WikiWirral is where great friendships are made.
Forum Statistics
Forums65
Topics76,352
Posts1,033,280
Members14,560
Most Online16,551
Feb 2nd, 2024
Who's Online Now
11 members (ArranGirl13, 3 invisible), 7,969 guests, and 428 robots.
Key: Admin, Global Mod, Mod
Top Posters
sunnyside 45,164
MattLFC 22,315
Mark 21,269
granny 17,788
_Ste_ 16,345
Newest Members
Kalena1962, rattus, Lisrush, Emotecreative, ArranGirl13
14,560 Registered Users
New General Forums
New Wirral History
Tall Brick Chimneys
by diggingdeeper - 16th Mar 2024 11:56am
D.I.Y. / Building Shop
by mharford91 - 3rd Mar 2024 8:53pm
1970's Murder in Central Park, Wallasey
by cutespam - 4th Aug 2018 10:28am
Old Hall in Higher Bebington
by Rhoobarb - 25th May 2010 5:55pm
Oleo Works
by Ducko - 14th May 2008 3:09pm
Top Posters(30 Days)
granny 14
casper 13
bri445 4
Topic Replies
Facial recognition coming in supermarkets?
by diggingdeeper - 27th Mar 2024 3:50pm
Bolts at Bebington Road
by diggingdeeper - 27th Mar 2024 2:33am
Hot Spot Chippie Broadway
by Excoriator - 26th Mar 2024 7:57pm
Tall Brick Chimneys
by diggingdeeper - 26th Mar 2024 5:43pm
Lost river (Well, brook really)
by diggingdeeper - 26th Mar 2024 3:09am
Old Hall in Higher Bebington
by diggingdeeper - 24th Mar 2024 5:58pm
Netflix 3 Body Problem.
by BultacoAstro - 22nd Mar 2024 8:04am
Wisper electric bike. 36v .
by Dilly - 21st Mar 2024 7:36pm
This is Elvis
by GingerTom - 21st Mar 2024 9:27am
March
M T W T F S S
1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31
Top Likes Received (30 Days)
Kylix 1
cools 1
Top Likes Received
bert1 14
Mark 4
granny 3
casper 3
Previous Thread
Next Thread
Print Thread
Joined: Sep 2011
Posts: 760
Norton Offline OP
Veteran
OP Offline
Veteran
Joined: Sep 2011
Posts: 760
From the BT News pages -

Self-driving cars can be brought to standstill by laser pointers, say security researchers
A hack using cheap laser kits can create phantom cars and people which force self-driving cars to stop.
Security researchers have found a flaw that could allow anyone to fool a driverless car into stopping – simply by using a technique involving laser pointers.

Sensors on board autonomous vehicles allow it to detect any obstacles, but Security Innovation say that a laser kit costing under £40 can be used to create phantom objects that force the car to come to a halt.

“I can take echoes of a fake car and put them at any location I want,” Principal scientist Jonathan Petit told IEEE Spectrum. “And I can do the same with a pedestrian or a wall. If a self-driving car has poor inputs, it will make poor driving decisions.”

The simple hack – also known as ‘spoofing’ – messes with laser ranging lidar systems, which most self-driving cars rely on to generate an image of its surroundings. But Petit was able to create several fake obstacles that worked within 100m of the lidar unit.

“There are ways to solve it,” he continued. “A strong system that does misbehaviour detection could cross-check with other data and filter out those that aren’t plausible. But I don’t think car makers have done it yet. This might be a good wake-up call for them.”

Petit will present a paper on the hack at the Black Hat Europe security conference held in November.

Google, which has led the way on self-driving cars, has experienced several accidents since hitting the road – although all have been human error. In July, one of the firm’s Lexus SUV driverless cars was rear-ended in Google's home city of Mountain View, California.

It makes me wonder what would happen if the police point a laser speed gun at it?

Google Ads
Joined: Nov 2003
Posts: 21,269
Likes: 4
Wiki Master
Online Happy
Wiki Master
Joined: Nov 2003
Posts: 21,269
Likes: 4
I hope if i stood in front of one it would also stop saving me £40 lol

Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
Self driving cars are most unlikely to take over.

I doubt very much whether a few decades of electronic development is capable of matching 3.8 billion years of evolution. Nobody's managed to make a machine capable of catching a ball as well as an average person yet, and as for clay pigeon shooting, machines are hopeless.

