At the end of the article, they discuss how the report stresses a need for more secure transmission systems on the CAN bus, as well as potentially including things like permissions-based security
It's just another biometric matching system. So it might be used as a form of sign-in to your car's network or data, better than taking your hands off the wheel to key a password. Or a LoJack complement that identifies your car's thief.
That doesn't make sense to me. The constraint is 15 minutes total. If drives vs drivers give different data, that could only imply that (a) each driver's 15 minutes spans multiple vehicles, or (b) they make conflicting conclusions for the same driver across multiple trips.
You're reading constraints into the experiment that don't appear to exist. My take is that after 15 minutes of driving the algorithm could correctly identify which driver was behind the wheel. That means: I get in, start driving, and within 15 minutes or so, the system analyzing the car's internal network will, with 90% specificity, say, "The driver was 'rosser."
I don't see where you're coming up with only 15 minutes total driving. That isn't implied anywhere in The Fine Article. But even leaving that detail aside, it would actually be more impressive than the reported finding if they were correctly identifying people that well after only 15 minutes total!
If each driver is run in twenty 15-minute (or longer) trials, you have 300 total data points to analyze. If the car was correct in its fingerprinting 270 times, then it was correct 90% of the time.
Yes, but the only way you get a number not divisible by 15 is if, for some number of given drivers, it must have been correct for some trips but incorrect for others. I would not consider those drivers to be "successfully identified". Remember, the goal of this is to isolate individual drivers.
> it must have been correct for some trips but incorrect for others.
No, it could have been inconclusive for the other 10% of the trips, simply stating, "Insufficient data for meaningful answer". So 90% of the time it said, "Oh, this must be DrScump" and the other 10% of the time it said, "I don't know who this is yet, let me keep analyzing the braking patterns"
EDIT: Which is different than the way I stated it in my initial comment. Your reply helped me realize that.
Yes, 10% of the time. The other 90% it did. So.. 90%. Not being snarky or passive aggressive, I just don't understand the issue you're having with that statistic.
FYI, those are different projects with vastly different quality requirements... But they do shoot themselves in the foot when they put buggy software right in front of their customer.
I don't believe the application is in surveillance or tracking. It's just useful to know who is driving out of say a few people so you can adjust vehicle parameters to them.
Modern cars will already learn about driving characteristics so they can better manage the car to match, but I don't know if they try to match individuals.
I think you're correct, re: adjusting parameters based upon the driver.
My truck has three "memory settings". I can adjust everything the way I want it -- different seat positions, mirrors, etc. -- and save those in "A". My girlfriend can then get everything the way she wants it (she's much shorter and smaller than me) and save her settings in "B". When I get in to drive after she has been driving, I just hit the "A" button to have everything adjusted back to exactly how I want it.
It's very handy when you have multiple people who drive a single vehicle.
Speaking of which, it's really maddening how the latest automatic transmissions try to learn about you, because, guess what -- that impedes ME learning about IT!
I rented a new Chrysler 200 a few months ago and just when I thought I had the throttle figured out to where I could trick the transmission into not upshifting when I wanted it not to, upshifting when I did want it to, etc... it stopped working. It was a bizarre experience, like some sort of arms race or adaptive-difficulty video game between it and me, with my goal to actually control the RPM and its goal being to prevent RPM from getting out of its preferred small range no matter how increasingly erratic-seeming my throttle foot behavior became.
As a result, I am unable to develop any intuition about how the engine will actually behave in response to my inputs, and in a split-second dangerous situation will be unable to do anything useful.
It's advertised as "helping fuel economy" but all it really is is "adaptively learning to ignore the operator's inputs" which might be nice for the first few thousand miles a brand new driver drives, but seems outright dangerous afterwards.
This is exactly why I only drive manuals when I can (not really ever a choice with rentals in the US) and it's getting worse with more gears in transmissions. With 4 speed torque converters, you'd generally mash in the gas on the freeway and it'd drop to 2nd from 4th w/ torque converter lock-up. With 6spd torque converter automatics, even the nicer faster-shifting ZF stuff, it'll drop to 4th first, then as you continue pushing the throttle in, to 3rd, then immediately try to go back to 5th or 6th as you've now backed off the throttle due to the car trying much harder than you wanted to accelerate.
I really don't know how people drive, let alone enjoy, cars with automatics these days. Paddle shift with dual-clutch autos might be loads better, but they don't generally have those on cheap rentals, so I haven't had much time to play with one.
I'm surprised it took 90 minutes; I'd imagine that it'd only take two or three. The data is very rich and continuous, and the knowledge is almost entirely reflexive. I would also think that identifying typists wouldn't take much longer. Does anybody know of any papers on that? You could pull that off with javascript.
