Autonews reports that programmers working on driverless cars have found that having made them perfectly law-abiding, unable to commit traffic infractions, the result is that they have double the accident rate of driven cars, as they cannot cope with the anarchic driving of humans.
They obey the law all the time, as in, without exception. This may sound like the right way to program a robot to drive a car, but good luck trying to merge onto a chaotic, jam-packed highway with traffic flying along well above the speed limit. It tends not to work out well.
As the accidents have piled up — all minor scrape-ups for now — the arguments among programmers at places like Google Inc. and Carnegie Mellon University are heating up: Should they teach the cars how to commit infractions from time to time to stay out of trouble?
“It’s a constant debate inside our group,” said Raj Rajkumar, co-director of the General Motors-Carnegie Mellon Autonomous Driving Collaborative Research Lab in Pittsburgh. “And we have basically decided to stick to the speed limit. But when you go out and drive the speed limit on the highway, pretty much everybody on the road is just zipping past you. And I would be one of those people.”
Undaunted by collisions with reality, the diagnosis is that the problem is human driving.
Driverless vehicles have never been at fault, the study found: They’re usually hit from behind in slow-speed crashes by inattentive or aggressive humans unaccustomed to machine motorists that always follow the rules and proceed with caution.
Or might it be, as a libertarian might say, that government rules setting arbitrary speed limits create conditions where collisions are more likely?
Frankly, I’m surprised that they’re not employing “fuzzy logic” in their programming, which would permit the program to “learn” and adapt with experience, rather than the “rules-based” approach they seem to be using. It’s not like fuzzy logic hasn’t been around for years, and been employed in all sorts of devices from cameras to washing machines. Automobiles would seem to be a logical application.
This does sound rather like your average government program, particularly those of a socialist bent: brilliant until it encounters humans. And naturally, humans are the problem.
There are many who work in such fields who will happily argue for human driving to be made illegal once they have managed to persuade the legislators to permit self-driving cars. Personally I’m in favour of self-driving cars, for many reasons, but I believe the self-driving cars should be made capable of adjusting to real world conditions rather than the make the world adjust to self-driving cars.
Laird, might there not be a great fear of building into a driverless car a means of breaching the law by exceeding the speed limit?
I am thinking if the potential civil and criminal suits that might follow, in all the various jurisdictions in the US.
Ah think about the work for the lawyers. You have a self driving car with a licensed driver behind the wheel, photographed going too fast by a speed camera. If you allow the self-driving car to exceed the speed limit, then is the licensed driver responsible for allowing the car to go too fast? Was he/she actually driving at that time? Can the authorities use the information contained in the self-driving car’s memories, which are owned by the driver, against him (self incrimination)?
There are still some countries that allow one to drive as fast as is reasonably possible. The ones with speed limits do so to cut down on all the arguments – if you exceed the limit, you are guilty. Now will we have different standards for machines?
Oh, how many lawyers will get to purchase big houses and big boats and airplanes and lots of other expensive toys with the money we will be spending to sort it all out?
Speed limits are not arbitrary: the State owns the roads, and can set whatever speed limits it likes. If roads are privatized, then the private owners can likewise impose speed limits.
If the problem is that people do not respect speed limits, then the solution is simple: right now people are fined if they go X mph above the speed limit; then raise the speed limit by X and fine all people who go 1 inch/hour above the speed limit.
Tim: to me, it looks more like AI meets the real world; though, coming to think of it, there are analogies to socialism.
Laird: fuzzy logic is passé. Bayesian logic performs better in all the performance comparisons that i am aware of; though i do not follow this closely.
For some reason, I am reminded of my experience of landing at LAX for the first time, hiring a car, and having to merge into highway traffic 5 minutes later, with a big sign saying: DO NOT STOP.
The law is an ass, but a supremely powerful one. I’m not terribly surprised that this is a real concern for these companies.
