Waymo van prang, self-driving cars still suck, AI research jobs, and more This week's AI roundup includes an alarming report from California's Department of Motor Vehicles about how shoddy autonomous cars still are, a Waymo self-driving car crash, and some news from Facebook's F8 conference and its new job posting. Uh oh, not another self-driving car crash It’s Waymo’s turn to be involved in a car …
It's a work in progress. There have been a few smashes, and unfortunately one fatality. As with any method of transport, or any situation where things are moving at speed independently, accidents will happen. To say "self driving cars suck" is a little harsh, though this is El Reg so a little rhetoric is expected.
In my opinion, the tech is already safer than human drivers, and I for one welcome our robotic vehicle overlords.
> Cars? Why on earth cars?
Because the automobile industry is highly competitive and there are heaps of money to be made building the car 2.0. Trains, not so much.
Autonomous cars are not a solution to some existing problem, they are a means to reshuffle the automobile market.
Supervised autonomy, of course. It only covers motoring and braking, none of the other operations such as PA, doors, CCTV, heating and lighting etc. The various modes such as full ATO, protected or coded manual, slow manual etc describe how either the driver watches the robot, or the robot watches the driver.
"Do trains first. Should be simpler, right?"
They did that years ago. Have you not used the DLR? That's fully automated with just an attendant to open / close the doors.
"Tell that to a Central Line driver on a rainy day."
We need to replace those overpaid button pushers with automation ASAP.
""Tell that to a Central Line driver on a rainy day."
We need to replace those overpaid button pushers with automation ASAP."
The Central Line very often has to be driven in coded manual or even restricted manual because the automation (full ATO) cannot cope with wet rails and signal loss due to heavy rain or storm conditions.
https://www.davros.org/rail/signalling/articles/central.html
Now, take your copy of the Daily Heil and go and shove it down a dark hole somewhere... a tube station, perhaps. Honestly... IT engineers, managers, programmers, eh? We need, ASAP, to replace those overpaid pencil and button pushers with... well... anything would be better really, wouldn't it? Surely we could build computerised systems which are 100% reliable and can cope with every circumstance, and be flexible enough to keep their services running in almost every routine eventuality... then we wouldn't need these people keep holding us all to ransom all the time with their £200 keyboard replacements and £79 battery replacements and ongoing software licensing fees. Oh, they can JUSTIFY that can they? Changing security threat profiles require a rapid response do they? Pull the other one, it's got "communication cord" on it. And just look at the TSB fiasco... it's a public service for goodness sake... people rely on being able to pay their bills you know...
Gimme the wages of 1/10th of the drivers on the line, and I'll build you an automated system that is immune to signal loss (why the hell do you LOSE SIGNALS just because of a bit of rain in BRITAIN! Of all countries!), able to cope with any London storm conditions, and where in the wet it will go slower and do a better job than any human could ever do.
Honestly, train automation is not hard. "Only drivers can possibly do this" is just blatant - and false - propaganda to justify extortionate wages backed by union intervention. Other countries, which everything from tsunami to tornado, manage this perfectly well with the same kind of stock on the same kind of rail, and often in much worse environments, and much more cheaply.
I will HAPPILY trade the current system for one that literally turns itself off in stormy weather if it thinks it's unsafe to continue. Because the savings of £30k+ per shift, per train, per year into perpetuity would immediately pay for everyone to have a personal London taxi in such rare instances. And an INCREDIBLY DUMB computer system capable of doing the job. You don't need AI to operate a train. Nor an always-on Internet connection to every device. Nor an intelligent signalling system.
"Gimme the wages of 1/10th of the drivers on the line, and I'll build you an automated system that is immune to signal loss (why the hell do you LOSE SIGNALS just because of a bit of rain in BRITAIN! Of all countries!), able to cope with any London storm conditions, and where in the wet it will go slower and do a better job than any human could ever do."
PROVE it, then. Spell it all out or show us a comparably-sized system that can handle everything including a full track break or obstacle.
Absolutely flabbergasted at the lack of support for automation.. I'm certainly not suggesting we retrofit all existing rail services with computers, but the bank example? Yes there have been a few high profile failures, a few consumers have been inconvenienced, but frankly the bank doesn't really give a shit about consumers anyway. I'd rather rely on a computer to process my bank transactions, like it has been doing millions of times per day since at least a decade before I was born (not yesterday!)