I don't know of any papers, but for verified certificates from Coursera, you validate your identity via a webcam selfie and typing their one sentence honor code into an input field.
Talk about doing things the hard way. With 15 people, I'd bet you could determine the driver simply by the weight sensor in the driver's seat. No need to do any actual driving. You could probably make some progress with radio station settings (volume/channel etc) or even how they adjust the seats/mirrors.
Not all sensors are available all the time. I was creating a prototype dash cam that recorded drivers driving habits and AR of obstacles on the path in addition to GPS routing.
The best way to "log on" a driver was analyze driving pattern and assign records to that person
Weight is only one dimensional though, so you'll have more collisions. If two people happen to weigh about the same, then your weight sensor is useless and your driving analysis is not.
And the article points out that they can already distinguish drivers with 90% accuracy based on the throttle sensor alone after 15 minutes of sampling. With extra sensors or more time, they get much closer to 100%.
So yes, you(s) are all correct. With one sensor you get decent accuracy, with more you get improved accuracy.
Noting that these are forensic assessments, and thus useless for personalising car settings when a new driver gets into the vehicle.
Was an Ex Ebay'r founded startup tried to recruit me. The mouse movement was part of their security product. Not sure what else they were doing or what happened to them.
I suppose it would depend on the driver. I know a few people I could fingerprint with the lousy IMU built into my ears with no computer analysis - some go constantly from the white line to the yellow line and back, some don't realize that the accelerator works at input levels below 10%, some second-guess the automatic transmission at every shift...but many drive, well, normally. Regarding the mouse or keyboard inputs discussed in this thread, there are a few hunt-and-peckers or over-correctors who would similarly be found. I don't think it's that impressive to get a few drivers out of a sample, they likely have detectable quirks.
The privacy concern is valid. It won't be long before all new cars are sending data from the sensors to third parties, and it would be unthinkable for the manufacturers to ask for your consent before collecting that data since they'll make money off of it.
The article mentions concerns with insurance companies. One I worked for tried to get customers to agree to having their rates adjusted based on data from their cars. It was mostly to do with how many miles the customer drove, but I think there was also an element that looked at speed coupled with GPS data to determine if they habitually exceeded the speed limit. There was a lot of pushback from customers and privacy advocates at the time and it didn't really go anywhere at the time (2007-08 iirc) but since that time I recall hearing of another company doing just that with some volunteers.
Now imagine the insurer has access to everything on the CAN bus. Your rates could go up based on taking curves too fast, braking too hard too often, or following too close.
The law enforcement angle is disturbing too.
Then there's the nightmare of the advertising people getting hold of it, complete with GPS data. God knows what they'd do with it exactly, but whatever it is, it will be sleazy and annoying by definition.
Time to include this on the list of things to oppose for privacy's sake.
>taking curves too fast, braking too hard too often
I was thinking about this the other day, people driving in San Francisco or Seattle hills would then have higher rates than people driving on flat Texas roads? People stuck in rush hours (where no matter how smoothly you try to drive, there will be more hard brakes) would have higher rates than people who can avoid it?
regardless of the number of car accidents in the area?
Obviously, you'd want to use some more (and more useful) features to calculate your price. (Otherwise, your insurance company will lose out to the competition that prices better. You overcharge low-risk drivers who will leave to a cheaper alternative for them, and are left with the high-risk drivers who cost you a lot.)
The area you are driving in is an obvious candidate. Or even the particular roads you are driving on at particular hours.
I can't see how this is a bad thing at all. Currently rates are set on arbitrary things that merely correlate with driving skill, like past record, age, and gender, etc.
I'd love to be rewarded for actually being a good driver. Rather than being lumped in with bad drivers who have a high chance of being in $50k accidents, which everyone has to collectively pay for.
Driving is also the leading cause of death of young people, so anything that encourages people to drive safer is a huge net good for society. What's more effective, a sticker that says "buckle up" or "drive slower", or knowing that your insurance company is keeping track of it?
Of course it should be voluntary and opt out. But there could be a huge gain for people who opt in.
>I'd love to be rewarded for actually being a good driver. Rather than being lumped in with bad drivers who have a high chance of being in $50k accidents, which everyone has to collectively pay for.
The whole point of insurance is not to lump you into good driver / bad driver categories. If you make bad drivers consistently pay for accidents, then you might as well not have insurance at all, and you'll pay perfectly in direct proportion to the accidents you are involved in.
