UK government using AI tools to check up on roadworthy testing centers The UK's Department for Science, Innovation and Technology (DSIT) has produced a list showing how the country uses AI technologies to perform tasks ranging from speeding up the planning process to prioritizing the inspection of MOT testing centers. Most vehicles more than three years-old are assessed annually by an MOT ( …
Hey [ Human | Ai ], can you make this sentence more readable?
"Most vehicles more than three years-old are assessed annually by an MOT (Ministry of Transport) test road worthiness, as required by local law."
Looks like editor/proofreader jobs are safe for the time being which ever source created that one.
Most States do not. Even California just does a "smog test" for emissions, but the basic vehicle safety features (steering, brakes) are not checked. Watch the many US-based automotive repair YT channels for the kinds of rotted-out deathtraps that are on the road.
https://en.wikipedia.org/wiki/Vehicle_inspection_in_the_United_States
The article (and probably the press release it was based on) is missing some key information. i.e. what metrics are being used to decide which MOT test centres need inspection... I mean, it's probably just an improbably high pass rate for older vehicles, in which case a human being could do the job with a careful database query without spending any money on AI.
I suspect that in this case the "AI solution" will be a database query as you suggest. Possibly one that is already being done under a different name.
Don't forget that "We implemented an AI solution" sounds much better than "We ran a database query", and scores more points on the buzzword bingo card.
If, however, the powers that be bring in one of the consultancy friends, then we can expect to see a multi-billion plan wrapped around the same query.
Yeah, I suspect this is going to end up being a set of rules which are either based on the number of passes or breakdowns since last MOT.
Finding out what those rules are is what the current generation of AIs is quite good for. It might be something like having too many tests done on one day which flags up an investigation but without a lot of testing data from good and bad test centers it would be difficult to identify the conditions.
I thought the MOT system is already rate limited. A testing station cannot submit MOT pass info faster than the time it is believed is needed to carry out a proper MOT. I've seen this happen. The MOT happens a bit quicker than expected, but I've had to wait because the system wouldn't let the garage log the details more quickly.
Of course, some places will game this, doing a 5 minute "MOT" for cash and then just making the customer wait until for the timeout.
The bay used for an MOT test is quite specialised so the rate will be based on number of suitable bays and however long it should take to properly conduct the test.
But none of this is new. Testers have always been tracked on how many vehicles they pass and fail and too high a pass rate has always been flagged up. As usual the sticking point won't be how effectively you can detect the fraudsters, it will be how big your enforcement team is.
Or you could just ask around in the pub. Won't take long to get directed to a dodgy tester. I'm quite happy to volunteer for this role....
Yeah, that was precisely what I was looking for, but wasn't expecting.
There are options: Collate the totals for all the pass, fails, and advisories for X MOT stations in geographical area Y, divide for the mean, and visit the stations Z points outwith the mean. That can be done with a pencil and paper, or the basic spreadsheet skills of a schoolchild.
Alternatively, they could engage an indepth analytical study of varying metrics and potentialities, produce the modelling, testing... and well, you get the idea.
The cynic in me says it will be dispense with the high school graduate on work experience, throw £££££'s at AI to do the simplest of queries and call it a day. I doubt there will be anything more complicated than that.
I could think of some analytical use for a collated database of MOT test results - not least trends for vehicle failures & parts, and possibly, just possibly a use for AI to trawl the data and extrapolate, but that's just running a query on already available data, and I kinda thought that was a solved problem with existing tools.
I imagine the MOT database is already being scraped and crunched by the motor industry to *ahem leverage pricing rather than for QA. If you know that suspension arm bushes on VW's have a high failure rate, then that's high parts turnover and so you can weld a few extra quid on them for ££££'s - huzzah!
"Or we could improve the design?".... "Get out!"
Similarly for Gov emissions regulations- Here is the mean for MOT passes, here is the max emission currently allowed for class, reduce the max to just under the mean and weld on the Tax! - huzzah!
Urban planning perhaps? This area has a high failure rate for vehicles indicating a lower socio-economic class -redraw the electoral boundaries, and divert spending to the affluent area. Get the poor off the road and on the buses and trains... and while you're at it, cancel the buses and trains. "It might just be bad roads, perhaps we should resurface them?"... "Get out!"
tl;dr You're not failing enough vehicles for your area, those a rookie numbers, weld on some more.
Test results are generally too vague to be much use to manufacturers. "Left rear tyre worn below legal limit". Doesn't say who manufactured the tyre or how long its been on the car.
In fact most advisories are so vague as to be useless, "Subframe corroded but not seriously weakened". What do I do with that information? Change it out immediately? Cross my fingers for another 12 months?
Regardless of what is used; query or AI, does this mean there is an enforced failure rate?
Some poor person with a perfectly good but old car will get failed so the test centre hits its reverse KPI? Or if the test centre is really stupid they'll fail a brand new car - "but I only came for an oil service".
A surprising number of cars fail their first test because their owners think they are still brand new and hence don't require any form of maintenance. Especially if they are on lease (most are) then they resent maintaining something they don't own. Consumables like tyres, bulbs, brake pads. On the other hand my 12 year old Honda sails through every time because I give it a quick once over before booking it in.
