Hacker News Reader

Comment by natch | Hacker News Reader

I think most people would agree that the middle zone is not ideal for most drivers.

However, there may be no other way to someday reach level 5 for general driving, other than to live with that middle zone for a while (keeping in mind that it’s already safer than humans driving alone) in order to conduct the training needed to reach level 5.

You don’t just wish, or regulate, level 5 into existence. You need to build toward it by having real systems deployed in the real world. Those real systems are not going to be level 5 at first or even level 3. This is where we are today, for general driving we only have level 2 and some level 3 if that.

Training is needed not only for the systems but also for drivers (including other drivers in traditional vehicles, they are also being trained unwittingly) and pedestrians, and regulators. Everyone is learning here. It’s not just an overnight, snap of the fingers kind of thing.

View on HN

> However, there may be no other way to someday reach level 5 for general driving, other than to live with that middle zone for a while (keeping in mind that it’s already safer than humans driving alone) in order to conduct the training needed to reach level 5.

Gotta call bullshit on this. If human-piloted cars were bedecked in sensors, recording drivers' inputs and outputs, we could collect all the training we could need for L5 driving. It would take a ton of stringent curation, but that's going to be necessary anyway. It's not like Tesla has magically learned not to hit emergency vehicles stopped in traffic yet.

Or, alternatively, we can slowly make roads safe for any AI driven cars. Use appropriate marking on the roads everywhere, clearly visible signs, clearly visible crosswalks, physically separated lanes for bicycles and vehicle similar to bicycles etc. etc. etc.

Funny thing that we will simultaneously make roads safer for all cars and pedestrians and cyclists too.

It doesn't snow where you live, does it? And your proposal doesn't remotely address the "emergency vehicles stopped in traffic" issue.

Yesterday I was on a two lane country road where I saw a man standing on the opposite side of the road holding up a bright orange baseball cap in his hand and his other hand in a "stop" motion. I slowed down enough to where I didn't hit the crashed car which was just around the bend, but I may have hit it without the advance notice because the turn had very poor visibility. If I were an AI driver, I probably would have considered the man "not-an-obstacle" and promptly ignored him and hit the other car. After all, I'm trained to slow down when I see people in hi-vis with proper "slow traffic" signs. Some random person whose neighbor crashed and is waving the brightest color thing they have is not in the training data.

Or in other words, there are signals humans can send to other humans which can be understood even if you haven't seen that particular signal before, and those will be the biggest challenge for an AI driver, at least if it resembles the current generation of AI.

Fleet gathering of edge case data is a thing, and it helps with many cases. But of course you’re right some general intelligence is needed.

One other huge disadvantage of L3. As driver physically drive less and less while models learn, their skills will wither and when the car gives up control it will be to out of practice drivers.

I believe the article speaks to the lack of data: we don’t know if L3 is safer. Safer than a drunk driver or a tired long haul, maybe. But a typical driver that would also use L3 responsibly, it may not be safer.

This training in the wild sounds scary at first, until i remembered that we've done something similar with air travel, basically "training" with live people and fatalities. But over the years we did reduce those from 40 crashes/mil to 0.1 crashes per million flights. I would not want to fly on those early jets in the 50's, but enough people actually took that chance!

40 crashes per million flights has still got to be better than current road safety, at a very rough calculation the US for example had 128 road deaths per million people, which I'm admittedly not sure how to convert to make a fair comparison.

Reply on news.ycombinator.com