[thedoc (OP)]

Cars that drive themselves are very nearly here! We
went to test out Volvo's prototype that's being released
to the streets in two years...
Read a transcript of the interview by clicking here

or [chapter podcast=1000890 track=14.10.28/Naked_Scientists_14.10.28_1002878.mp3] Listen to it now[/chapter] or [download as MP3]

[David Cooper]

The sooner the better - it'll be a lot safer on the roads once human drivers are banned (or are overridden by software whenever they step out of safe limits - people will still be allowed to drive, and children will be allowed to take the wheel too, a red light coming on whenever the computer has taken control away from them so that they can see how well/badly they're doing).

[syhprum]

Driverless cars have been around since the eighteen twenties we call them railway carriges

[Monox D. I-Fly]

The sooner the better - it'll be a lot safer on the roads once human drivers are banned (or are overridden by software whenever they step out of safe limits - people will still be allowed to drive, and children will be allowed to take the wheel too, a red light coming on whenever the computer has taken control away from them so that they can see how well/badly they're doing).

Can we trust AI, though? After watching an episode of Yu-Gi-Oh! Vrains where it was revealed that Blood Shepherd mom got paralyzed badly after the AI driving their car malfunctioned, I don't think that driverless cars are good things.

[David Cooper]

Human drivers often malfunction though, so once the statistics show self-driving cars to be safer, human drivers will have to be banned (unless they can pass ever-more advanced driving tests and display superior skill than the AIs), and by banned, I mean that if they are driving, the AI of the car will always be allowed to override them at any time when it considers it necessary.

[jeffreyH]

The recent Boeing fiasco shows just why letting a machine decide what to do might just be a bad thing. People die.

[David Cooper]

The Boeing fiasco was not AI, but AS (artificial stupidity). However, even then, suppose human pilots were so bad that Boeing's AS made their system safer than having human pilots do everything, it would still save more lives by using it than it costs, and that's the important point: when the statistics say it's safer, you have to use it, no matter how imperfect it is.

[jeffreyH]

Banning humans from taking over?

[alancalverd]

The Boeing antistall system has, to date, crashed more 737s than those not fitted with it. Likewise the fatality rate per mile for driverless cars has, to date, exceeded that for human-steered vehicles. Beware of statistics!

Fact is that cars and planes were designed to be piloted by humans, and our transport systems have evolved around that fact. You could start from scratch and design a pilotless drone that shut the doors, started the engine and took its cargo to wherever, but even the simplest shuttle train finds it difficult to cope with the unexpected. The great advantage of human drivers is their ability to turn off the main road and/or stop, and a human pilot talking to a ground controller can organise a diversion with whatever medical or technical facilities he thinks necessary, without fouling up the entire system or crashing into anyone else.

Fact is that humans, though fallible, are exceptionally flexible and generally motivated by whatever motivates their passengers - survival and comfort above all else, and if you can't get there today, just land and make a phone call. Even amateur pilots spend a lot of training time flying by the seat of our pants: cover up half the instruments, disable one or two controls, work out what's happening from what's left, and configure the aircraft for the least damaging arrival. Worth watching "Sully" to work out how you might program a machine to decide whether and how to ditch in the Hudson with two dead engines. Or study a famous aerobatic sequence  (it's on You Tube somewhere) where one wing folded upwards, so the pilot flew back to the aerodrome upside down, using only power, rudder and elevator control until he could roll into a controlled crash from which he walked away. You could in principle write a program to do either, but would you?

[Halc]

Fact is that cars and planes were designed to be piloted by humans, and our transport systems have evolved around that fact.

Good point.  The AI cars would be more sensible if they all were that way, more predictable, but even then, you point out some interesting scenarios that would likely never be handled by manually programmed software.

The software in cars and planes is not AI.  AI teaches itself the same way humans do. They don't run an algorithm like the cars do.  I've seen very little actual AI out there, but what they've got is pretty impressive.  I'd still not want it to drive me somewhere any more than I'd like my quite intelligent child to.

Or study a famous aerobatic sequence  (it's on You Tube somewhere) where one wing folded upwards, so the pilot flew back to the aerodrome upside down, using only power, rudder and elevator control until he could roll into a controlled crash from which he walked away. You could in principle write a program to do either, but would you?

Got my own anecdote on that theme: TWA_Flight_841 (1979)
I witnessed that one. A 727 does a dive with double barrel roll, losing control as it broke the sound barrier. The solution was to drop the landing gear, creating enough drag to bring the speed down to where the controls responded again.  Minor injuries, mostly due to higher G forces than they paid for.  My vantage point was Davidson, MI, where the sonic boom could be heard.  Yea, it made you look up, but it was long over by the time the sound hit you.

They did try to blame the software (the 'AI') on the 737 anti-stall thing, but it didn't stick.  The blame came mostly down to safety being compromised if they thought they could save some money in the process.
https://moneymaven.io/mishtalk/economics/boeing-737-max-major-design-flaws-not-a-software-failure-rVjJZBVzZkuZLkDJn3Jy8A/

[David Cooper]

If x, then y. --> If not x, then not necessarily y.

If the statistics show something to be safer, it's safer. If the statistics fail to show that it's safer, that is a different case from a case where the statistics show it to be safer. Clearly those planes were not as safe as others, and the statistics bear that out. With self-driving cars, we're getting close to the point where they're safer than human drivers, even if they occasionally go badly wrong and kill people. Human drivers occasionally go badly wrong and kill people too, so it's all about which system causes more of those deaths. There is one other factor though that needs to be considered, and that's the possibility of terrorists or a rogue government hacking all the cars and programming them all to crash at the same moment in time. That possibility affects the safety statistics even if it never happens: the potential for it to happen is what matters, and the probability of it happening. That would also be hard to calculate, so we won't know when it's time to ban human drivers and replace them with AI. There will be a large guess involved. There will also be a long delay while they work out how to prevent self-driving cars being used to deliver and detonate bombs (although I expect they'll allow them first and then have to ban them again after a few bombings have occurred, and then they'll say that no one could have predicted that).

[alancalverd]

But you just did!

[David Cooper]

That's exactly my point: they achieve a lot by throwing money at problems, but they're reckless, dim, and they don't listen.

[alancalverd]

So far, driverless (or at least "driver-assist") vehicles work well on motorways and similar controlled, predictable environments. Things go wrong when you have to make a choice.

Dog in the road. Stop, drive round it (which way?), blow horn, or slow down and drive straight at it? The last is probably the safest as the dog will recognise the threat and move, but you don't know which way. But a bull or a ram is quite likely to hold its ground to defend its family, so your car needs to recognise species (Been there! We stopped, and the ram charged at the car!)

Arrive at destination, turn off the main road onto a building site with temporary signs that have been changed since yesterday. Can your car read handwritten "office closed - report to traffic marshal for parking"? Is the truck in front of you stopped to allow other traffic to pass, parked, or unloading?

Motorways aren't a problem anyway. People die on side streets, city roads, country lanes and work areas (including supermarket car parks), where programmed vehicles tend to get confused.

[mermaid12]

maybe in 50 years... But I'm scared of this