Insurance adds fuzziness to the otherwise "perfectly rational" outcome because people that do not behave rationally with respect to accident avoidance and financial planning will be at a severe survival and status disadvantage to those who do. It's like putting safeties on guns to prevent careless people from shooting people or getting shot. It makes the world more forgiving of individual mistakes at public cost.
The whole point of insurance is to make you pay for the risks you take, and not for when you get lucky/unlucky.
That we often lump people together is just one implementation of insurance. But not a necessary feature. Otherwise, there would be no insurance for one-off events---but that happens all the time.
There's nothing inherently 'social' about insurance.
The entire point of insurance is to pay based on risk. Someone with a bad driving record should, and does pay much more for insurance than a normal person. An old person should and does pay more for health or life insurance. A wooden house costs more to get fire insured than a brick building.
The point of insurance is not to let bad drivers get away with their bad driving. It's to make sure you aren't punished for bad luck. But if we can exactly predict the probability you will get in an accident, there is no luck involved, you are just a bad driver.
I habitually exceed the speed limit, drive very aggressively, take turns very quickly, regularly drive on canyon roads pushing my vehicle close to the limits, and have never caused an accident. What should my insurance rate be?
It's currently very low as I'm a safe driver according to the records (no accidents, no points, no tickets in the past few years).
If the insurance company tracked my driving, I'd be paying much much more, even though I'm not actually costing them any more, while "safe" drivers (who regularly impede traffic and push others to make more risky maneuvers; consider the Prius in the left lane on 101 going 5 under) would get cheaper rates.
This would also kill the sports car industry. A very small number of sports car buyers ever drive their car on the track. Most have fun away from cops, on the streets. If you knew that you couldn't exceed the speed limit, corner with more than 0.2g of force, etc without getting a ticket in the mail or otherwise penalized, it would surely keep people in compliance for the most part, but utterly destroy any demand for vehicles capable of exceeding those limits.
I don't doubt that these types of changes are coming, which is why I choose to have fun now. Back in my day, they used to let us actually pilot our own vehicles!
Or what about if the people who take corners.faster make LESS claims?
PS: not trying to defend aggressive driving, I'm not an agresive driver, just saying it'd be interesting if a lot of our assumptions were thrown out the window with that data.
That depends on what he was trying to do with that comment.
I believe that he's trying to make a distinction between the number of claims filed by people who take corners faster and the size of the claims made by such people when they do file a claim.
Generally, "fewer" is used when talking about things that come in discrete amounts, and "less" when talking about continuous things. For example, if you were comparing the amount of water in two glasses of water you would say that one has less water than the other, but if you were comparing the number of ice cubes in the two glasses you would say one has fewer ice cubes than the other.
However, "less" is often used even for discrete things when there is a unit attach, such as with money. E.g., we might reasonably say either "the waller contained less than $5" or the "the wallet contained less than $10".
Before the late 18th century "less" was used with discrete things. As with many things in grammar, some prescriptivist decided that people were doing it wrong and made a new rule, and we got stuck with it. And like other such bogus rules (such as not splitting infinitives and not ending sentences with prepositions) current usage is tending away from following the rule.
The main argument I've seen that we need different words for comparing discrete things and comparing continuous things is that it is less ambiguous when it would not be clear what word the comparison word applies to. For instance suppose there were two studies of possible cancer treatments. Study A produced 2 promising treatments and study B produced 4 promising treatments. All of the treatments from study A look like they will be better than all of the treatments from study B.
Using the prescriptivist take on "less" and "fewer" we could say that study A has "fewer promising results" (we are comparing the number of results) but study B has "less promising results" (we are comparing the quality of the results).
Anyway, I think that is the distinction he was trying to make.
I strongly suspect that most fast acceleration, fast cruising, fast cornering, etc. is undertaken 1) in cars that are designed to handle it, 2) by drivers who are 100% in the moment focused on enjoying the act of driving.
I expect these behaviors to be at least an order of magnitude less dangerous than
- having slept less than 8 hours
- texting
- eating fast food
- having a conversation with passengers or on the phone
- interacting with small children in any way
- driving a car which is not up to the fun you're trying to have with it
- driving a car with serious damage
And yet of these, only texting is frowned upon.
I'm thinking of all the times my mother turned her head away from the road at 60mph to mediate a fight in the backseat and feeling significantly less bad about my cornering habits.
I scratched my head reading this thinking "Why would using the clutch change things" also made me think of two things, one being that yes i believe that swapping your primary foot your using would most likely cause different driving patters much like swapping where your hands are sitting on the steering wheel would have a similar effect. And two is i wonder if there would be an improvement in recognition time if the car being tested had a manual transmission adding common shift patterns to the equation.