To drive a car as well as a human being requires a huge degree of intelligence - well beyond what machines are capable of. There are moral judgements to be made as well. Do you protect the occupants in an emergency at the cost of killing a pedestrian for instance? Do you drive on regardless with an engine fault which might destroy it when you are taking a sick person to hospital?

There is a lot more to driving than not hitting things!


Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
I think you seriously over-estimate human ability and under-estimate where robotics are up to.

Better than ball catching, some robots are even catching odd shaped objects better than an average human.

Ball catching itself is a relatively trivial thing for robots (depending on what technological limitations you specify), not much money has gone into it because of the concentration on more difficult things eg playing table tennis which is possibly as hard as it gets because of spin that has to be sensed and compensated for.

Google self-drive cars have an exemplary safety record, far in excess of humans.

When all the majority of cars are self-drive, speed of journeys will be a lot faster, safety will be much better and they will use considerably less fuel.

Robots don't design their own limitations, the limitations are mostly human design or price now. Moral judgements are not that difficult on the grounds that humans can't even agree on what is right and wrong.

There are numerous technological solutions to your emergency transport/engine breakdown dilemma and no human could make the correct decision in your scenario, they would make a guess based on very little information.

The organisations that will fight this technology the most will be insurance companies.


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
I think you seriously underestimate evolution!

Google don't have enough miles under the bonnet to claim better safety. There is always a human in charge ready to take over but even if there weren't it would require literally billions of miles of experience to make any claims as to whether they are more or less likely to have an accident.

There is also the problem of hacking. Recently I think it was GM who had to recall thousands of Jeeps because you could stop them by hacking in via their radios! Very easy to fix, of course. But that is not the real problem, which is seeing ALL the possibilities of how things could go wrong.

Finally, I would say that software control of anything is potentially dangerous. It is - in practice - impossible to produce software of any complexity which will never crash. Testing ALL possible combinations of states and input conditions would be necessary for this, and that would take longer than the life of the universe to do. The practical approach is to track down as many bugs as you can find until you reach the stage where curing them produces more than you cure. Then you ship it!

That is why any complex piece of software is liable to crashes and freezes - a common experience with smartphones in my experience.

You want to trust this sort of contrapted approach to engineering, fine. Personally, I think it is folly.


Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
Google have over one million public road miles under their belt, they never been responsible for one crash but have been hit by other human driven vehicles over 12 times.

If someone maliciously wants to cause an accident, of course they can whether it is human driven or computer driven car they are always vulnerable to malicious activity.

Humans don't learn - how many times do you see cars driving too close, how many times do you hear people say that the car in front braked too hard/fast.

The evidence is that google is already in a different league of safety compared to human drivers. But hey, saving lives isn't as important as robot bashing.


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
A million miles is far too small to draw any conclusions. It would need to be a thousand times that - at least.

I have driven not far off a million miles and never had an accident. What does it prove? That I am a very good driver? Or perhaps a lucky one? Or perhaps that all the other drivers have covered for the mistakes I've made by anticipating them? I can't say for sure, and neither can anyone else.

As for humans not learning, they learned how to make cars and computers all by themselves which is more than you can say for any computer!

And you might factor in the consideration that maybe Google has a big investment in these things, and may be glossing over any mistakes by their car or blaming the accidents on others!

I suggest you read Roger Penrose's 'The emperor's new mind' before being so sure of what artificial intelligence can and (more interestingly) cannot do!

The best approach would be to use neural networks which are trained rather than designed, but how they do it is impossible to work out, unpredictable, and therefore hardly suitable for a safety critical system. Google's approach of which I know a little is fundamentally flawed in my opinion.

Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
The driving test is done only on a few miles which is the required standard. Humans have a pass rate of just less than 50%, Google cars would unquestionably exceed that.

You are exceedingly lucky not to have had an accident, most of Google's cars' accidents have been rear end shunts which corresponds to my experience having been shunted on a number of occasions (usually when I was stationary).

Neural networks and other artificial intelligences are a game of probability, its the statistics that matter and the stats for Google cars are impeccable, the stats for humans are not.

No doubt a google car will have an at-fault accident at some time, it still won't upset the balance of probabilities that they are safer than human drivers.