I’ll be happy to share the road with driverless cars if they make the roads a little bit smarter. Smarter traffic lights would be a start.
The answer isn’t going to be in individual smart cars but with every vehicle being in communication with every other vehicle as well as the highways and the whole system integrating in real time. It has to be a co-operative system and then the speed limits will also be smart.
As long as governments control the roads, nothing about traffic will be smart.
I can’t imagine self-driving cars will ever be programmed to break traffic rules. Consumers will expect manufacturers to pay for tickets.
Got a ticket for going 93km in a 90km zone while driving through France this summer. Can’t wait to see this play out. BTW, how would a driverless car contest a ticket?
In Illinois, a driver can be ticketed for driving at the limit when surrounding traffic is all going faster. (I think under the general category of “obstructing traffic”.)
Kevin B @ December 18, 2015 at 10:38 pm:
That’s what we have now, with the communication being between drivers and limited to “body language”, signal lamps, honks, and occasional had signals.
Driving in traffic is in fact a social activity: it includes rudeness, courtesy, and rage.
When I got my licence in the UK you were allowed to exceed the limit for a short distance, three tenths of a mile, to safely overtake. Is that no longer the case? It seems like a simple solution to this problem.
“…government rules setting arbitrary speed limits create conditions where collisions are more likely?”
Speed limits are not arbitrary. There are obvious reasons for not driving 80 mph down a small road in a residential area. Perhaps higher or lower speed limits do render collisions more or less likely in certain areas, but that is a separate point.
Here in the south of Taiwan, most of the accidents I see (and I see them regularly) are, so far as I can tell, nothing to do with flaws in the rules of the road imposed by government, and everything to do with the poor driving education, stupidity and superstitions prevalent throughout the driving culture.
Got a ticket for going 93km in a 90km zone while driving through France this summer. Can’t wait to see this play out.
Assuming they catch up with you, it’ll be about 40 euros and no points. They take speeding in a 50kph zone more seriously, and also if you were over by a large margin. Quie sensible, French speeding rules.
I think it is worth pointing out that the merging-onto-the-motorway problem has nothing to do with the being-rear-ended problem. Also that the merging-onto-the-motorway problem is not solely about speed. I am sure we’ve all been in the situation where we have had to muscle our way into a lane because there hasn’t been a big enough gap between vehicles. This, of course, is because humans make such lousy drivers. The sooner the robots take over, the better.
This story does not appear to be about traffic rules or any kind of Azimovian law violation.
They’re usually hit from behind in slow-speed crashes by inattentive or aggressive humans
Wait. So the tail-gaters get caught out, the sh*ts that drive up on your boot/trunk, the intimidating idiots that want you to break the law to satisfy their own inhibitions?
As for this slow witted bumper jockey who is just as guilty, I would have happily paid for a Fiat Panda with City Break as my last car had I been able to also fit two adults in the back of it.
I for one welcome out new driverless car overlords, roll on technology, if you want to avoid accidents with driverless cars, then get a driverless car!
Brake, not Break (duh!)
No, it’s a Fiat, stay with Break.
As F.A. Hayek and M.J. Oakeshott were fond of pointing out – a lot of knowledge is tacit. People just learn how to do some things that it is difficult to put exactly into words.
And if it can not be put exactly into words (or numbers) it can not be written into a plan – or computer code. This does not mean that the humans are wrong for breaking “the rules” – the written instructions.
This is going to have to be understood in the future design of driverless cars.
When all cars are driverless there will be no problem with speeding cars at all.
OH, the State would love to have everybody in driverless cars that couldn’t break the laws, as the State would then start changing the laws to make disfavored people unable to use their driverless cars. A similar analogy is the way the State wants to put a mandatory kill switch into smartphones, ostensibly with the idea that if you report your phone stolen, the police can turn it off for you centrally. You just know the police would misuse such a thing to make people unable to use their phones to record the police doing illegal things in their operations.
