Comment by ClumsyPilot | Hacker News Reader

Comment by ClumsyPilot | Hacker News Reader

Parent Full thread

I am appalled at the way we are attempting to introduce driverless cars.

Cities install traffic lights with sensors and an internet connection so that it can be controlled from the city traffic management HQ. There are dozens of professionals in front of giant screens, and a supercomputer at their disposal. Could that information be sent directly to cars?

No, instead we cram a computer and a camera into every car and hope it can find the right pixels on the picture in fog and rain.

Smart approach would be to invest in infrastructure - standard location beacons that tell cars exactly where they are and what is ahead, wireless communication between cars so they know locations and bearing of every vehicle within a square mile, wireless tags could be installed on all bikes and cycles, etc. We have an entire department for traffic management in each city that could suggest a route to every car?

We already installed speed cameras, number plate recognition, pedestrian counting, we have vehicle sensing sensors inside the road that count and weigh cars, and none of this information is being used. If cloud computing makes sence for crappy apps, then it definitely makes sence to assist cars.

What do you do when something falls on the road? Should everyone walk around with a wireless tag if they don't want to be run over by a car?

A centrally automated system could — and certainly would — be backed up by per-vehicle AI. Exactly the same way that self-driven cars are backed up by human intelligence.

(Note, I’m not advocating for either approach. I think personal automobiles should be banned, in fact. But I’m at least willing to be honest about how such systems would obviously be designed.)

Obviously not. In such a case, a much better option would be to fall back to the camera system or, worst case, force the car to go into manual mode with a human driver if it detects that nearby systems are unresponsive

Then the fallback must be so good that you don’t need primary system in the first place.

No, that would not be a smart approach. Then you can only drive where beacons are installed.

If you have no problem with humans driving by hoping they find the right pixels on their retinas, why would you worry about AI doing the same? If anything, AI has a perfect reaction, is never tired or sleepy or unattentive. It just needs to be taught properly.

> Then you can only drive where beacons are installed

Uh no, then you can only use automated driving systems where beacons are installed.

"It just needs to be taught properly."

So just like all miracle solutions, .i.e. communism, every time we tried it, we just weren't doing it right.

Technology, not miracle. It's being actively worked on, and there's an undeniable progress. YouTube has plenty of videos of Tesla's self driving system and its progress over the years is very obvious.

No, the problem is that AI systems are unpredictable, unverifiable, and their correctness is unprovable. There is a million academic articles detailing this issue.

When we need something that works 99.99999% of the time, it is written in ADA or similar safe language and tested exhaustively. You know how long processing will take to the microsecond, you use a special realtime OS, etc.

AI will have its own failings, some of them totally unpredictable, so yes, boiling it down to 'just needs to be trained properly' is terrifying when peoples lives are at stake

Look at this attack, and tell me how is this goonf to be avoided with 'proper training' https://www.theregister.com/2018/01/11/ai_adversarial_attack...

> the problem is that AI systems are unpredictable, unverifiable, and their correctness is unprovable.

Same can be said about humans.

None of what you listed matters in the real world. Self-driving systems just have to be better than average humans, on average, to be useful.

> boiling it down to 'just needs to be trained properly' is terrifying when peoples lives are at stake

You have no problem with humans being trained to drive. 38.6 thousand people died in car crashes in the US alone last year.

An AI which “just needs to be taught properly” is the same as a genie or magic pixie which also exist but for “just needing to be taught properly”

Transportation infrastructure investment is fairly zero-sum. A city full of self-driving cars is only marginally better than a city full of regular cars. The ‘smart approach’ is to reduce car friendliness in cities and rather invest in public transit and walk/bikeability.

The issue with this is that it requires such high capital investment before the first driverless car can be introduced.

What you are describing is actually probably the end game - but I think musk and others have correctly identified that if you wait for infrastructure to be ready then you will be waiting a long time, and if you want to see self driving cars anytime soon you have to assume current infra.

If this is where you want to go we could just install souped up train tracks instead.

The centralized approach you describe would work over less than a few percent of the land area of the USA, namely big city cores.

Those are also, incidentally, the hardest places for autos to exist in.

It's much, much easier to navigate a lonely highway in the middle of nowhere without much traffic than it is to navigate the perilous labyrinth of the suburbs and downtown traffic.

It can also be done much more simply by laying down copper in the road and making it a part of infrastructure so that the cars can sense it and figure out that it's the road. Stop signs, traffic lights, and interchanges can be marked using changes that the car can detect. The rest can be collision avoidance.

Ultimately I think the issue is that the government would have to take the lead and deploy the infrastructure to make self drive work better. That US government barely does infrastructure anymore. Nobody wants to take that lead, take the political risk, or pay for it. That leaves private sector cowboys to innovate with no help.

The US Government is trying to do infrastructure right now. To the tune of $3.5Tn over 10 years to make up for the lack of investment over decades. And it includes charging infrastructure and more. The budget has been balanced and paid for, thereby not impacting the deficit.

> Biden is proposing a $174 billion investment into electrifying cars and trucks in the United States. It would establish grant and incentive programs for state and local governments and the private sector to build a national network of 500,000 electric vehicle charging stations by 2030.

https://www.nytimes.com/2021/03/31/business/biden-electric-v...

America's infrastructure in such bad shape that there's a good chance that the arteries of US commerce will start having an increase in catastrophic failures leading to death.

But it's being pushed back against, because well, reasons. This spending, to transform America's degrading infrastructure, will come out to $350Bn/yr. The Middle East & Afghanistan wars come out to, $355Bn/yr at - $6.4 trillion over ~18 years. https://www.cnbc.com/2019/11/20/us-spent-6point4-trillion-on... . The tax cuts under the last administration come out to $200Bn/yr. Neither of these - the wars and the tax cuts - have been balanced and are coming out of America's ballooning deficit.

If you would like the Government to do more of this, please call your representatives and vote for people who invest money towards a better future for everyone.

And if anything, governments are in a privileged position to create and enforce these standards across multiple large regions.

Reply on news.ycombinator.com