Surely what most of us here are paid for is essentially working to automate mundane tasks carried out by humans so they can spend more time doing things which require a human brain. Unless you work in SEO or something, in which case, take Bill Hicks's advice and kill yourself.
Perhaps suggesting that anyone kills themselves was bit out of order. I wasn't being serious.
"The Central Line very often has to be driven in coded manual or even restricted manual because the automation (full ATO) cannot cope with wet rails and signal loss due to heavy rain or storm conditions."
So we need to get that fixed, install full automation and fire the button pushers ASAP. £60K for an unskilled job using 1 finger?! They are forever striking over nothing too.
There is NOT automation on the Central line at the moment - all trains have drivers.
People freaked out over automated elevators/lifts. It wasn't until there was a labor shortage that automation in that area picked up. It took almost 50 years from viable automation to widespread adoption. It isn't the technology that prevented the uptake, but people's opinion. People haven't changed, so I expect a similar lag in acceptance of self-driving cars.
Have you not used the DLR?
No, but I have used the Lille Metro(1). Fully automated trains with no door-button operators, no "supervising drivers" and a sizeable fraction of the line length above ground and therefore in the rain or the snow or the wind or whatever.
There are locked panels at the two ends of each train that are clearly intended to be able to be opened to reveal train controls, but I have never, ever seen them in use.
(1) Indeed, I still do, five days a week.
Lille Metro 1 was built from scratch in 1983 and is a rubber tyred metro system, not a rail system. It's a great achievement for sure, but the de novo approach means that you can walk alongside one of the trains down the tunnel on a lit, segregated path in the event of an evacuation, something that you cannot do in the 12 foot diameter metal pipes of the London Tube.
The fact that LM1 has manual driving provision indicates that drivers are still required. They still have to be trained, paid, they still have to practice the various techniques required. The trains still need maintenance, they still need cleaning. The tunnels and guideways still need inspecting. It's still a controlled environment. There's no getting away from workers withdrawing their labour and causing disruption - the DLR was shutdown by strike action recently after decades of good management/worker relations. It was down to changes made by a new management company taking over. Good relations is what makes a service strike free, not threatening to sack everyone and replace them with robots.
"Then I suggest you sign up. They are recruiting at the moment. I hope you make it through the pre-interview questionnaire at the very least."
I'm sure at least I would spell the form correctly which probably beats most of the incumbents.
Existing railway staff that are recommended get priority. (So basically it's a typical union closed shop):
https://www.telegraph.co.uk/news/general-election-2015/politics-blog/11730449/Want-to-be-a-Tube-driver-Well-you-cant.-Heres-why.html
Yet another reason why automation is needed to replace this model.
"I'm sure at least I would spell the form correctly which probably beats most of the incumbents."
I don't think they need to you write the form for them. I'm pretty sure they have one all ready-prepared for candidates to fill out.
As for the union arrangement, well there is an agreement that driver vacancies are offered internally first, yes. There are basically three manual labour "camps" on the Tube; stations, trains and depots. Most stations are RMT, most drivers are ASLEF and depots are a mix of RMT, Unison etc. So there's no one union that is set to gain by internal transfer priority arrangements, is there?!
So why would unions and TfL agree to recruit drivers internally first? Well, firstly stations and depots are all certified with the basic competencies mandated by law for working on the railway - track walking, evacuation, platform operations etc. That's 12 weeks worth of training and testing required there already. The re-evaluation process is much quicker than the full training, so basically it shortens the time you are paying a driver not train instead of drive. Secondly, this training process is quite arduous and the drop out rate is quite high, even higher amongst non-railway-exposed trainees. Dropouts waste money.
Secondly, it allows a staged recruitment into driving. Want to be a driver? Best bet is to get into stations or depots first. Get to know people, find out what the job is really about. It's not a bar on the general public - apply for a CS role first. 6 months in, apply to train operations.
Honestly, there's no telling some people, is there?
"I don't think they need to you write the form for them. I'm pretty sure they have one all ready-prepared for candidates to fill out."
I am referring to filling out the form. As I would have thought was obvious. You are either a bit dim or some sort of pedant.
"So there's no one union that is set to gain by internal transfer priority arrangements, is there?!"
No, there are 2 that stand to gain.
"It's not a bar on the general public"
That's exactly what it is. And if you did join for six months in a menial role and dont join the union, you won't be getting a driver job.
"In the end, it doesn't matter how dumb the mistakes made by self-drive cars are. The only thing that matters is that they cause fewer accidents and injuries per mile than human drivers do."