Getting back to the cars automatically obeying speed limits, I’m reminded of 15 or so years ago when a lot of the international broadcasters were still on short-wave, and Radio Vlaanderen Internationaal out of Belgium would run a report in one of its weekly features about the people who wanted to make speeding impossible by putting transponders along the roads and speed governors in the cars that would respond to the signals from the transponders. I replied that I couldn’t wait for the hackers to break into the system and force cars to go no faster than stalling speed along all the routes to the North Sea resort towns, as well as force the cars to do 100 mph in school zones and mow down children.
Oh, and Alisa wins Quote of the Day.
Don’t see the problem. If the driver is at fault then give him/her a ticket. Drivers will soon learn to give them a wide berth.
On the twisty, narrow country lanes near me there have quite a few fatal crashes where nobody would have died had the drivers obeyed the ‘arbitrary’ speed limit.
One thing I’ve noticed is that the young men involved (and it nearly always is a young man) were all, according to the reports in the local press, wonderful, caring, bright people with a great future ahead.
It is odd that no thick, inconsiderate and selfish people kill themselves and others with their libertarian approach to speed limits.
Not sure about your theory of causation. It might have something to do with attempting to share the same piece of tarmac as another vehicle coming the other way, and not bothering to approach a corner with a view to thinking ‘‘Where do I go if there is something coming the other way?’. Such crashes can be fatal well below the speed limit.
As for speed limits being ‘arbitrary’. Of course they are. Why is there a ‘national speed limit’ in the UK? Why is it 60 mph, and not 63 mph?
The point that I was making is that the problem with driverless cars interacting with speeding dickheads is that the cars fail to account for dickheads. Remember that dickheads are always with us, some blog, some become politicians.
Firstly it sounds as though driverless cars are programmed to to react to what other road users should be doing rather than what they are doing. So what happens if they meet a horse, say, which is easily capable of exceeding the 30 mph limit especially if it has panicked, and indeed could be on any part of the road. Do driverless cars actually know how to drive past horses without frightening them?
What happens if one driverless car meets another with a program malfunction- inevitable if they become common.
Presumably they are already programmed to drive below the posted limit in slippery conditions or tight bends- else each and every one would stack up on an English rural road.
Having worked much of my life for a highway authority I can assure one and all that speed limits are extremely arbitrary.
The 30mph limit for built up areas was introduced in 1935 on no evidence (other than that only a few high end cars of 1935 could actually exceed that limit anyway).
The definition of a built up area as one with street lights every 200 yards was extremely arbitrary, several rural parishes erected street lights on country lanes just to impose a speed limit.
The 70mph motorway limit was introduced in the 1960’s after one meeting with no research (again I suspect that the fact that only a few high end cars could sustain a higher speed was the deciding factor).
The speed limit on a rural road is usually effectively set by a parish councillor with a bone to pick- he gets his speed limit in the hope of shutting him up. In urban areas 20 mph limits are similarly imposed.
In my neck of the woods many people stack themselves travelling 20mph below the speed limit- I seriously doubt that Rossi could get round some bends at 60mph.
Speed limits were introduced to deal with a problem of proof. PC copper says you were driving dangerously, you say you weren’t. Proof beyond reasonable doubt is difficult if not impossible without film of the offence. Proof of exceeding a speed limit is far easier, and in days of yore was used as a substitute for proof of dangerous driving.
I’ve found that on English roads physics often sets a much lower speed limit than the law does.
One thing we often forget about the purpose of speed limits is their intent to give everyone some certainty about how fast other cars are going. If I’m crossing a road with a 30 speed limit, I will do it differently than if the limit, and thus expected, speed of other cars is 70.
I’d be interested to learn about a wreck that happened because all drivers involved failed to violate the speed limit.
