Autonomous Cars Are Not Imminent | Hacker News Reader

Autonomous Cars Are Not Imminent

recode.net

This article is pretty bad. There's no reason that automous vehicles have to be linked to electric vehcules, it just happens to be the case that Tesla is pushing both. The Lexus cars that Waymo uses are not electric for example.

Automous also cars don't need communication with other cars to hit the market. It'd be nice for faster driving, but it's in no way a requirement.

Hacking is a real concern, but I don't think it will stop adoption. It's also essentially unsolvable, companies will just need to have sufficiently good security measures that it occurs rarely enough to make automous cars safer than non-automous.

I think there are roughly two camps about autonomous cars: the people who think that a car that's 70% as good as a human, but significantly better for that 70%, will be good enough in the beginning, and the people who think that an autonomous car won't be okay until it's as good at a human in every aspect, e.g. requiring an AGI.

Both camps have valid arguments, but I think that the safer option is the one where humans don't need to think or drive at all, essentially not requiring a steering wheel at all. I don't think the Tesla model is so good because if you let a driver essentially not worry about what's going on for 95% of the time, and then something needs their attention quickly, their reaction time will be significantly slower because they're not primed for it, unlike during a normal driving session. You can argue that humans can be trained for this, but unlike pilots pretty much anyone can get a driving license and it's not difficult: for example, in Florida, many municipalities just have you do some simple stuff in a parking lot with no real world, as in not in an actual road, driving test. For those people you can't expect them to pay attention and allow a car to drive, the human brain just doesn't work like that. This argument doesn't even talk about the things humans don't even realize they perceive, like the intention of other humans in the area, that is really, really difficult for a computer to understand, another huge barrier.

In my mind there will be two areas that self driving cars will impact: long haul trucks, and regular car features. Truckers are trained pretty well and I could see a semi-automated truck work for them, especially if they were trained a bit more on the specifics of the automated tech. Regular cars will likely get an improvement as well, maybe a better version of cruise control that takes care of stop-and-go traffic better, or something that can sense accidents further away and warn the driver to take an alternative route before even being near the accident. These would have real, and positive, impacts but won't be truly self driving.

There is a third camp.

The ones that think cars need to be better than humans and that one death in a self driving car is too many even if there would have been thousands of deaths by human drivers in the same time.

The amount of time spent debating issues is huge. Like how a computer would handle situations that are unique to human drivers (for example, not leaving enough room to stop because you're driving too fast and following too close) or things that a human driver would perform poorly on too (should I drive off the bridge or run over those children?).

It's fine to debate issues but to use them as "show stopper" by default is irresponsible.

There is a serious concern to keep in mind about adaptive cruise control or automatic braking, which is, their inelegant failure modes. For a simpler example, consider that new cars have started including blind spot sensors, which light up an indicator on the side-view mirrors if a car might be in the corresponding blind spot. The thing that baffles me about these is, if that system fails, then the indicator will be off regardless of whether there's anyone in the blind spot, which is exactly the wrong thing to do.

Likewise, if adaptive cruise control fails, you'll probably rear-end someone. So then you get into the same problem, that there's a rare situation in which the failure is potentially fatal. So then you have to pay attention for something that's unlikely, which we're pretty bad at, so what value is added?

Naysayers are a dime a dozen. What's valuable: someone who can see how to make it happen.

Expected change takes longer than expected. And unexpected change happens before you know it.

This article is one long non sequitur. I learned absolutely nothing.

Reply on news.ycombinator.com