Hacker News Reader

Comment by georgemcbay | Hacker News Reader

I'm fully convinced self-driving cars will be sold commercially within the next 10 years. However, it is going to be a very long time until they are allowed on the road sans human (even if they are self-driven 99% of the time, I'm positive legislation will require a human behind the wheel at all times and car makers will have to attempt to enforce this technically), which puts a serious dent in your taxi plan.

You're right long term, but self-driving cars that are allowed to drive without any human behind the wheel is more like a 50 years out thing, IMO. Part of this is just cultural and political, and part of it is due to real technical challenges (there are still going to be extraordinary driving situations in which the obvious thing to do would be clear to any reasonable human but not an autonomous car without serious general purpose AI).

View on HN

>You're right long term, but self-driving cars that are allowed to drive without any human behind the wheel is more like a 50 years out thing, IMO.

Why not just outsource the driver to, say Eastern Europe, Mexico or India? Rent out a room, buy bunch of computers, screens, tables, chairs, and then hire a bunch of "virtual drivers" with enough experience on the road to control those vehicles remotely. Furthermore, program the car to stop on the side of a road in case should it lose the wireless connectivity. I guess the ping could be a problem, but haven't most of the people driving on the roads today got slower reactions to events than 100-150ms? Later on, as the tech progresses, teach the software to direct the driving to that virtual driver on hold only when the "AI" wasn't be able to figure out the current situation.

"there are still going to be extraordinary driving situations in which the obvious thing to do would be clear to any reasonable human but not an autonomous car without serious general purpose AI"

I believe that any such situation that could not be dealt with by simply coming to a safe halt is currently very badly handled by humans. I believe that they could be better handled with the _current_ state of the art, let alone advances we see over the next few decades.

I'm not arguing that they could be perfectly handled; just that the machines can do better than the average meat-bag.

Even simply coming to a safe halt is problematic. What if some troll traps a self-driving car with a few empty cardboard boxes?

It stays still until someone removes them. If it's a cab it's more than likely to simply call for help.

And while it is still, it is now an obstacle for other empty self driving cars. Soon the troll has blocked an entire intersection or freeway using a combination of empty boxes and other trapped self-driving cars.

You underestimate the effort a good troll would be willing to put into this.

There are lots of simple things that one could do to a roadway to impede or endanger traffic. Consider, for example, the danger of leaving something sharp or slick on a tight turn where people often speed.

I've thought about this for a few years, and as near as I can tell it's either much harder than it seems, or there are fewer vandals than I'd have expected.

the real test of driverless cars comes shortly after they start to become widely used. A driverless car will be forced into a trolley problem situation. That is, it will have to choose between hitting two things without the option to hit neither. Expect extreme scrutiny in the ensuing legal case.

At least we'll be able to reliably determine the reason, and fix it if necessary, rather than relying on human testimony.

Reply on news.ycombinator.com