The self-driving car that will never arrive | Hacker News Reader

The self-driving car that will never arrive

theoutline.com

10 pointsedward9 comments

I don't disagree with this article's premise, but I do disagree with it's overall diagnosis. AI's near perpetual boom/bust cycle stems from a lack of clear expectations on what AI is. AI runs the gamut from "a poorly understood pile of linear algebra" to "this will automate all work and free us from the shackles of labor". This, of course, will fuel hype cycles.

Self driving cars has a clear benefit, when all the hype shit is scraped off. Motor vehicles killed ~33,000 people in the US in 2010, and nearly 38,000 people in 2016. Accessibility without a car in large parts of the United States is extremely difficult. These are real problems that can be solved. Will they arrive with Uber, or Waymo, or whatever startup-of-the-month that promises revolution? Almost certainly not. But like any research, each failure hopefully pushes us forward.

What about the people that are currently being driven around Arizona by Waymo's cars with no person sitting in the driver's seat? Do those rides not count?

https://www.bloomberg.com/news/features/2018-07-31/inside-th...

>What about the people that are currently being driven around Arizona by Waymo's cars with no person sitting in the driver's seat? Do those rides not count?

You have to remember that Phoenix was specifically chosen as the number one place on earth that is most friendly to autonomous driving, for a lot of reasons. The problems with autonomy discussed in this article are exactly what will make the transition to operating in a place like Boston incredibly harder, if not impossible.

This article doesn't really discuss, in any reasonable depth, the problems with AI. Its primary example of self-driving car's failure to perform as promised was the Uber incident, and we all know that the Uber incident was most likely the fault of Uber's reckless engineering practices rather than some fundamental issue with self-driving cars in general.

Self-driving cars seem to work reasonable well in Phoenix. That's already miles better than this article is portraying the situation. No one who has been really paying attention would think that we're immediately going to from no self driving cars to self-driving cars that work in all situations. Getting them working in inclement weather will just be another iterative step.

I really think that once self driving proves itself on interstate highways in sunny weather areas, that the transition to other places will have to include additional technology such as beacons that get installed along the roads. Then self driving will only be turned on when those beacons are available.

Aside from a few one-off demos, those rides all have a person sitting in the driver's seat. Their first demo ride without a driver was in October 2015.

https://www.washingtonpost.com/local/trafficandcommuting/bli...

> It’s easy to forget how quickly we can overextend technology, which is so good at solving some of our problems, into a good way of solving all of our problems. But eventually the self-driving wheel turns and we realize we don’t have as much command to reduce the whole complex world to a set of yes or no answers, let alone predictions, and our grossest capitalistic dreams are thwarted yet again, but not without a cost.

If the argument is that when viewed through the lens of today's technology these are seemingly insurmountable problems, then it is a weak argument for the case that the self driving car will not arrive.

It's easy to forget that what these cars can do today is leaps and bounds ahead of what self-driving technology could do a decade back. When these problems will be solved, it will be through other breakthroughs that we are not aware of today, so this is a premature obituary for self-driving AI.

This is presuming it is actually AI and not a bunch of very smart though not intelligent algorithms. (essentially an expert system for car driving)

This is similar to how chess human players were soundly beaten using very unintelligent algorithms and learning algorithms are only more getting close with a lot of CPU/TPU effort...

Car control, routing, computer vision and 3D vision enough for driving cars well are not in essence AI-complete problems.

(compare to say making music, writing sensible text or learning complex games with hidden stage and deferred reward or exploring unknown situations)

> actually AI and not a bunch of very smart though not intelligent algorithms

That is a relatively old philosophy question that won't be resolved any time soon.

Reply on news.ycombinator.com