Well, Paul does it again. On Dec. 19 at 10:30 a.m. he wrote a penful, and this part of it really does need to be recognized as part of the Wisdom of the Ages:
Absolutely true. We work constantly (as private individuals, in our own heads, as well as in our professional capacities) to narrow the cracks between the areas of our known knowledge, if I may put it that way — so as to be able to think properly about two related but disjoint facts or situations or whatever, to learn to articulate the points they have in common, the relationships between them, so as to unite them into one piece of terrain navigable to the intellect.
But what we recognize and grasp are categories and patterns, and often we have some “feel for” or “intuition about” or “tacit knowledge of” how they are related; not just in our minds but in the real world.
I know how to use my fingers to press keys on this keyboard, but I could not begin to tell you how to go about tightening and relaxing which muscles when and how much in order to accomplish the act.
Er, Paul at 10:40, not 10:30. :>(
In “off-road” driving, the catch-phrase I learned was, “Drive to the conditions”.
Utterly unsurprisingly, this same concept applies to “proper” roads, be they freeways or country lanes and in all manner of weather.
This is VERY applicable with motorcycles.
Even then there can be nasty surprises, like ice-slick paint on zebra crossings, random oil patches, especially near traffic lights, and so on. The BIGGEST hazards are idiot engineers who, amongst other sins, build roads with inadequate or even NEGATIVE banking on curves and drivers who are utterly unaware of their surroundings.
Speaking of “engineers”, one thing I noticed when driving in the western US, was that the entry ramps on “expressways” are, generally, easily long enough to SAFELY reach traffic speed before merging, and the exit ramps allow drivers to decelerate to “street” speed on exit and to accommodate a reasonable number of vehicles doing so.
In Oz, on MOST “freeways”, the ramps are ludicrously short and “merge” / “split” lanes almost non-existent.
Paul wrote (and Julie echoed) “People just learn how to do some things that it is difficult to put exactly into words.” Precisely. And this is why typical Boolean logic and “rules-based” programming will not work well in driverless cars. Bayesian logic, while perhaps better, still fails in this setting because so many of the decisions needed to be made while driving are not easily susceptible to probabilistic analysis. Hence my call, way back at the top of this thread, for the application of “fuzzy” logic here, which is proven to work in many applications.
I second Bruce’s rant about “idiot engineers”. When my son told me he was planning to study engineering I said I’d pay for any type other than civil; if he wanted to go into that field he was on his own. (He ended up a math major anyway.)
I have always had a libertarian approach to speed limits and have so far racked up 40 years of driving with a 100% safety record. I suspect than many others here can say the same.
There have been a couple of times in my life where ‘speed limits be damned’. Medical emergencies come to mind. Saving someone’s life ( or my own ) means that speed limits will be ignored. What if my driverless car essentially causes or contributes to my or someone else’s death simply because there is no way to make the damned thing go faster than 30 MPH ?
The trial lawyers in this country will have a field day ( and get VERY rich ) from ‘accidents’ ( which lawyers do not believe in – SOMEONE MUST be responsible and pay $$$ ), deaths, mishaps, ANYTHING that can be blamed in any way on ‘driverless’ cars, and believe me when I say they will be unimaginably inventive when coming up with ‘reason(s)’ for the ‘driverless’ car to be at fault !
Paul, Julie and Laird are all correct in pointing out the present shortcoming in trying to build a machine that functions like the human brain. Currently we can’t and are not even closet ( A.I. is a MYTH – the human mind is not a sequential machine –
computers are !
Keep dreaming about driverless cars but until there is some fundamental change in the functioning method of computers ( base 2 ) and their programming, sorry, not me. Too many decades with these machines to EVER trust my life to one !
Paul Marks gets it right, and Julie is right that Paul is right. (Did i just quote Blazing Saddles??)
That is why learning algorithms are so fashionable in AI, i suppose: so that all the tacit knowledge can be learned from experience (as we humans do) instead of being coded. I am not entirely happy with this fashion for learning, i must say: i feel that learning cannot entirely substitute for thinking.
