Hacker News Reader

Comment by jacquesm | Hacker News Reader

I really respect your skills as a hacker but it is actually quite scary that you would not understand that DMV issuing these permits is actually a strong sign of progress, and that 'user videos on YouTube' would somehow be the yardstick by which you believe we should measure such progress.

Does the same hold for the FAA? The FDA?

View on HN

It depends what your opinions of those agencies are, and if you believe they have technical competence or not. After the 737 MAX fiasco, I've begun to seriously doubt it.

"you would not understand that DMV issuing these permits is actually a strong sign of progress" explain to me why you believe this. Videos on YouTube convince me something is real and exists. A government press release convinces me that some process was followed, but perhaps one having nothing to do with the solution to the problem.

Self driving is almost entirely a software problem. Testing that software is equally as complex. Do you believe the CA DMV has an army of good software engineers? Data scientists? If not, how did they evaluate it?

Assuming your argument is that all you need is cameras, at what point does it become entirely a software problem? Once you have cameras that cover a sufficient FoV around the car? Or can pivot so that they can, like human eyes? How do you know that the cameras you have meet that requirement?

Even then, I'm not sure how you can say its always just a software problem when no one has accomplished it yet.

Agreed that the 737 MAX did not help, but this is also a failure of all of the other regulatory agencies the world over that rubber stamped the FAA approval instead of doing their own. It certainly looks as though this served as a wake up call - it should! - and that a re-occurrence is relatively unlikely in the near future (until everybody becomes complacent again).

Self driving may be a software problem, but that does not necessarily mean that we have the means to make that software a reality today. Given that fact that we know that intelligent beings make plenty of errors and that they are much smarter than your average self driving system, the enormous range of conditions and inputs and the relatively limited amount of computing power that can be thrown at this problem I highly doubt that it is a 'mere matter of software engineering' at this point, taking into account the state of the art.

Agreed that testing it is super complex, in fact, I think that testing is the key: without a fully representative test equivalent to a driving test none of this software should be on the street. That's what we agreed upon: pass the driving test and you are allowed to drive. American driving tests are ridiculously simple, so it should be easy to pass such a test and yet that hasn't even been made a mandatory requirement for certifying a self driving solution. Until then I don't think any of this stuff should be on public roads.

I don't think the CA DMV needs an army of sofware engineers or data scientists, they are first and foremost a regulatory body that can hire external experts to give them guidance, and I suspect that they genuinely believe that self driving systems from these manufacturers are now solid enough to be given wider latitude during this experimental phase. I personally disagree with the decision, but I do think it is progress, that a regulatory body would dare to attach their name to such a decision rather than to play it safe, this is a clear shift in perception. Time will tell whether or not it was the right decision, if it turns out it wasn't there will be a bigger set-back than this was a step forward.

I don't think it's necessary to believe that the DMV's evaluation means anything about the quality of self-driving (and I don't think it means much).

The fact that they're permitting it means it will be happening in practice, which will generate lots of data points, which (a) will help the companies improve their software and (b) assuming it's not a disaster, will get people used to seeing self-driving cars that really work, which (from a practical-politics perspective) will make it much easier to get permission from other regional governments, and to expand the permits.

I'm nobody and I have huge respect for both of you. I do think some transparency would be nice in regards to how they made the decision and with how much data.

The shortcomings of Comma.AI and Tesla FSD are very clear due to YouTube. If Cruise has a much safer system, that is great, I would just like to see more data.

The difference between FAA/FDA and DMV is that they have subject experts whose jobs is to test and evaluate the technology.

DMV consists of bureaucrats that seemingly just rubber stamp the permits hoping for the best.

In that context, people actually using the technology, testing and evaluating its performance are like FAA/FDA and not DMV.

Subject experts can be hired, and I am assuming there are close contacts between DMV and the likes of NHTSA.

https://www.nhtsa.gov/technology-innovation/automated-vehicl...

I mean, if you just look at the offices the DMV works out of, they look like jails. Classic white painted brick (wtf, who paints brick except for jails) and no windows.

You can't expect the people working there to not be driven to insanity.

Reply on news.ycombinator.com