Of the 3 cars I've owned over the last 8 years all 3 have previously failed MOT tests for worn tyres or worn wipers (before I owned them I hasten to add). They then pass the same day after the garage has repaired the fault.
My guess is that the most common failure on the books will be tyres, followed by brake pads, followed by wipers, then brake light bulbs, then headlight bulbs.
Mechanics don't seem to want to/can't due to the rules just replace these consumables and pass the car, they have to log a fail then re-test once fixed.
"Mechanics don't seem to want to/can't due to the rules just replace these consumables and pass the car, they have to log a fail then re-test once fixed."
I asked a friend about this a few years ago when he worked in a testing centre. Apparently the regulations state that the MOT is for the state of the car "As at the time of the test" so any failures are logged even if it is a wiper or a bulb, it also allows the most common failures to be tabulated. He did say that the re-test for things like that are generally just a check box of "Thing that was broken has been replaced/fixed/"
This is where I get confused with the use of AI or as it should be known ML and I've given this a little thought.
We know what an MOT is and the output of an MOT is one of three things. Pass, Fail or advisory. They do record other information such as milage and emissions data etc...
The input to this test is of course the car. However, the car is completely random as are the results. Most people take their car to the same garage every year but sometimes you'll have to use a different one. Personally I've had a few occasions when my car is in for something else and I ask they give it quick look as my MOT is due soon and I'll get a couple of the jobs done then like if they tell me a wheel is nearing it's end of life I'll get replaced or brake pads or whatever. That means that if the logic is at x number of miles you are expecting job y to be done it doesn't work because it's already been done and won't be recorded on the MOT. It's also based on luck. As one garage might get all passes for a couple or months and another get all fails.
The question here is how are they going to train it and what are they going to train it on to mark a garage for inspection. Personally I think the all garages get inspected method is the best method for this but I know from experience that will be purely a cost saving exercise and they will half the inspections they do which ultimately puts people lives at risk or am I missing something here?
I would expect that DVSA (the agency who inspect the MoT testers) have a very good idea of what they're looking for in the data. Looking at the data isn't about whether a given car is a fail or not, it's about patterns in the data, and whether that then triggers targeted inspections. Obvious examples would be a garage where a much higher proportion than normal of older, high mileage cars being given a pass with no advisories, or more newer lower mileage cars are getting failed, and the like.
But this is yet more government AI window dressing. The big problem with the MoT is that the statutory fee has been fixed at £54.85 for 15 years. Allowing for a test bay costing £30-40k, the other garage overheads, and tester time, where's the margin? It's not really possible to do a good quality test for that fee, so it has become more and more cursory. You might get lucky if you are on very good terms with the tester, but its easy to understand why the quality of MoT testing has fallen.
"I would expect that DVSA (the agency who inspect the MoT testers) have a very good idea of what they're looking for in the data"
My last job was working for a company that made devices that each and every MOT testing station in the UK is <U>required</U> to have by the DVSA. Some time ago, along with several other items of tests lane equipment, ours was required to interface with the new 'connected' way-of-working, meaning that the test results were required to be passed to DVSA systems automagically[1] and without human input, or as little as possible. It was my job to ensure the migration from 'human input' data to 'highly automated' data went as smoothly as possible, from a business (Employer's and DVSAs POV) and the poor VTS ('Vehicle Testing Station') staff (My POV).
They did indeed have a very good idea about the 'data' they held, in terms of what information they could reliably extract, and - crucially - what their 'data quality' metrics showed. (You'll note I don't say anything positive or negative re the data quality, just that they were aware of their DQ 'scores' wrt their metrics).
* [1] The 90s called and want their terminology back
"of marginal benefit"
From the point of view of mechanical safety, that's always been the case - an MOT test 'pass' is just a statement showing that the vehicle passed on *that day only* and says nothing about the following 364 days.
From the POV of data available to the public, it is now possible to see dates, test numbers, failure reasons and and 'advisories' noted on line... very handy if you are considering buying a secondhand car!
Time taken for a test.
When the tester starts the test, the time is logged. Then when the test is concluded (sans entering all the bits the car failed on), the time is logged. It already flags up to the authorities on how long a test took to perform.
So it might be that a test centre tends to take a short amount of time to do the tests on average. While the cars are random, the time it takes to diagnose a duff CV joint (as an example) isn't. So it might be that a Honda Civic (as an example) is taken to 2 garages. One does the test in 60 minutes, the other in 45 minutes. If it goes to a 3rd garage that takes 55 minutes, the 45 minute garage could be an outlier and it might indicate that the tester may not have performed the test properly, or even seen the car. The latter does happen. Absolute death traps of a car give clean MOTs are on the road and they've never seen a test centre.
They could well be looking at car history too. A car that had a clean pass 12 months ago could then present to a different testing station and be an absolute write off of a car.
The MOT is not perfect /I've come across 2 vehicles of people I know where there has been a defect on a brake disc - warping in one, a chunk missing in the other. In both cases, passed. Even though the defects would be indicated during the brake efficiency test where the reading on the gauge will change/pulse.