And yet the most common faults seem to be GPS and sensor errors. The sort of things that should already be a done deal, unlike the AI components.
I wouldn't expect a train driver to understand the complexities and reach of the jobs of programmers or computer specialists or financiers etc etc etc.
It's clear that there are a vast number of people who don't drive trains that are very happy to relate their expertise on the job they don't do, loudly and forcefully. And if anyone tries to tell them different, or even point out that there may be room for ignorance in their view of it all, they'll get really defensive and irate about it.
If you refuse to listen when someone says "actually it's more complex than that..." then I really don't want that kind of know-it-all building the transport systems which will be taking myself and millions of others around the capital. Thanks, but no thanks.
"actually it's more complex than that..."
Pushing a button to drive a train really isn't any more complex than that. Yes you have to have a few weeks training to stop at red signals, know what to do in an emergency, etc, but its basically unskilled manual work.
And because they have London over a barrel, it's overpaid unskilled manual work that is unreliable as they strike for any reason they can think of. For instance incompetent and dangerous staff being prevented from driving trains (and not even fired or disciplined!):
https://www.standard.co.uk/news/transport/tfl-tube-strike-2018-fury-as-district-line-staff-walk-out-in-support-of-driver-who-went-through-a3812231.html
It needs to be fully automated as soon as possible.
Accept people keep saying that and not doing the math.
Waymo 5 million miles 30 crashes humans are just over 4 per million so about 21 crashes for the same 5 million miles.
Don't get me wrong I am for it in places it makes sense, but lets not tout it as successful before it is out of R&D or praise its promise before it is delivered.
"All you need is a PhD in Machine Learning, AI, AI ethics and Alchemy, law or policy and some good research experience."
FTFY
It's interesting to note that Zuckerborg is keen on replacing his 20.000 moderators with a bot to moderate a Social Media site that depends on humans for it's product, anyone would think he doesn't like people.
Or paying them money.
The brain drain has left people worried about who will be left to educate the next generation of AI engineers.
Given that the current trend is to utilize AI/ML to design the next generation of devices, there might not be much of a need for the next generation. Engineers, or humans for that matter.
Okay, harsh yes, but this is something on trend I've seen since the 1980's when I was operating in the field. As I put it, I've been trying to automate myself out of job since I was a teen.
Isn't that what all programmers/engineers are trying to achieve? Replacing humans with machines to make our lives "easier" (it totally does). I don't see this as a bad thing, as long as we're not so good at it that the machines take over.
We automate the mundane stuff so we can concentrate on automating the more complicated stuff, and so on.
ROTM icon obligatory.
Isn't that what all programmers/engineers are trying to achieve? Replacing humans with machines to make our lives "easier" (it totally does).
Does it? Seems like we're always racing to keep up with changes and replace obsolete systems. It's a fashion-conscious field. Maybe that's inherent in the malleable nature of software. We tweak it to death because we can.
Seems like we're always racing to keep up with changes and replace obsolete systems.
Like replacing the scythe with a combine harvester? Them combines are totally in vogue right now.
If you prefer to grow your own wheat, harvest it by hand, separate the grains, hand mill the grains, hand knead the dough and bake the bread (which is by no means a bad thing) - cool.
print("Hello World!")
Surely that after looking at today's world and all the computers running it, the tech execs, etc., those have to go down in history as the scariest words in the world. Hell, FB alone gives me the shivers and not a product.. err.. user.
And this ultimately is the real problem that no one is prepared to address. Automation the removes jobs with NO REPLACEMENTS is going to cripple society. Everything at the moment is about short-term gain and mega-profits for the few, usually already very rich corporations. Once the bubble has burst and there is no longer a majority to consume, use or whatever, they are stuffed. At that point all the mega rich will be on their yachts, islands or whatever without a care.
What happens when the working population is essentially menial jobs that cannot be automated (and do not contribute to tax) and very wealth corporate types (who also do not pay tax)?
Automation as it stands is going to destroy the so-called civilised world. The divide between the "haves & have not's" is going to get ever larger and recent history has already shown us that those in the new generation of "haves" have not a jot of concern for the outcome of any decision/action they take if it does not directly benefit them.
"[Glorified] Cruise [Control] said some of the data incoming to the car’s many sensors did not quite match, giving conflicting information and causing the car to behave erratically."
Are these people so unedumakated that they're unaware of the work of Stratonovich, Bucy, and Kalman?
Apparently, Kalman filtering was knitted into the Apollo computer.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