Laird and Jerry are also right to say that Paul and Julie are right. (More Blazing Saddles.) I beg to differ on their reasons why Paul and Julie are right, however.
Laird does not seem to realize that Bayesian theory is just a principled form of fuzzy logic: any other form is just handwaving; and btw it doesn’t work.
As for Jerry, it’s true that “we can’t and are not even close” to match the capabilities of the human brain, but surely he knows enough about computation to know that hardware is not the problem: software is the problem.
About merging into highway traffic: my first time in LA, i had specific problems:
just got off the plane;
driving a car i had never driven before;
only the 2nd time i rented a car with automatic transmission (never owned one);
never before having had to merge into such heavy traffic.
Having said all that, the discrepancy between speed limit and traffic speed would have been a major problem — if i felt duty bound to stick to the speed limit: iirc speed limit was 50 mph, traffic speed was 70 mph. There were large enough gaps between cars, but any driverless car sticking to the speed limit would have been a major hazard.
I also note that my experience in a car with an English friend and colleague, and also driving scooters on a Greek island with said colleague, leads me to suspect that the concept of “safety distance” is alien to the English mind.
I wonder how the driverless car would handle some of the crazier conditions here? I am thinking of the Macquarie Pass, just north of Kangaroo Valley. Along one stretch of road, the speed limit keeps altering continuously. Curvy road, 55kmph. 100m further on, 80kmph on the straight stretch. By the time you reach 80kmph, curves ahead 55kmph. That sort of thing.
@Pat: indeed.
I have often thought that Google’s robot cars will undoubtedly perform OK on American roads: straight, all bends are right-angles, all junctions have lights (not roundabouts), most drivers are fairly law-abiding.
By contrast, on British rural roads, there are no straights; corners and junctions can be of any conformation you care to name and frequently combinations of same; give-way rules cannot be deduced from the geography and signage may be seriously decayed; there will be random pedestrians, some of them drunk; there will be horses whose riders think they bought the road when they bought the horse, and cyclists ditto but with more attitude; there will be farmers driving strange machines with large pointy bits sticking out, at 10mph or less (but sometimes at 50mph), and there will be nice deep ditches either side of the road.
Now that’s going to be a test of someone’s programming skills. I can’t wait.
Paul is absolutely right and his statement about “tacit” knowledge goes way beyond driving. I have solved complicated mathematics without “thinking”. On Key West 9 years ago I hired a bike. I hadn’t rode a bike for about 15 years. I got on it and rode it without thinking. If I may be so bold we see this with the national curriculum and other drivel. The simple idea that if it ain’t taught formally it isn’t learned. So how did I learn computer programming? From reading books and magazines and just doing it.
Oh, and I started at Notingham University (which the NUS, at the time, lamented, lamented) was the most politically apathetic campus in the UK. What did I learn there… Physics and mathematics mainly. A bit of formal logic on the side. Sex, cannabis, women’s shoes, fencing, rowing (I once dated a cox*) and just stuff… Excellent university. At least when I was there (graduated in ’95) there was little or no politricks – not in the faculty of science. What the red-flying-fuck they got up to in the department of “Social Policy” I dunno. I didn’t want to know. Oddly enough the guy who taught me 1st year QM (Prof Challis) was instrumental in refurbishing George Green’s windmill in Sneiton, Nottingham. George Green. Green was almost entirely self-taught. His work is still on the curriculum in any maths, physics or related degree course. He was that brilliant. I didn’t do an A-level in maths but blagged my way into physics from biology. 2 1/2 weeks late and playing serious mathematical catch-up I managed through the toil and the kindness of (what were) relative strangers. So, a bit like Green, I know a bit about self-educating in Nottingham.
In the gent’s toilets in the science library at Nottingham was a graffito that read,”I’ve sucked Prof Challis’s dick – know why they call him Challis the Phallus”. The library is named for George Green. We had better graffiti in the Green than the arts library. The best I recall was, “I’d sleep with Gerry Adams but I’d be thinking of Martin McGuinness”.