Apparently, that is not a failure, or even an advisory. On the road, this would be felt by the vibration during braking.
The chunk missing off a rear brake disc was on a Mercedes, and that vehicle went through 2 years of MOT at a Mercedes main dealer without getting picked up.
Eventually, after the next time I was in that car, got the owner to take it to a local independent mechanic who replaced the defective disc
The MOT Test operates primarily as a test of function rather than condition. If the component or system being tested functions within the parameters layed down in the MOT Testing Manual, it will pass unless it is assessed as being in a seriously enough deteriorated or damaged condition as to be dangerous. If the tester notes that the part that has passed the test has nonetheless some minor damage, or is becoming worn (eg brake pads becoming worn), he may add a 'Minor' (formerly an advisory) to that effect to the the test comments on the certificate, but this is principally at the discretion of the tester. In this case the specific limits for that wear before the pad is considered defective and in danger of failing to function correctly is set at 1.5mm minimum pad thickness or worn down to the wear indicator, and such a pad will result in a 'Major - dangerous' as a reason for rejection.
I can't really see how supposed 'AI' is going to achieve much if anything that the DVSA inspectors are not already seeing themselves from the test data that they recieve, and I suspect they are already using queries to the database to look for anomalous results.
Since the MOT Test data logging went completely online a few years ago, there is less scope for cheating the test than there used to be, although I do know there are still 'dodgy' MOTs available from equally dodgy people, it is less easy to issue a false pass certificate now that a vehicle has to be logged in to the system and the data electronically recorded during the course of the test. It is also (because of this) no longer possible to take a vehicle which has failed a test to another testing station and have it pass, as the earlier fail will immediately show up to the tester as soon as he logs it onto the system.
I'm not sure you couldn't just go to a different station though - there's still an element of discretion involved. I've seen two testers at the same station have different opinions on the same component, in which case it would be down to who tested it whether it was pass or fail.
A friend of mine works at a test station, and there is quite a list of things they aren't allowed to do when testing - like they aren't allowed to poke and prod at corrosion to see whether it's surface or all the way through, for example.
This article is written from a GB, not a UK perspective - in Northern Ireland (part of UK but not GB) MOT Test Centres are run by the government themselves, the DVA (Driver and Vehicle Agency), not by private garages. There are 17 DVA-run test centres that cover Northern Ireland ("https://www.nidirect.gov.uk/articles/driving-and-vehicle-test-centres").
"Most vehicles more than three years-old are assessed annually by an MOT (Ministry of Transport) test road worthiness, as required by local law."
In Northern Ireland an MOT for a car is required once the car is four years old, not three years old as in GB.
MOT delays have been a mess in Northern Ireland since 2019 due to stress fractures discovered in vehicle lifts, arguments between the DVA and the lifts supplier about these faults (Supplier: "you didn't maintain the lifts correctly", DVA: "Well *you* maintained them for us under contract"), the closure of test centres during part of the COVID lockdown, staff strikes, etc. Several times in recent years cars in NI due MOTs were issued with TECs (Temporary Exemption Certificates) as a way of coping with the extended delays in test appointments: https://www.infrastructure-ni.gov.uk/news/minister-announces-way-forward-reduce-mot-waiting-times
When I tried to book my most recent MOT approx 5 weeks before it expired the earliest date available was 2 months *after* the MOT was due to expire (which was be "ok" as government has agreed with Police and Insurance companies to "permit" this). However that earliest MOT appointment would have also been one month *after* my vehicle's Road Tax was due to expire (and automatic renewal via Direct Debit would not occur due lack of MOT) and so I had to use the DVA's service of phoning them 10 days before Road Tax was due for them to find a last minute MOT appointment (i.e. someone elses cancellation).
There are still significant delays in booking appointments: https://www.belfastlive.co.uk/news/northern-ireland/dva-update-northern-ireland-mot-30211613
Due to these ongoing delays the Department for Infrastructure in NI are consulting on switching MOTs from being annual to being bi-annual (once they start): https://www.infrastructure-ni.gov.uk/news/biennial-mot-testing-consultation-begins
The are potential option with the approach that are positive.
Remember the garages that do MOT’s all sit somewhere on the Goldilocks scale, as do the punters that take their cars in and the cars themselves.
Daddy bear garages will fail more cars, but that doesn’t mean to say they are better at an MOT than anyone else, quite frankly they may be ripping off their customers.
Baby bear garages will pass nearly everything, that doesn’t mean they are allowing unsafe cars on the road, it may be that the cars they examine are maintained better during their service cycle.
Where cars move between garages you will begin to see patterns, looking at the difference between the number and type of “advisories” you will see patterns, How you scale these patterns remains to be seen, but if somebody could dip into vehicle service details (even the tiny garages now use online recording systems, which also help them with billing and accounts) you may suddenly be onto a really good thing.
Obviously this won’t all happen on day one. Worryingly somebody may decide they are happy with a partial view and accept the road accident because of faulty vehicle statistics as being cost effective, but we can but hope that the approach provides better visibility than the existing MOT of MOT stations that seems to have its own Goldilocks problems.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