So... I'm expecting surveillance like driver fingerprinting + that shit Progressive is doing with "Snapshot" to be used to really penalize drivers in the future for the slightest mishap.
I hate driving - especially in California. Automate all the things.
70 comments
[ 2.9 ms ] story [ 121 ms ] threadHow would this enhance privacy? It seems like it would do the opposite?
With 90 minutes of data, they could pick out one driver out of a sample size of 15. How meaningful is that, really?
So, the claim is 90% accuracy in determining which one of the fifteen drivers they are looking at. So, they were correct on 13.5 drivers.
Um, what?
I don't see where you're coming up with only 15 minutes total driving. That isn't implied anywhere in The Fine Article. But even leaving that detail aside, it would actually be more impressive than the reported finding if they were correctly identifying people that well after only 15 minutes total!
If each driver is run in twenty 15-minute (or longer) trials, you have 300 total data points to analyze. If the car was correct in its fingerprinting 270 times, then it was correct 90% of the time.
No, it could have been inconclusive for the other 10% of the trips, simply stating, "Insufficient data for meaningful answer". So 90% of the time it said, "Oh, this must be DrScump" and the other 10% of the time it said, "I don't know who this is yet, let me keep analyzing the braking patterns"
EDIT: Which is different than the way I stated it in my initial comment. Your reply helped me realize that.
Honestly it seems like everyone reading these kinds of papers/articles does a ctrl-f "sample size" before anything else. Bit cynical.
It's been easy to do face identification for a very small population for a while. But scaling that to thousands of people has proven quite difficult.
https://www.theguardian.com/technology/2016/may/17/findface-...
and the HN discussion
https://news.ycombinator.com/item?id=11712197
Modern cars will already learn about driving characteristics so they can better manage the car to match, but I don't know if they try to match individuals.
My truck has three "memory settings". I can adjust everything the way I want it -- different seat positions, mirrors, etc. -- and save those in "A". My girlfriend can then get everything the way she wants it (she's much shorter and smaller than me) and save her settings in "B". When I get in to drive after she has been driving, I just hit the "A" button to have everything adjusted back to exactly how I want it.
It's very handy when you have multiple people who drive a single vehicle.
I rented a new Chrysler 200 a few months ago and just when I thought I had the throttle figured out to where I could trick the transmission into not upshifting when I wanted it not to, upshifting when I did want it to, etc... it stopped working. It was a bizarre experience, like some sort of arms race or adaptive-difficulty video game between it and me, with my goal to actually control the RPM and its goal being to prevent RPM from getting out of its preferred small range no matter how increasingly erratic-seeming my throttle foot behavior became.
As a result, I am unable to develop any intuition about how the engine will actually behave in response to my inputs, and in a split-second dangerous situation will be unable to do anything useful.
It's advertised as "helping fuel economy" but all it really is is "adaptively learning to ignore the operator's inputs" which might be nice for the first few thousand miles a brand new driver drives, but seems outright dangerous afterwards.
I really don't know how people drive, let alone enjoy, cars with automatics these days. Paddle shift with dual-clutch autos might be loads better, but they don't generally have those on cheap rentals, so I haven't had much time to play with one.
This has implications for "which family member is it?" or "you claimed you lent it to your friend, is that really true?" etc.
The best way to "log on" a driver was analyze driving pattern and assign records to that person
So yes, you(s) are all correct. With one sensor you get decent accuracy, with more you get improved accuracy.
Noting that these are forensic assessments, and thus useless for personalising car settings when a new driver gets into the vehicle.
The article mentions concerns with insurance companies. One I worked for tried to get customers to agree to having their rates adjusted based on data from their cars. It was mostly to do with how many miles the customer drove, but I think there was also an element that looked at speed coupled with GPS data to determine if they habitually exceeded the speed limit. There was a lot of pushback from customers and privacy advocates at the time and it didn't really go anywhere at the time (2007-08 iirc) but since that time I recall hearing of another company doing just that with some volunteers.
Now imagine the insurer has access to everything on the CAN bus. Your rates could go up based on taking curves too fast, braking too hard too often, or following too close.
The law enforcement angle is disturbing too.
Then there's the nightmare of the advertising people getting hold of it, complete with GPS data. God knows what they'd do with it exactly, but whatever it is, it will be sleazy and annoying by definition.
Time to include this on the list of things to oppose for privacy's sake.
I was thinking about this the other day, people driving in San Francisco or Seattle hills would then have higher rates than people driving on flat Texas roads? People stuck in rush hours (where no matter how smoothly you try to drive, there will be more hard brakes) would have higher rates than people who can avoid it? regardless of the number of car accidents in the area?
