IS/CT/RC Discontinued in UK (and Europe?)

[ganzoom]

On 11/4/2020 at 10:02 AM, Linas.P said:

where is fun looking at AI "learning", cannot really see anything more boring than that. 

 

You're bored by probably the biggest technological revolution in living memory, that says it all really. 

DeepMind alone have published 17 paper in Nature since 2015, if you honestly don't get what a big deal AI is then I honestly don't know what else to say.....

[Linas.P]

4 minutes ago, ganzoom said:

You're bored by probably the biggest technological revolution in living memory, that says it all really. I suspect the Toyota/Lexus board shares your sentiments :).

TRUE AI and FULL Automation will be big revolution, sad reality is that Tesla does not have either and is not even close to developing it.

[ganzoom]

2 minutes ago, Linas.P said:

TRUE AI and FULL Automation will be big revolution, sad reality is that Tesla does not have either and is not even close to developing it.

Do you even know what you mean by 'TRUE AI' ?? Writing in capitals don't have any meaning?? Why not start reading up on the subject before the world overtakes you. 

[Linas.P]

2 hours ago, ganzoom said:

Do you even know what you mean by 'TRUE AI' ?? Writing in capitals don't have any meaning?? Why not start reading up on the subject before the world overtakes you. 

This is ironic that you say it... for many reason... I understand that you are hyped by some inspirational speeches, but reality is much more "normal" and sadly not as exciting as you may think. 

What I mean by "True AI" (also known as "Strong AI") is exactly that - artificial intelligence where computer can learn and improve by itself in general sense, to some extent becoming same as human or adapting to circumstances similarly as human, rationalising it's environment etc. This is just hypothetical construct which does not exist - some say humans will cease to exist if one is created (namely Mr. Elon himself).

What we generally confuse with "true AI" is "machine learning (ML)/deep learning (DL)". Basic example of ML would be "smart" light which has ambient light sensor and could learn when to come ON based on ambient lightning conditions and user behaviour. In this example it would learn when to turn itself ON based on when user turns it ON, light would work on code similar to this "WHEN Ambient light => X, Turn the light on, WHEN Ambient light =< X, Turn the light off". All light would learn is lightning level to replaces the "X", that is all - just a line of code. DL would go further and would check if there is anybody home before running the above command. DL is not better than ML, just has more lines of code to check more conditions. Not only that both fundamental rules should be defined from beginning e.g. light would not know what to do when bulb fails if there are no specific rule to tell it to monitor this condition. Equally, it would completely fail because it could not understand why user turns it off at 1AM when there is no ambient light, because it could not rationalise that user is going to sleep, unless such condition is specifically defined in the code. 

This (simplification) is similar to how Tesla cars works - they have no "True AI", they just have million lines of conditions, which based on sensor data decides when to flip the switch. Then you need to understand  the concept of "big or mass data". You see the only time "machine learning" can take place is when algorithm encounters exception, meaning that until something happens the algorithm cannot predict or be prepared to react to exception. So the Tesla's solution is to collect massive pool of data and just throw it at their algorithm to train it. Hence you often get these when browsing web:

They are designed so that we help the algorithm to learn what is "bicycle" in this case. As a humans with our intelligence we can tell what "bicycle" is even if we see just part of it or picture is blurred or it is some kind of weird one. Algorithm can't do this until it is trained. Meaning that it will run bicycle over unless it has been previously trained to know it does not taste nice.There are many difference between "machine learning" and "deep learning", but for that you will need to do your research as it will take very long time to explain.

This is what makes Tesla autopilot dangerous to use on the road - if it has not encountered certain exception previously, it won't have "rule" how to deal with it - in AI development this is called "novelity factor". That was cause of several deaths when car could not identify concrete barriers or white truck against sunny background and just simply "run them over".

With enormous set-of data you could only train algorithm to be safe with certain margin for error of say - 99.9999999999%. 0.0000000001% being possibility of "novelty" situation seems like that is good enough. Well, now consider that drivers could be making as much as 25 decisions every second when concentrating, but long term average for human is ~2 decision per second. Ok that does not tell us much, but put this in perspective - if there is 1 million self-driving cars are on the road, each on average being driven say 500 hours a year, then the combined fleet will need to make 3,600,000,000,000 decision every year, meaning that even with 0.0000000001% chance of unknown situations there will be 3.6 times per year when 2 tons projectile travelling at 70MPH hit something. Ouch!