Artificial intelligence is as about definable as human intelligence - both are concepts without any particular rules or boundaries, its mostly a sales gimmic.

There is no computer made that does not use computers to design and make it, its a human-computer partnership, one can't happen without the other. The computers were clever enough to make themselves indispensable but us egalitarian humans believe we are in control wink


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
Sadly, the stats from Google are far from impeccable. The mileage is far too small to draw any conclusions as I have tried to point out.

I understand, too, that the car gives control back to the driver under conditions that it cannot handle. If there were no driver, an accident would seem pretty likely, but it will not appear in Google's statistics as such.

To get a real comparison you would have to let the car run around unmanned, and see how many accidents you get, but it is obviously unethical to put the public's life at risk.

And I am old enough to have designed computers without computers to help. I have also used them to design other computers, and they are only an aid to removing the non-creative donkey work. Tools. Glorified screwdrivers. The design is always entirely human.

They are a long way from artificial intelligence. But on this subject, I would ask a seldom asked question about AI which is "How would we recognise artificial intelligence if it did occur?" It may be so different from our thinking as to be unrecognisable. It may even be already here!

But it certainly isn't living in CAD tools which do simple tasks very fast, but are otherwise pretty dumb!


Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
How do you calculate what mileage is acceptable, the google stats are sufficient to get above 95% uncertainty. The trials haven't finished, google are still gaining 10,000 miles a week, this rate will increase soon as they start trials in other countries including ours.

How many real miles do you think aircraft flight computers do before becoming accepted, I guarantee it is less than one million miles.

My trade is electronics, I've been a programmer since 1975, I worked for one year in AI in 1990 and then again for 5 years from 1995. I wrote a complete multitasking windowing operating system before Windows was released in 1981.

My statement "There is no computer made that does not use computers to design and make it" is in the present tense, modern CPU's cannot be designed by hand, nor can they be built by hand.

As you say, AI is indeterminate, who is using who, computers or humans. We are both engines, our body and mind is only a chemical engine as is a computer.


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
I don't know what you mean when you say '95% uncertainty'.

The UK has about 30 million cars which do on average 15,000 miles a year each. There are about 1,700 deaths a years due to RTAs - one death every 265 million miles. A million miles of experience by google is totally meaningless. If they could do a billion miles and kill less than four people then you can say they are better than humans with a rather poor degree of confidence.

Aircraft use multiple computers and take a majority vote because software is considered untrustworthy for safety critical function. I'm given to understand that they run on different microprocessors, the code is developed by different teams, and I believe the programmers come from different universities in some cases in order to minimise the chances of a common error affecting more than one of them. It would be far too expensive to do that with cars. Google certainly doesn't.

As to computers being used to design computers, it is certainly true, but only in the most trivial sense. They contribute no more to the design than does a pencil! I worked for a time designing microchips. The early ones were laid out by hand, but as the number of gates went up this became impractical. The last chip I designed had five million gates, and had the masks been printed out to a size where manual methods could be used it would have been bigger than a tennis court!

A computer is essential, but its intellectual input is nil. It is simply used to place and interconnect components as defined by a hardware language or a schematic. A mindless but repetitive task well suited to a computer. From memory it took me about three months including simulation and a few 'tweaks' to improve its performance.

I will begin to take your claims about driverless cars more seriously when Google have a billion unmannned road miles under their belt. So far they have none. There is always a driver at the wheel although not necessarily doing any steering.


Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
I don't believe a one million gate CPU could be hand designed for operation at multiple GHz, it is no longer a logic/electrical design, it is an RF design, every path is incredibly critical for length, resistance, capacitance, inductance and cross-talk. Then there are the tolerance to impurities, thermal mapping, radiation tolerance etc etc.

You are extending the criteria to accidents that cause deaths which happen every 265 million miles, you could extend it further and specify for innocent party deaths which would be a lot higher or even more criteria, not dissimilar to a political filibuster.

Keeping it simple and quantifying for any at-fault collisions and you come down to around 150,000 miles. If there are no at-fault collisions then there are no deaths.

The current Google cars are not being developed as driver-less car, though of course we have driver-less trains/trams in this country already.

Electronic stability control is now compulsory on new cars, this has an obvious ability to cause accidents but does not go to the degree of testing that Google is undertaking.