Oddly enough whilst the science library is a fitting memorial to George Green the most prominent statue on campus is of DH Lawrence who hated the place (it is in his writings) and never even got a degree – he didn’t have the Latin so he got a “teaching certificate”. He eloped with the wife of a professor of German who he copped of with because she wanted him to bugger her.
*Said cox (a theology student – gave me my first heads-up on Islam – she was CofE) managed to wreck a Poly boat by cox-ing straight through it. It caused GBP20,000 damage and this was in 1994. We always won the Trent boat race anyway.
Humans cannot drive. We’re shit at it. In making the roads safe for our incompetence, and overconfidence, they’ve been made dangerous for people walking, or on bicycles. People have responded by retreating from risk, inside armoured boxes. The sooner cars drive themselves and human driving is consigned to history, the better.
“What was it like when people drove themselves grandad”
“Well my boy, everyone spent all their money on cars, and millions of people died every year”.
Yes millions. http://www.who.int/gho/road_safety/mortality/traffic_deaths_number/en/
“It can only be attributable to human error.” -HAL 9000
I would have put it that the vast majority of humans who drive badly choose not to do so, disregarding their fellow humans and the rules of the road, and basic common sense. That is how they are, and if they die or injure themselves, then that is effectively their choice in the face of environmental selection. However, it is the fate of the innocents caught up in the dickhead drivers’ mess that trouble me.
Mr Ed: you seem to assume, implicitly, that we are able to accurately assess our own individual driving skills. As a matter of fact, well over 50% of drivers think that they are better than average. I am right about myself, of course, but not everybody is.
Snorri, that old chestnut that statisticians trot out when asking people if they are above average drivers and most say ‘yes’ is not on point. The point is that many people choose to drive badly, their skills are not the issue. Even a bad driver (i.e. one not adept at taking corners and judging tyre adhesion), can drive safely with the will and learning the customs (not necessarily the rules) of the road, but many choose to be sloppy or simply arrogant lunatics. At a rough guess one motorist in five I come across seems to have an attitude issue, not a skill issue.
What makes people inherently bad drivers would be visual defects, impaired concentration, a lack of co-ordination or unbearable terror overwhelming reactions. Such people are few, but the rash lunatics are far too common. Quite why people are like that is another matter, but that is the world as we find it.
Discussing this in the pub last night it occurred to me that this is essentially an Asimovian problem; First and Third Laws vs Second. It could also be framed as a Second Law problem – could an 3 Laws compliant robot be ordered to break the law, given that the law constitutes orders from a country’s worth of people. I shall have to find a moment to scan synopses to see if he ever wrote one along those lines…
“Driverless vehicles have never been at fault … ”
Why does this make me think of something else?
“I am an H.A.L. 9000 series computer. No H.A.L. 9000 series has ever made a mistake. Our operational record is perfect …”
Yeah, those pesky humans…
Donning my flame-resistant suit, I’m going to be contrary on this point — just a little bit. While the above is literally true, it contains an implied premise that is not true: That the tacit knowledge of human drivers is the thing that has to be explicitly built into the algorithms. While that can be done (and is being done), it’s not the only tool in the engineer’s arsenal. Purely correlative, non-algorithmic modelling can be extremely powerful in mimicking human behaviour. If you doubt that, consider that this is exactly how Google Translate works — not with rules of grammar and syntax, but rather by finding the statistically most likely solution based on a huge (and growing) database. By this method, autonomous vehicles can “learn” not just from the behaviour of human drivers but also from each other.
You might object that Google Translate isn’t nearly as good as a human translator who knows both languages well, and I agree. But Google Translate is as good as a human translator who doesn’t know either language very well, and that’s a more fair comparison to the relationship between autonomous vehicles and human drivers. They don’t have to be perfect; they don’t have to even been very good; they just need to be, on balance, somewhat better than the average human driver, which is setting the bar very, very low.