[jeffreyH (OP)]

If a car leaves the road then looking at the tree ahead is not the best thing to do. It makes the tree a target to aim for. In the same way, focussing attention on the existential threat of A.I. is also making that a target to aim for. There has just been an accident with a Tesla vehicle. It crashed into a fire truck stopped at a red light. It was travelling at 60mph when it collided. This is not a good reflection on the developers of this system. How much of the failure was because of a focus on the negative?

[jeffreyH (OP)]

The wired.com take on this situation doesn't inspire confidence.
https://www.wired.com/story/tesla-autopilot-why-crash-radar/

[chris]

Quote from the article you cited above:

"Volvo's semi-autonomous system, Pilot Assist, has the same shortcoming. Say the car in front of the Volvo changes lanes or turns off the road, leaving nothing between the Volvo and a stopped car. "Pilot Assist will ignore the stationary vehicle and instead accelerate to the stored speed," Volvo's manual reads, meaning the cruise speed the driver punched in. "The driver must then intervene and apply the brakes.” In other words, your Volvo won't brake to avoid hitting a stopped car that suddenly appears up ahead. It might even accelerate towards it."

So why is this? What's the problem with determining that an object is not moving relative to you, so you should stop...?

[evan_au]

So why is this? What's the problem with determining that an object is not moving relative to you, so you should stop...?

The problem is that the wavelength of practical car radars is too long to form an actual image. You just get a lot of reflections from the things around you - the best you can do is to have a beam that projects most of the energy straight ahead, and gets the best returns from straight ahead.
- This is great for cruise control that can slow down to match the speed of the car ahead of you
- But it is always getting reflections from the road, signs, trees, poles - stationary stuff
- So to reduce this "ground clutter", the designers suppress all the signals coming from "stationary" things (apparently including a stationary fire truck, in this unfortunate incident)
See: https://en.wikipedia.org/wiki/Clutter_(radar)

In contrast, cameras use very short wavelengths (visible light) and are able to form an image; in principle, by detecting objects that are growing larger in the visual field, a camera should be able to detect imminent collisions, and put on the brakes, regardless of whether the object is stationary, moving in the same direction or the opposite direction. Just don't trust it in rain or fog.

But we aren't quite there with AI, yet...

[chris]

I'm still incredulous that systems can be built to recognise my face at an airport and let me through customs (although not, apparently, at Perth, WA, where the computer said I didn't look like my passport and sent me to see the nice lady at the desk, who blamed my haircut) but we cannot built a system capable of registering a stationary truck... I don't get it.

[evan_au]

There are several different ways to construct a radar:
- Continuous wave radar is very simple, and can measure the velocity of objects with very low power. But it can't really estimate range. If you ignore stationary objects, this kind of radar is very likely to run into a parked truck.
- Pulsed radar is more complex, with high peak power and low average power. It can measure the distance to objects, but not their speed - this needs to be extracted by comparing adjacent pulses.
- Then you can get quite complicated and have pulse-Doppler radar that can measure both speed and distance
- But to aim radar, you need a big steerable dish, or a complex phased-array antenna
- Airport radar can get away with relatively low power, and a relatively small dish by requiring every aircraft in its airspace to have a transponder which reports its position whenever interrogated. This will gradually arrive in new vehicles over the next decade (V2V or Vehicle-to-Vehicle communications).
- I am sure that all these things will become much more feasible as 5G mobile and home WiFi pushes up into > 20GHz frequencies, with high frequency electronics and flat, electronically-steerable MIMO antennas.
See: https://en.wikipedia.org/wiki/Pulse-Doppler_radar

Today's LIDAR is effectively a radar operating at optical wavelengths, so producing a narrow beam and beam steering is quite compact (at least in the most modern versions; earlier versions looked like a vertical windmill mounted on the car's roof). LIDAR can measure distance, so in theory it could detect that you are heading straight towards a parked truck. But would it also balk at heading straight towards a tunnel entrance at high speed?
https://en.wikipedia.org/wiki/Lidar

In the end, the data from these sensors needs patterns detected and extracted from it. The 4 or so layers of nerves in the retina extract an immense amount of data from the visual field, compressing it down for transmission through the optic nerve. This enormous processing power is the envy of computer vision experts!
See: https://en.wikipedia.org/wiki/Retina#Retinal_layers

[Bored chemist]

All we need to do is make the driver-less car disaster less bad than the driven car disaster.
http://www.who.int/gho/road_safety/mortality/en/

(anything else that killed a million people a year would get called a disaster)

[jeffreyH (OP)]

So killing people less efficiently is a good target? What about not killing anybody at all? It is easy to weigh risk when the consequence is impersonal. I am sure that relatives of the million dead would have their own views. Don't put unfinished technology into service because that is just nuts.

[Bored chemist]

So killing people less efficiently is a good target?

The use of the word "so" like that suggests that what follows was in some way implied by what I wrote.

How did you come to that conclusion?
What I did was point out that there's every chance that autonomous cars will soon get good enough at it to be less dangerous than human drivers.
They don't have to kill "nobody"; they have to kill less than a million people a year (for the same net mileage or pro rata).

[jeffreyH (OP)]

The fact that Tesla rely on humans to intervene when their technology is no good implies that they are OK with the possibility that human error is an acceptable part of their strategy. That has nothing to do with what you said. The problem is that accepting one bad thing over what we consider a worse bad thing is not a good way to proceed. It allows technology companies to view the deaths of people as collateral damage. I don't consider anyone I know as suitable fodder for their reckless experimentation.

[Bored chemist]

The problem is that accepting one bad thing over what we consider a worse bad thing is not a good way to proceed.

Would you like to cite examples where  replacing a bad thing with a less bad thing turned out not to be a good way to proceed?

For the record, I'm using definitions of good and bad that reflect actual quality, rather than a mistaken perception.

[jeffreyH (OP)]

The nazi's determined that shooting Jews was having a really adverse affect on their troops. To solve this they established concentration camps to make the disposal of human beings more efficient. This had less psychological impact on their troops and was beneficial to their plans. Was it a good way to proceed?

[alancalverd]

I think the goal here was efficiency and costeffectiveness (slave labour, recycling gold teeth, etc.) rather than military  psychology. All the eyewitness reports suggest that most people enjoyed killing Jews or at least thought it was a sufficiently Good Thing that they re-elected Hitler on an antisemitic platform. What really annoyed German soldiers was starving and freezing in the marshes around Stalingrad, the lack of water in Egypt, and the fact that everybody else (including their own commanders and comrades) loathed them. It is the inalienable right of squaddies to grumble.

Not that it has anything to do with the current thread, but as every effective leader has discovered throughout history, most humans are disgusting and will happily kill others if given an excuse such as religion. We even enjoy watching indiscriminate or sadistic killing of our own species - Westerns, war films and horror films are hugely popular replacements for gladiators, even though the producers take care to point out that "no animals were harmed" in the production.

I guess the advantage of driverless cars is that they can't detect the religion of the people they kill. But neither can the average islamic suicide bomber - it's the killing that counts, which makes them even more irrational than Nazis..

[Bored chemist]

The nazi's determined that shooting Jews was having a really adverse affect on their troops. To solve this they established concentration camps to make the disposal of human beings more efficient. This had less psychological impact on their troops and was beneficial to their plans. Was it a good way to proceed?

For the record, I'm using definitions of good and bad that reflect actual quality, rather than a mistaken perception.