By uncertainty I meant confidence interval, 95% and 97.5% are two common measures that are considered highly unlikely, 95% confidence interval is not the same as 95% probability which would not be considered highly likely/unlikely.

My biggest concern are other companies that are also bringing in autonomous driving, Google has invested a lot of time and money and clearly lead the field, if Google gets approval other companies that haven't been as thorough can use the Google precedent to short cut the route to market - especially as Google would have a monopoly.

Some of the entries into the DARPA grand challenge are exceedingly worrying.


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
Well in high speed microchips, not ALL paths are critical, but when the computer has done its layout it probably won't work at the speed you want. At that point you have to do some investigation and move stuff around to minimise critical path lengths etc. You can, with most tools, specify what paths you want short, but this slows the machine down and its generally quicker to let the computer do its best and then go in and tweak it.

I am using deaths because they are recorded and the data on them is reliable. Not all collisions are reported to police or insurance companies so the data on non fatal accidents is suspect. If you are concerned about safety, it is deaths that one needs to look at anyway. In my opinion a system that caused MORE minor scrapes but fewer fatalities would be perfectly acceptable. Sadly a computer program is likely to achieve the opposite! When things go wrong in a piece of software, they go badly wrong in my experience.

I am not sure how a failure in an electronic stability system can cause an accident. they are designed to be 'fail safe' or at least 'fail - no worse off than not having it' I would question the safety of any system interposed between driver and wheels that is capable of overriding what the driver does.

But a system that actually controls the vehicle is certainly capable of killing people if it goes wrong, and there is no way of ensuring that a software failure could have such an effect. The ugly truth about software is that it is untestable and entirely unsuitable for safety critical systems. Google with a background in highly redundant non safety critical software systems is unlikely to appreciate that they are naively venturing outside their area of expertise.

Joined: Jul 2008
Posts: 14,343
Likes: 19
Wiki Master
Offline
Wiki Master
Joined: Jul 2008
Posts: 14,343
Likes: 19
Google buy in expertise, at least one specialist company was bought by them and also individual programmers eg from the DARPA challenges have been very quickly snapped up by Google.

Google probably run more computers than any other company in the world (the only other likely candidate is Microsoft), they know a lot about fault tolerance, redundancy and error checking.

I'm confused how you are saying that electronic stability systems are designed to be fail-safe then saying that autonomous driving systems can't possibly be fail-safe?

Electronic stability systems do over-ride what the driver does, it brakes, releases brakes and steers the car by individual wheel braking. Over-riding the driver is their sole purpose in life, when the driver operates the car into an uncontrolled or unstable state, the ESS jumps in and attempts to mitigate the drivers mistakes.

Inside the CPU its not just path lengths, its edge timing and waveform shape, multi=layer chips make this easier but then you increase the thermal problems which almost invariably is the prime limiting factor on speed, any slow edge rise-time produce heat, any over-shoots produce heat, ringing produces heat. Its all got to be time and impedance simulated, this cannot be done by hand.

Any mechanical, chemical, biological, electrical, electronic or software based system can have faults, the advantage of software systems is that they can be tested and improved much more quickly and cheaper during development.

Look at satellites, the majority of significant avoidable faults that occur are mechanical/physical, I can't even recall a significant software fault occurring on any satellite. They are very thoroughly tested in extremely small production runs (often only one is produced).

An interesting debate that occurred a few years ago at a conference was:- a software product has four independent programming teams available, how many teams should be allocated to the production software and how many to the testing/simulation software. The (eventual) general consensus was one team on the production software and three on the testing/simulation software, almost everybody agreed on that but there was strong disagreement on the reasons for that choice.

I think I have raised every point I can think of now, thanks for the well behaved discussion, I value that greatly. I agree with your concerns to an extent but it is a matter of balancing risk against progress - there are a lot of crashes, the risk is currently higher than most of us would like, there is obvious scope for improvement using technology, nothing will be perfect.

Thanks again thumbsup


We don't do charity in Germany, we pay taxes. Charity is a failure of governments' responsibilities - Henning Wehn

https://ddue.uk
Joined: Jan 2010
Posts: 1,841
Likes: 3
Forum Addict
Offline
Forum Addict
Joined: Jan 2010
Posts: 1,841
Likes: 3
I too have enjoyed the discussion. It is a pleasure to have an exchange of views which do not involve the ad hominem attacks that spoil so many posts on the internet.

