Oh Dear Another Dead Road User Killed By Computer

[ToraToraTora]

http://www.bbc.co.uk/news/business-43459156
When are we going to accept that our current software ability is insufficient for this application? at least Uber has the sense to halt their tests.

[ToraToraTora]

"For someone who claims to work in IT/AI, you appear woefully unaware of developments in this field. " - err no, you and others here are woefully unable to grasp the basics of computing.

[RobNorth]

If this stuff Is that good, why are we about to spend billions on infrastructure and signalling etc. for HS2 which presumably will have drivers...and run on rails.....early 19th century technology..... Cars are late 19th century technology and have barely changed in a century. The methods of propulsion, control and stopping them is the same as it ever was and they have no infrastructure to speak of, in the main sharing what is there with bicycles, pedestrians, horses and much other just as they have since day one. Politicians the world over are obviously hoping that a few geeks can fix all this with a box of switches screwed to every car, but I fear that they have a very long and disappointing wait...

[ToraToraTora]

good points Rob, driverless trains are at least feasible.

[Hymie]

Looks like the future is already here:-

https://en.wikipedia.org/wiki/List_of_automated_urban_metro_subway_systems

Keep up, TTT.

[ToraToraTora]

the go article, if you read it honestly actually confirms a lot of what I say above. As for trains, I say right above they are feasible but they are a million miles from automated road traffic.

[ToraToraTora]

the problem is hymie, you want it badly so you'll go along with anything. In reality we will have a flirtation with this then it will be shelved for a century or so when they realise it's futile.

[Zacs-Master]

‘Digital photography is hardly comparable’

Erm.....I think you missed the point. It was the photographer’s insistence that there was no way the tech would ever be good enough which was the analogy. Not the fact it was photography.

Ring any bells captain Luddite?

[ToraToraTora]

Deepmind for the GO program uses 3 TPU's (NOT CPUs) with 135 Tflops and 1800Gb of storage, with a clever algorithm maintaining a nural net, huge storage and power much more than deep blue that beat Kasparov at chess. QED, clever algorithm huge power can give the illusion of thinking. Basically they threw processing power at it.

[ToraToraTora]

ZM: "I think you missed the point. It was the photographer’s insistence that there was no way the tech would ever be good enough which was the analogy. " - no, I think you'll find they are saying that a digital picture would never be as good as an analogue one. This is a million miles from AI.

[jim360]

I am not sure that anyone who goes around saying "there is no such thing as machine learning" is qualified to tell us who has and has not grasped the basics of computing.

I mean, at the very least, you should appreciate that the term "machine learning" refers to a very specific type of program where the software is able to improve its ability to process information given to it within a certain context. It may not be the same as "learning" for humans but it's still learning in a sense, and a very meaningful one. The AlphaZero software was able to become better than any human or other computer at Go (and Chess and Shogi) with only a single (adaptable) piece of code, as long as it was told what the rules were (and with enough computing power thrown at the resulting neural network).

I strongly suspect, all the same, that your computing ability and knowledge far outstrips mine, but to dismiss machine learning is, at the very least, an abuse of language.

[ToraToraTora]

Luddite? realist mate.

[jim360]

Also, RobNorth's BA is a bit of a non sequitur. Government spending plans are behind the cutting edge of technology all the time, either because it's impossible (or impractical) to keep up or because certain interests lobby to promote the outdated in favour of the new.

[Zacs-Master]

Tora, are you aware of the theories that human intelligence is based on simple algorithms?

https://futurism.com/intelligence-may-stem-from-a-basic-algorithm-in-the-human-brain/

This is my reason for attempting to try to get you to explain the difference between human and machine intelligence. They grow ever closer every day.

[Zacs-Master]

Here, look! A ‘coder’ admitting it:
https://www.quora.com/Does-a-human-brain-have-an-algorithm-If-the-algorithms-of-the-human-brains-are-figured-out-what-happens-Is-it-used-in-artificial-intelligence

[jim360]

I suspect that what TTT isn't aware of would fill entire libraries, to be honest.

[ToraToraTora]

"It may not be the same as "learning" for humans but it's still learning in a sense," - yes it's how a machine "learns" by remembering information and algorithms that are clever enough to utilise that information. There is no intelligence or reasoning, no "thinking"- You can write an algorithm for closed systems where the rules are known and that the opposition will obey the rules, eg GO, Draughts, Chess. Not so for infinite variables and rules that are at best interpreted differently at worst ignored at will with no consistency, as in the highway.

[ToraToraTora]

the title says "may" ZM, just another theory I'm afraid. nHave you considered starting your own thread on thinking?

[Zacs-Master]

Again.....so what is ‘thinking’? As I’ve pointed out, there are theories which say it’s exactly what you describe as the processes involved in AI.

[Zacs-Master]

Well, you seem to be coming round to at least acknowledging the importance in discerning the difference between AI and human ‘thinking’. If there is a difference!

[jim360]

Not all human drivers use much intelligence either.

I mean, it seems to me that the problem is, still, that you (and anyone else against self-driving cars) seem to be holding them to impossible standards, while deliberately overlooking the even grater failings of the human drivers they are eventually going to replace. I don't for a second mean to minimise the death of this pedestrian, but thousands of others like them died last year when self-driving cars weren't even involved, and I am not sure why this isn't relevant.

If you can introduce a measure that can reduce road deaths by even a small fraction, it seems well worth the effort. And, in time, I think it's safe to expect self-driving cars to reduce road deaths by a great deal more than that. That's what the testing is for.

There are many tests they will still fail. I'd venture to suggest that so do rather a lot of human drivers.