Reg lecture asks what’s so scary about 1.5 tonnes of metal with a mind of its own? The technology to run driverless cars is ready. Whether humans are ready for driverless cars is quite another matter. Scientists and philosophers are scratching their heads working out who should be responsible when things go wrong. And you can guarantee someone is considering the future of the drive-time show when there are …
Well indeed. At least these things won't speed, turn left across my path (I'm a cyclist), pull out in front of me and so on. Of course there's also their inability to get pissed, get distracted by the telephone or any similar worthwhile cause (children, wasps, food, gorgeous bodies) in or out of the car.
"I'd trust the code more if it was open to peer review and not in the soul domain of a mechanic to install and update."
Yes, well that would depend on whether your mechanic's soul was heavenly or demonic, shurley?
Ahem. Aaand back to the article...
The UK’s Transport Research Laboratory? I hope this isn't the same organisation that tried to make leg protectors on motorcycles compulsory a few years ago, even though research indicated that they would reduce highly survivable lower leg injuries by 25% but increase far more lethal head and pelvic injuries by 50%.
I'll trust driverless cars the day a computer can understand why I am happy. Or sad. Or angry. Etc.
At first glance a somewhat flippant remark. At second glance it is actually a very good summary of the degree of "situational awareness" that AI will need to demonstrate for it to be better than human drivers.
Have an upvote.
I wonder if they will have optional personalities for your car?
Audi - Disables indicators and reacts furiously to being overtaken
Mini Cab - Ignores all other road users and pulls out in front of them repeatedly.
White Van - Uses the car sensors to detect females, slows down and beeps the horn repeatedly while passing.
Sunday Driver - Drives at least 10 mph below the speed limit and takes extra precautions at every junction doubling the gap a normal driver would need to pull out.
Boy Racer - Increases revving by 300% at junctions and if it detects another car in this mode has a race putting every other road user at risk, sat-nav can only be set to the back of supermarket car parks.
Idiot - Not programmed with any road rules. Unable to use with a valid insurance document, Tax or M.O.T.
I'd trust code over the meat-sack equivalent.
If us meatsacks were that bad then the accident statistics would be far worse than they are. A couple of scenarios spring to mind:
The ability of a car with a meatsack behind the wheel to stop if a pedestrian suddenly appears in the vehicle's path is limited by two factors; reaction time and braking time. Now an autonomous car might have a reaction time of zero* but it will still have braking time and thus braking distance, and if the braking distance is compromised by the pedestrian appearing within it then an accident will happen. Who gets the blame?
Secondly Mrs Commswonk and I live very close to a primary school (NOT a good idea, IMHO) and the nearby roads easily become congested at school start and finish times. The road width(s) will allow cars to travel in both directions if there is nothing parked, but a stationary vehicle requires "negotiating". Add numerous parked vehicles and the progress of traffic in each direction requires slalom action. It is not uncommon for everything to grind to a halt with two cars facing each other (or very close to it), unable to move forwards or backwards because of following vehicles. Will AI be able to find a way of untying the knot? Or will it have to resort to what the meatsacks do; simply sit there hooting at each other?
*As it happens I am of the view that a zero reaction time will have undesirable effects as well.
It does not take a lot of thinking to reveal a significant number of very common scenarios where AI is going to have to be extremely clever if it is to be better than human drivers.
I quite like the idea of the blame being put on the pedestrian in the case you highlight - it's clearly their own fault. Right of Way be damned, if you walk in front of a car like that you have to accept culpability.This, coming from a non-driver.
The other situation is far more interesting. Yes, this would be beyond any AI, if only because it's dealing with meat. But just imagine where all of these vehicles are networked and able to negotiate their way through?
But just imagine where all of these vehicles are networked and able to negotiate their way through?
Interesting concept. As a meatsack I can determine the other drivers with whom I need to "network", and hopefully so can the others. As a car how do I work out which other vehicles in the vicinity are part of the problem and which are not? Over what range do my network capabilities have to operate? 50 yards? 100 yards? (The use of Imperial measurments is entirely deliberate.)
I am currently of the view that "networking" would have to be a major part of any autonomous vehicle scheme, and that networking would require roadside infrastructure on a massive scale because without it so much of the information that a driver observes (or should observe) which is in real time and may be well ahead of the actual current location, would be missing.