The much more sad reality is that current algorithms are nowhere near 99.9999999999% correct even in "known" situations, never mind "novelties". To begin with humans can get away with making 2 decision per second, because we can drive cars subconsciously, we don't need to constantly think about angle of steering wheel or accelerator pedal, we only need to make "conscious decision" every so often (twice a second) and even in accidental situation we can react with trained reflexes. Algorithms don't have such ability, they cannot rationalise their surrounding, they can't really predict slippery road, they can't predict that toddler on pavement can fall into road and they are absolutely reactive to their surroundings. Whereas humans could be proactive and take steps to avoid potential danger even before something happens.

Further issue is that our roads are simply unpredictable, we have a lot things happening which should not happen - potholes, crazy cyclists, toddlers, random rubbish, british road workers, suicidal males jumping of the bridges and many more things. Individual drivers or cars are unlikely to encounter most of them, but collectively we know about it - thus when we see cyclist on the road we know we need to give them the widest breadth possible... ML/DL does not! If our roads would be much more structured and simplified, clearly marked and cleansed from various unpredictable risks then ML/DL could work much more reliably - in reality they sadly could not. 

Overall, it is generally accepted that for full automation (level 5) we need to develop near levels of "true AI", at very least in areas of driving skills motorisation and rationalisation of road environment. Neither Tesla, nor anyone else (in any field) has come even close to "true AI" - so certain Mr. Elon claiming they will "soon" have full automation are dreams at best, but most likely well calculated lies to boost company shares and his personal wealth. 

The only argument to contrary is that humans are known for making less than irrational decisions, especially when under certain influences. So I tend to believe that in near future (maybe 10 years) algorithms will be able to drive cars statistically as a good as we do collectively. Meaning they will drive as well as hybrid between drunktard and experience driver... And all this is not even considering things like somebody else crashing into you, or systems simply failing. If that is sufficient level for you to trust your life to algorithm, then I wish you good luck!

@royoftherovers - I promise you - this is interesting read if you have enough time to go over it 😁

[PaulWhitt20]

I though we had had bonfire night. [emoji854]


Thanks
Paul

[Linas.P]

16 minutes ago, PaulWhitt20 said:

I though we had had bonfire night.

I thought it was appropriate to celebrate my 4000th well researched post 😅

[royoftherovers]

I did read my friend and was impressed.Lets see what your correspondent makes of it.

Congrats on 4k.

[ganzoom]

8 hours ago, Linas.P said:

 

What I mean by "True AI" (also known as "Strong AI") is exactly that - artificial intelligence where computer can learn and improve by itself in general sense, 

Still haven't read the DeepMind AlphaGoZero paper I see :).

The current neural nets already 'learn' by it self, no human actually understands how it comes to the conclusions it does.

Ask your self is a single ant or bee 'intelligent'? Does a single ant or bee know how to build a nest or hive? 

Is a single human brian cell 'intelligent'? What about 2 or 3? 

We are on the cusp of the biggest technological change the world has seen. The fact my car parked on the driveway right now will get access to some of the most cutting edge AI coding is simply crazy.

Ofcourse you than come to the billion dollar question, does 'driving' actually require intelligence or is if just procedural? Do you find you need to concentrate more when driving or playing chess? 

The fact a car, something you go to work, do the shopping in, take stuff to the tip in can engage people to have these fascinating conversations is why I love our Tesla so much more than our Lexus. There really are nothing else on the roads like them. If it wasn't for ££££ we would have already replaced our IS with a Model 3.

[dutchie01]

Now back from Artificial to Human Intelligence. About 3 yrs ago i was adviced not to operate a motorvehicle for about 2 wks as my inner ear was causing lightheadedness. During that second week i felt ok so took the car for a drive and only then i experienced what data the human brain has to decipher in order to cross a busy roundabout. EVERYTHING moves. in different directions, at different speeds, different shapes, sizes. You have to predict where that all will end up in a split second. Of course i failed hopelessly and felt totally in the wrong spot, started sweating and slowly drove home. In short, my own processor was not able to cope with the massive input of data and could not process this in time to come to a result. It was then i realized that driving a car is way more complicated than i always thought and i came to the conclusion that selfdriving cars will take years and years to develop.       

[PaulWhitt20]

Isn’t it great how a thread can meander from topic to topic with interesting debate.

As an ex software engineer who wrote automation software, I don’t see how a computer can learn for itself. It will always follow rules programmed into it by the programmer.

In the example above the program can learn what light level to turn on based on recording the light level when it was manually turned on but that algorithm has been programmed into it by a human. I can’t understand how a computer can program itself, to be able to adapt and do new things that have not specifically been programmed into it.

If we did get to this scenario when computers could learn for themselves then we would certainly need the three laws of robotics.


Thanks
Paul

[royoftherovers]

1 minute ago, PaulWhitt20 said:

Isn’t it great how a thread can meander from topic to topic with interesting debate.

