autonomous cars

dannyrst

If hackers in your main concern with driverless cars, you need not worry.

As shown above, you can already hack into cars and take over the brakes. If I were someone who wanted to hack a system, I'd certainly not waste my time hacking a few cars, causing some crashes and getting nothing from it (other than sick thrills).

It wasn't that long ago that the news was full of stories of someone taking control of a plane from a passenger seat using the entertainment system.

victor2

I'd be more concerned about a car being driven by Microsoft software than hackers. Gives Blue Screen of Death a whole new meaning...;)

almillar

What about when something malfunctions on the car and you'd need to take over to avoid an accident.

There are multiple explosions happening in front of you every second when you're driving, and you're likely to have gallons of flammable fuel just behind you. And loads of tech in between.
What if some of that goes wrong?

Or a situation develops that a human driver is capable of anticipating and avoiding, but the self driving car isn't

That's why they're being tested so thoroughly.

Seriously, have a look around the roads. People are so lazy they can't be bothered turning their neck around, looking in a mirror, or even looking at the green light in front of them. Autonomous cars are already better than the standard of many drivers.

DominicH

Humans, even lazy ones, are still way better at handling novel situations than AI. Unexpected massive pothole in the road? Humans will usually see it for what it is and take avoiding action. AI might not see it, or be unable to tell if it's just a shadow or dark patch of road, and plough straight into it. There are countless other situations that will catch AI out. Really, it all comes back to the basic challenge of these kind of AI systems - endowing them with "common sense", which we haven't even come close to yet.

I know that the Google cars have done a lot of miles, but almost all in a relatively small area that they have mapped in enormous detail (way beyond Google Maps level of detail).

Tobster86

DominicH wrote: »

Humans, even lazy ones, are still way better at handling novel situations than AI. Unexpected massive pothole in the road? Humans will usually see it for what it is and take avoiding action. AI might not see it, or be unable to tell if it's just a shadow or dark patch of road, and plough straight into it. There are countless other situations that will catch AI out. Really, it all comes back to the basic challenge of these kind of AI systems - endowing them with "common sense", which we haven't even come close to yet.

I don't buy any of that; as a completely normal driver who, like everyone else, has smashed through many an unexpected pothole resulting in some blasphemy despite my best intentions. Yes, many drivers will claim to have never hit a pothole in their life and that the sun shines out of their !!!!. Whatever.

As for common sense? There are many a video on the internet of utterly poor, and in some cases, non-existent reactions to unexpected situations from drivers despite the options available to them to mitigate the circumstances. I'd say this is far more common that someone taking appropriate action.

Facing an absolutely dire situation, for example a HGV crashing over the central reservation onto a head-on collision course; an AI control system would unprejudicially do absolutely everything it could to avoid a collision; making precise calculations and reassessments at microsecond intervals all along the way. Many drivers would simply scream and shut their eyes, and those that didn't would probably not have complete control during their evasive makeovers, or react as quickly.

bazster

The only damage-causing pothole I ever hit was hiding underneath a puddle. I'd like to think that an AI system wouldn't have gone speeding through it with the same cavalier disregard for what might lurk beneath that I exhibited that day.

DominicH

Tobster86 wrote: »

I don't buy any of that; as a completely normal driver who, like everyone else, has smashed through many an unexpected pothole resulting in some blasphemy despite my best intentions. Yes, many drivers will claim to have never hit a pothole in their life and that the sun shines out of their !!!!. Whatever.

As for common sense? There are many a video on the internet of utterly poor, and in some cases, non-existent reactions to unexpected situations from drivers despite the options available to them to mitigate the circumstances. I'd say this is far more common that someone taking appropriate action.

But there are many other situations that we take it for granted that a human can handle, but which are very difficult for AI. For example, you're driving along and see a policeman at the side of the road flagging cars down, trying to stop traffic. A human can see that it's a policeman, and stop. Current AI would, at best, maybe recognise that there was a person there. Maybe. It's not going to see the policeman mouthing the word "stop".

Any changes to the terrain, such as lane closures or unexpected obstacles, are problematic. AI has to fall back on its general purpose terrain handling, which is far inferior to how it handles known, mapped terrain.

Parking is another thing that is difficult. Car parks are inconsistently laid out and have lots of hazards, and usually require a fair amount of written signage and directions to help human drivers. AI is bad at understanding writing and language - another thing that falls under "common sense", which shouldn't be conflated with "being sensible".

colino

You seem to be making the mistake that people who develop these things just replicate human behaviour and add a bit of fuzzy logic. Not only can they be safer, less prone to accidents and focused, but they can be much more economical too. This isn't the only thread on the subject where people appear to be dying to get the man waving the red flag back in front of cars.

Tobster86

colino wrote: »

You seem to be making the mistake that people who develop these things just replicate human behaviour and add a bit of fuzzy logic.

This is about the most relevant point.

Autonomous vehicles aren't just about replacing a human driver with a computer that does exactly the same stuff. The whole concept of transport, and society itself, will change to embrace it.

Motorway drives are long, boring and tedious. My answer to that is a big, leather-seated car with a thirsty V8 to get the whole endeavour over and done with as undisruptively as possible; because my time is precious. Put a computer in charge of that, and suddenly my time becomes my own again.

You reclaim that time back, because you can work, drink, skype, sleep, or generally entertain yourself (let's not get too imaginative there!) without the burden of responsibility of driving; so the fact that you're in transit is much less relevant. In this circumstance, the only thing I'd need to keep would be the leather seats. The car would otherwise need the quietest, most economical engine and to potter along at about 50mph.

All that said, the engineering; size, shape, drivetrain, just about every aspect of vehicles that are tuned to human usage would change. At the moment, most drivers !!!!! about things as frivolously different as CVT transmission.

welfayre

I don't think it'll happen for a long long time.

Aside from all the technological and insurance issues you have the whole moral conundrum of how an AI will decide the value of life. Presumably they'll be programmed to avoid injury to the passenger of the vehicle but say there's going to be a head on crash that will likey result in the death of the passenger. The only way to avoid it is to swerve onto the pavement but in doing so it will hit and kill a child, how will it decide who to kill?