud Detected: "If" you have some clear "Yes this is for sure fraud" metrics then this tier can result in automatic action. (That can be a Big "if")
2) Fraud Almost Certain: Trigger an investigation and depending on your definition of "Almost" maybe some "light" action (Protect from further harm if possible)
3) Fraud Possibility: Depending on the order of magnitude and your staff resources this could also be "Trigger an investigation" but if the number is too large more like "flag as suspect" -> repeated indicators elevate this to #2 and maybe the fancy version lower level investigation like the support equivalent of an off-shore call center.
Sounds like they are treating everyone as a #1 and with a success rate of 80% they are SOOOO far from that being acceptable. Note most banks have a version of #2 for bank activity.. Pretty sure most of y'all have gotten the "We've detected questionable activity on your account.. please verify these transactions" ... dealing with Gov't benefits that's a different situation entirely but if these mails being sent out were very gentle like "The Department has questions regarding the status of your benefits call this number -> XXX-XXX-XXXX" --> That call center I was talking about that could be acceptable.
Re: (Score:2)
Often the fact that the computer says something primes the person reviewing the situation to believe it, creating a bias. In fact even just filtering out all the ones that are likely to be fraudulent can create bias, as the perceived amount off fraud goes from 4% to nearly 75%. Suddenly fraud isn't rare any more, it's the norm.
Re: (Score:2)
Re: (Score:2)
Australia's welfare system sent automated letters to everyone they detected as being over paid. Cutting them off, or at least setting up a way to claw back the overpayment.
Problem was of course that the data wasn't perfect, and the system double counted some people's salary information going back 5 years.
Re: (Score:2)
The story is meaningless without knowing what is the ramification of those false positives. What happens, is it just a letter, can the person seek compensation for unfair psychological stress, do they need a lawyer, how about compensation for wasting their time. What happens, obviously 20% false positives is entirely unacceptable and if it generates 20% false positives, how many does it miss. Is it all a typical scam, the automation contract worth exorbitant amounts and how much profit it generates is the s
the real story (Score:3)
The real story is how a 20% Error Rate is even close to acceptable in any automated system, especially in an environment where addressing issues through bureaucracy is practically impossible?!?! On top of that every run will yield a new chance to be falsely accused so the threat always looms. Seeing this in my own USA is a much deeper fear for me than any terrorist activity, and 40% of the public will welcome it, including people who will be ruined by it. It's insane.
This is part of the thick edge of the wedge. Beware all.
Re:the real story (Score:4, Insightful)
So they analyzed 1,000,000 claims, and detected 40,000 of those were fraudulent. Of which 8,000 were "incorrectly" marked as fraudulent. That sounds like a pretty decent first run to me. Unless you want to manually look through 1,000,000 claims, or pay a team to look them all over. This just reduced the cost to identify fraudulent claims by 25x.
Not perfect, no. But having a team analyze the 40,000 claims it kicked out is a heck of a lot cheaper than analyzing the 1,000,000 that went in.
Re: (Score:2)
Been a while since I did stats, but I think you're right.
Tossing a coin will give plenty of both.
Re: (Score:2)
Problem is that 8,000 people were royally screwed by this. People literally starving, having to rely on charity hand-outs to feed themselves and their children because the system made a mistake. That's pretty callous.
Best thing now would be for all 8,000 people to submit GDPR review requests. The GDPR allows for decisions made by machines to be reviewed by a human on request.
Re: (Score:2)
The GDPR allows for decisions made by machines to be reviewed by a human on request.
That's limited to decisions "which produces legal effects concerning him or her or similarly significantly affects him or her." [gdpr-info.eu].
Here, there are not yet any actual effects. At this juncture they're just receiving initial letters saying that the council suspects they're falsely claiming a tax credit, and have an opportunity to show evidence to the contrary -- evidence which a human then would have to review. I highly doubt that would fall under the above definition.
Re: (Score:3)
8,000 people get letters - the horror!
Needs-based benefits deserve to be challenged periodically, and recipients should be able to defend their claims at any time.
Re: (Score:2)
Disability fraudsters are stealing from the disabled. They ARE SCUM.
The USA has been gaining 1 million additional 'disabled' every year since welfare reform passed. The whole system is routinely gamed. Lawyers make careers out of getting disability for the able bodied One million more per year, while the big demo bump (baby boom) ages out of SS disability and onto regular SS.
Re: (Score:2)
Problem is that 8,000 people were royally screwed by this.
They were royally screwed by getting a letter? I can only assume you jumped to conclusions, and didn't read the article before you jumped on the sky is falling bandwagon. No one is literally (or figuratively) starving because they received a letter. They have up to ONE YEAR to respond to the letter before anything happens.
But yes, filing a GDPR review request would likely be a good idea.
it depends on how the follow up is handled (Score:1)
Re: (Score:2)
How would you feel if your government accused you of fraud in an official letter ?
Like an American citizen. Last year 0.86% of all tax returns resulted in an audit.
This new system incorrectly flagged 0.80%. And they got a letter. I'm fairly certain that responding to the letter is both easier and faster than getting audited by the IRS in the US.
Re: (Score:2)
I've always been bothered that all police in the US carry guns - where in general the police in the UK don't, at lest in the England of my youth.
I've been bothered with the idea that the police in the UK don't carry guns. There's a near complete ban on the private ownership of guns in the UK but does that stop the criminals from being armed? I seem to recall a few instances of people using knives and vehicles as weapons. Sure would be comforting to know that at least the police were able to shoot the assailant sooner rather than later, and minimize the death and destruction.
Getting back on the topic of being tracked by automation. If you want th
benefits cliffs just make people work less or not (Score:2)
benefits cliffs just make people work less or not work at all.
Why work MC's part time as if you end up working 22 hours one week vs say just 20 you lose benefits that take a long time to get back.
Developer here (Score:2)
I thought they said frog detection [youtube.com].
So what is acceptable error rate? (Score:2)
Critics say the 20% error rate is unacceptable as around 8,000 people will receive letters wrongly accusing them of fraud.
And if you say Zero then you have to prove to me that the prior system had zero errors.
Demanding perfection in replacing a flawed system is ridiculous.
I guarantee there is a appeals process after detection - people will not be thrown out of their homes instantly.