As an ex software engineer who wrote automation software, I don’t see how a computer can learn for itself. It will always follow rules programmed into it by the programmer.

In the example above the program can learn what light level to turn on based on recording the light level when it was manually turned on but that algorithm has been programmed into it by a human. I can’t understand how a computer can program itself, to be able to adapt and do new things that have not specifically been programmed into it.

If we did get to this scenario when computers could learn for themselves then we would certainly need the three laws of robotics.


Thanks
Paul

Re your last para Paul---Asimov was ignored by HAL in 2001 A Space Odyssey, was he not ?

[Mincey]

15 hours ago, Linas.P said:

 

@royoftherovers - I promise you - this is interesting read if you have enough time to go over it 😁

I enjoyed it too!

[Linas.P]

7 hours ago, ganzoom said:

(1)The current neural nets already 'learn' by it self, no human actually understands how it comes to the conclusions it does.

(2) Ofcourse you than come to the billion dollar question, does 'driving' actually require intelligence or is if just procedural? Do you find you need to concentrate more when driving or playing chess? 

(1) Don't you see that as an issue when creating dependable system on which reliability depends millions of  lives?

(2) I think it could be both, but currently it is latter. Take for example trains - their environment is isolated and they can be fully automated with few basic safeguards. Would it still kill people - yes of course if they manage jump in front of speeding train, but that won't be an issue with train technology - it would be suicidal people abusing it. For cars we currently can't do it, because our roads are too chaotic - we allow too many people to use them who should not use them (at all, even if we ignore driving automation for a second). If we redesign our roads the driving automation could be made less intelligent and more procedural, but there are always trade-offs. Current trade-off many developed countries have made - they made roads very accessible for all sorts of incompatible "objects", in the alternate reality where roads are more exclusive and access is based on suitability and purpose I can see automation as much more likely outcome.

It is good that technology and your car excites you, however we need to put things in perspective - Tesla did not create some panacea, in fact they are behind the race nowadays. The only difference is that Tesla in this case is that "teenager on social media" sharing every moment of their lives to the world, whereas other automakers keeping their technology for themselves until it is ready. If Tesla is winning anything it is only the marketing - they manage to market their level 2 system in such a way that many people believe it is more advanced then alternative level 3 or 4 systems.

When the systems and laws will be ready I will be very excited to use automation, sadly I have looked to all alternatives currently and there is nothing fit for purpose.

1 hour ago, PaulWhitt20 said:

If we did get to this scenario when computers could learn for themselves then we would certainly need the three laws of robotics.

You can automate "the learning" to some extent, but you are right - at the moment it would still be limited to scope defined by human programmer before hand. This is what differentiates "true AI" and limited scope ML/DL systems. When machines going to really start learning by themselves I am afraid it will be irreversible process, very exciting times as well, I am just not sure we humans going to be part of them.

 

[ganzoom]

40 minutes ago, Linas.P said:

(1) Don't you see that as an issue when creating dependable system on which reliability depends millions of  lives?

 

Not all, if it does what it suppose to do why do humans need to understand it?

The software Tesla is currently putting out is a massive step forwards interms of whats available to mass public. You claim Tesla are behind on the AI software front, can you find me another car you can buy today that will do this?? Not a tech demo, an actual car you can purchase and use in a unrestricted non geo fenced situation? 

The rate at which Tesla make progress is now the key thing to watch, I cannot wait to see how things develop. 

 

 

[Linas.P]

Just now, ganzoom said:

The software Tesla is currently putting out is a massive step forwards interms of whats available to mass public, can you find me another car you can buy today that will do this??

Exactly that - their technology is not better and the only difference is that it is available to mass public. They use public as test dummies, other makes don't.

If I sell under-cooked pies when other don't, it does not make my pies better - they are still raw!

[ganzoom]

1 hour ago, Linas.P said:

Exactly that - their technology is not better and the only difference is that it is available to mass public. They use public as test dummies, other makes don't.

 

Here is a Tesla going on road with no lane marking, overtaking cars double parked in the middle of the road, doing unprotected left turns, than highway speeds, and another example of Tesla FSD going around SanFranciso in at night. 

Please post a video of showing 'better tech', oh any £40K Model 3 can do this (infact any Tesla made since Dec 2016), can a £40K Lexus get even close? We all know no 2016 Lexus will even move 1 inch without human input. So why would you pay £40K now for a new Lexus when for the same money you can access something which is so far more advanced?? 

 

 

[Linas.P]

It can only do that because the human is supervising it, making it completely pointless - if you human is fit to supervise the system, then human is fit to drive the car as well. I would not pay as single penny premium to have this is my car over Lexus LSS+.