My point about electronic stability systems is that they are limited to controlling the power to the wheels of the car. If they fail to work, the driver is no worse off than if they didn't exist. Yes the car will require more careful driving, but that is all.

This differs from a self-driving vehicle, in that the vehicle is under full control of some kind of software. If it fails to work, an accident is almost inevitable.

I am well aware of the problems you mention in designing microchips. The tools include signal integrity software which is capable of modelling these effects, but only up to a point. I have never known a case where human intervention is not involved except where the design is extremely simple and non demanding. The point I was making, however, is that these tools are not involved in the design of the device. They are simply used to implement it. The design is always done by a human being, who uses the computer to plod through the sort of tedious but well understood computations necessary to implement it. I am not suggesting computers are not essential, but rather that they do nothing original. They simply crunch numbers according to human ingenuity.

I think also that intuition is important in driving. One can often see that someone is in a hurry, and pull over to let them pass, simply from their style of driving. And on can tell by the fact that someone waits at a junction where they have priority that they are allowing you to go first etc. I can't really see how any currently available technology can be expected to handle interaction like this. We do it without having to think about it. A machine is deterministic and requires a defined universe to function properly

Finally, I think the fact that real time software is - in practice - impossible to fully test is worrying. It may be very unlikely to go wrong, but when it does it can do whatever it likes to the vehicle, and what it does is unpredictable.

I've done a lot of work over the years on equipment which is safety critical, and although software has been involved, the safety critical bit was always done in hardware. If you are careful you can give strong engineeering guarantees of the reliability of the system with hardware which is impossible in software.

I too, think the debugging phase requires far more work than the original code writing, by the way. This has a highly undesirable impact on development, as if there is a design change late in the debugging phase - even a simple one - it sets the debugging team back to square one and you lose a lot of time. The hardware can accommodate late changes very quickly.

We'll have to wait and see if self-driving cars ever become accepted on our roads in large numbers. Personally, I think it will be many decades or centuries before this comes about, and I am unlikely to see it.

Good discussion!


Moderated by  Mod 

Link Copied to Clipboard
Random Wirral Images

Click to View Topic.
Newest Topics
Facial recognition coming in supermarkets?
by Excoriator - 27th Mar 2024 10:52am
Hot Spot Chippie Broadway
by Excoriator - 26th Mar 2024 7:57pm
Bolts at Bebington Road
by Excoriator - 23rd Mar 2024 9:48am
Netflix 3 Body Problem.
by BultacoAstro - 22nd Mar 2024 8:04am
Tall Brick Chimneys
by diggingdeeper - 16th Mar 2024 11:56am
For Sale & Free
Wisper electric bike. 36v .
by Dilly - 21st Mar 2024 7:36pm
This is Elvis
by GingerTom - 17th Mar 2024 2:37pm
Member Spotlight
diggingdeeper
diggingdeeper
Wirral
Posts: 14,343
Joined: July 2008
Today's Birthdays
There are no members with birthdays on this day.
New Wirral Info
Facial recognition coming in supermarkets?
by Excoriator - 27th Mar 2024 10:52am
Hot Spot Chippie Broadway
by Excoriator - 26th Mar 2024 7:57pm
Bolts at Bebington Road
by Excoriator - 23rd Mar 2024 9:48am
Get your Money Tree ready
by keef666 - 22nd Mar 2023 12:27pm
Restaurant/pub with outdoor seating - Bromborough
by CarterUSM - 7th Aug 2020 7:59pm
News : New Topics
Gaza
by diggingdeeper - 29th Oct 2023 9:28am
Lost river (Well, brook really)
by Excoriator - 10th Sep 2019 8:50am
New Enthusiast Forums
Netflix 3 Body Problem.
by BultacoAstro - 22nd Mar 2024 8:04am
Xbox
by TudorBlue - 3rd Mar 2024 8:48pm
Wiper mechanism needed
by Peter0787 - 20th Feb 2024 4:07pm
Popular Topics(Views)
5,069,750 WIKI WALK CHAT
4,013,797 Spotted!
Powered by UBB.threads™ PHP Forum Software 7.7.5