I simply can't see that happening, and without it I can't see automonmous vehicles happening either.
Thinking beyond our current limitations is a Good Thing
That is true beyond challenge. Recognising that there are, or may be, limitations to what we can currently achieve with IT systems is imperative, or we may finish up wasting a collosal amount of money and have nothing useful to show for it at the end.
I, for one, will on principle refuse to join the crowd who sing the praises of the Emperor for his choice of clothing; it will take a great deal to convince me that autonomous cars can be a practicable option within my lifetime, which (hopefully) has still got some time to run.
You step out onto the road in the path of an oncoming vehicle and you are not a pedestrian crossing giving you right of way at that point in time then you are to blame. Why would it be any different to the current situation?
Also all teh very *COMMON* scenarios are easy to think up and program rules into the system to deal with. That just leaves the uncommon scenarios, but remember the self driving car only needs to be better than the *AVERAGE* driver to reduce accidents and deaths on the road. I know most people, men especially think they are all the Stig, but the reality is quite different.
Interesting second scenario, however it can't be solved by networking the cars as they will all be on their own proprietary system which may have been programmed to handle the situation differently.
That leads to another scenario, car turning right at lights is past the white line, car behind over/on the white line, the lights change. Front car decides not to go and the rear car does. I know the sensors should technically override the first decision but how well are these things going to be programmed that such a scenario wouldn't cause a crash either software or actual.
What about single lane roads with laybys for passing. What happens when both cars go too far forward how does the system decide who should reverse or do they both reverse and how do they know where the layby was and how far they are from it? That would require knowing you're on a single lane road and not a road full of parked cars or just a narrow part of a road.
What about a road full of parked cars, who gets or takes right of way? The current method of waiting and flashing can't be applied to computers because what if both signalled to go or not? It would be an endless loop of going reversing signalling.
I can probably think of a 100 scenarios where the programming is going to be a sod if not impossible.
I can probably think of a 100 scenarios where the programming is going to be a sod if not impossible.
Not difficult, is it?
I am still puzzled about why exactly. I know any research programme can have spin - offs with benefits that were not envisaged at the outset, but I cannot see the current intention. Reduced accidents? Perhaps, and certainly a desirable outcome; the downside is that the cost per accident prevented is likely to be far, far greater than the current cost of the accident taking place, although obviously the loss of a human life, or life - changing injury cannot sensibly be costed.
If implemented on any meaningful scale (which would really mean in the great majority of vehicles) then the cost of those vehicles themselves might be such that for the majority of people they are unaffordable. As yet nobody has tried to argue against my assertion that no such system could work without more or less continuous roadside infrastructure, and that too would come in at a truly eye - watering price, and someone would have to pay it; "general taxation" one might say, but where does that come from..?
And it couldn't just be limited to major roads; basically everywhere would have to have the infrastructure available; drivers who are no longer having to actually drive will neither develop the essential skills of hazard recognition and spatial awareness nor retain them, with the result that "drivers" will be hopelessly ill - equipped to cope when the autonomous system says "your turn now, pal" and promptly cedes control back to the human occupant.
If the intention is reduced accidents then I would suggest the following: (a) much tougher driving tests, (b) stricter enforcement of road law, (c) much tougher penalties for offenders and (d) really serious sanctions for those who are convicted of driving offences but cock their noses and carry on driving in the same way anyway.
So perhaps someone can up with some sort of figure for the cost:benefit ratio of making autonomous cars actually work - I'd very much like to see one.
OTOH perhaps it's all a plot of Uncle Sam's to force us to use buy their technology whether we want it or not or need it or not; it would make their keeping track of what each of us is up to sooo much easier, what with the on - board inward - looking camera and microphone...
I'd prefer a car that had a mind of someone else's - in particular, a someone else who I chose to rely on.
I think the danger of control unit / firmware replacement on vehicles that are out on the street overnight or are multi-tenant (driverless taxis and hire cars, for example) has been underestimated. Consider a replacement that was coded to work perfectly until a random time (perhaps years later) when the vehicle was doing 60+ MPH on a motorway, at which point it was designed to pull over to the hard shoulder, execute a braked turn, accelerate back down the hard shoulder until it was doing at least 70mph then pull out and aim for oncoming traffic.
Given the old adage that "if you can reach the hardware, you own the machine", I don't know how one can prevent this.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