The area you are driving in is an obvious candidate. Or even the particular roads you are driving on at particular hours.
I'd love to be rewarded for actually being a good driver. Rather than being lumped in with bad drivers who have a high chance of being in $50k accidents, which everyone has to collectively pay for.
Driving is also the leading cause of death of young people, so anything that encourages people to drive safer is a huge net good for society. What's more effective, a sticker that says "buckle up" or "drive slower", or knowing that your insurance company is keeping track of it?
Of course it should be voluntary and opt out. But there could be a huge gain for people who opt in.
The whole point of insurance is not to lump you into good driver / bad driver categories. If you make bad drivers consistently pay for accidents, then you might as well not have insurance at all, and you'll pay perfectly in direct proportion to the accidents you are involved in.
Insurance adds fuzziness to the otherwise "perfectly rational" outcome because people that do not behave rationally with respect to accident avoidance and financial planning will be at a severe survival and status disadvantage to those who do. It's like putting safeties on guns to prevent careless people from shooting people or getting shot. It makes the world more forgiving of individual mistakes at public cost.
https://www.statefarm.com/insurance/auto/resources/high-risk...
That we often lump people together is just one implementation of insurance. But not a necessary feature. Otherwise, there would be no insurance for one-off events---but that happens all the time.
There's nothing inherently 'social' about insurance.
See eg http://www.slate.com/articles/sports/sports_nut/2015/03/warr...
The point of insurance is not to let bad drivers get away with their bad driving. It's to make sure you aren't punished for bad luck. But if we can exactly predict the probability you will get in an accident, there is no luck involved, you are just a bad driver.
It's currently very low as I'm a safe driver according to the records (no accidents, no points, no tickets in the past few years).
If the insurance company tracked my driving, I'd be paying much much more, even though I'm not actually costing them any more, while "safe" drivers (who regularly impede traffic and push others to make more risky maneuvers; consider the Prius in the left lane on 101 going 5 under) would get cheaper rates.
This would also kill the sports car industry. A very small number of sports car buyers ever drive their car on the track. Most have fun away from cops, on the streets. If you knew that you couldn't exceed the speed limit, corner with more than 0.2g of force, etc without getting a ticket in the mail or otherwise penalized, it would surely keep people in compliance for the most part, but utterly destroy any demand for vehicles capable of exceeding those limits.
I don't doubt that these types of changes are coming, which is why I choose to have fun now. Back in my day, they used to let us actually pilot our own vehicles!
PS: not trying to defend aggressive driving, I'm not an agresive driver, just saying it'd be interesting if a lot of our assumptions were thrown out the window with that data.
I believe that he's trying to make a distinction between the number of claims filed by people who take corners faster and the size of the claims made by such people when they do file a claim.
Generally, "fewer" is used when talking about things that come in discrete amounts, and "less" when talking about continuous things. For example, if you were comparing the amount of water in two glasses of water you would say that one has less water than the other, but if you were comparing the number of ice cubes in the two glasses you would say one has fewer ice cubes than the other.
However, "less" is often used even for discrete things when there is a unit attach, such as with money. E.g., we might reasonably say either "the waller contained less than $5" or the "the wallet contained less than $10".
Before the late 18th century "less" was used with discrete things. As with many things in grammar, some prescriptivist decided that people were doing it wrong and made a new rule, and we got stuck with it. And like other such bogus rules (such as not splitting infinitives and not ending sentences with prepositions) current usage is tending away from following the rule.
The main argument I've seen that we need different words for comparing discrete things and comparing continuous things is that it is less ambiguous when it would not be clear what word the comparison word applies to. For instance suppose there were two studies of possible cancer treatments. Study A produced 2 promising treatments and study B produced 4 promising treatments. All of the treatments from study A look like they will be better than all of the treatments from study B.
Using the prescriptivist take on "less" and "fewer" we could say that study A has "fewer promising results" (we are comparing the number of results) but study B has "less promising results" (we are comparing the quality of the results).
Anyway, I think that is the distinction he was trying to make.
I expect these behaviors to be at least an order of magnitude less dangerous than
- having slept less than 8 hours
- texting
- eating fast food
- having a conversation with passengers or on the phone
- interacting with small children in any way
- driving a car which is not up to the fun you're trying to have with it
- driving a car with serious damage
And yet of these, only texting is frowned upon.
I'm thinking of all the times my mother turned her head away from the road at 60mph to mediate a fight in the backseat and feeling significantly less bad about my cornering habits.
https://www.progressive.com/auto/snapshot/
(among others)
I hate driving - especially in California. Automate all the things.