So indeed I cannot post the video of £40k car doing it, but it is irrelevant, because other automakers do not risk their customer life like this and do not use them to test such technologies.

They do not sell it, because it is not ready for sale... Tesla autopilot is not ready for sale either, however they choose to sell it because they know that  they have brainwashed customer base to pay for it. Same like apple knows they can sell stupid monitor stand for £999 . Ironically - difference being that apple stand is at least ready for sale and won't kill you...

[Overland]

14 minutes ago, ganzoom said:

 

Here is a Tesla going on road with no lane marking, overtaking cars double parked in the middle of the road, doing unprotected left turns, than highway speeds, and another example of Tesla FSD going around SanFranciso in at night. 

Please post a video of showing 'better tech', oh any £40K Model 3 can do this (infact any Tesla made since Dec 2016), can a £40K Lexus get even close? We all know no 2016 Lexus will even move 1 inch without human input. So why would you pay £40K now for a new Lexus when for the same money you can access something which is so far more advanced?? 

 

 

Amazing 

[ganzoom]

26 minutes ago, Linas.P said:

So indeed I cannot post the video of £40k car doing it

Thanks for confirming your statement about Tesla been 'so far behind' is totally incorrect :).

[Linas.P]

1 minute ago, ganzoom said:

Thanks for confirming your statement about Tesla been 'so far behind' is totally incorrect :).

If that pleases you more - fine, take it as you wish!. I never said Tesla is far behind, I just said that they are "not far ahead" at all. Tesla does not have any technology which is ready to use or adds any value and I don't need their half-baked beta crap system.

By the way Mercedes have cancelled their automation project. The goal of their project was level-5 and they are by now have level-3 automation together with BMW.

They have concluded that "car as a service" model makes no sense for Luxury brands where drivers values their individuality, car ownership and the driving experience and I agree with them. Shared self-driving boxes on the wheel is value proposition for car ownership for poor. However, in Luxury car market that is not a value which is important.

Not sure if BMW will cancel their part of project as well.

[ganzoom]

@Linas.P We'll just have to agree to disagree, ultimately at the end of the day the next few years will be very telling.

Just seen this article, it appears the head of Toyota thinks like you.......But posing posing in front of something that's made for the 1960s, whilst I did my shopping run today in something looks like its come out of a Sci Fi film.

Maybe your right, these companies aren't competing, its pretty clear one is still living the past whilst the other is delivering the future. I know as a consumer which company I would rather associate with.

 

 

[Linas.P]

Can you remind me if I need to be in the car, hold the wheel, be fit to drive and try not to fall a sleep when using Tesla "Autopilot"?

[ganzoom]

8 minutes ago, Linas.P said:

Can you remind me if I need to be in the car, hold the wheel, be fit to drive and try not to fall a sleep when using Tesla "Autopilot"?

As I've already shown you, the car can actually drive it self with no one in it, so you can do what you want. Here's another example. 

Can you find me a Lexus that will do the same? Again that's a Model 3, £40k, you can go and buy one tomorrow.

 

[Linas.P]

Just now, ganzoom said:

As I've already shown you, the car can actually drive it self with no one in it, so you can do what you want. Here's another example.

You didn't answer the question...

Doesn't matter what car can do... there are many things which could work in theory, but do not work in practice. Many cars can do 200MPH, but what is the practical application of that? Here BMW drives it self as well... so what? -  Do I need to be in the car, hold the wheel, be fit to drive and try not to fall a sleep when using Tesla "Autopilot"?

 

 

[dutchie01]

I know as a consumer which company i would rather associate with.

That sounds a bit like apple to me. Whatever they market at whatever price the fans will sleep in the streets to have it, i guess they want to be associated with, be part of the brand want to belong with the modern high tech independent intelligent exclusive image ( you get what i mean )

Enter Tesla. Where i live a Tesla is not exclusive, they are literally everywhere. Believe me or not but about a month ago during rushhour when entering the highway there were 3 behind me and 1 in front of me. It is the preferred companycar. Why? because of the advanced autopilot system? Almost nobody has this as it is a 7500 euro option. The only reason wht Tesla sold some 35k cars in the netherlands last year is fiscal stimulation it is with a big distance the most cost effective way of motoring. You can spend the saved money to your family.

Think of it the other way. The saloon market is melting away in favour of SUV's And what is the best selling car? A saloon the tesla 3, does not really add up does it? Take the fiscal savings away and Tesla is dead in the water. I do admire the progressive approach of the company which brought them success so far but i do find the styling bland and the buildquality dubious.

When i see the pic of Mr Toyoda in front of the hyper exclusive Century GRMN i know what i will choose and it is not a Tesla. There will be more technical expertise in the opening of the doors of the Century than in the entire Tesla3.