New Waymo data shows superhuman safety -- they're ready

Topic: 
Tags: 
Waymo collision detail.

In what is perhaps the robocar story of the year, Waymo has released a detailed safety report which shows 6.1 million miles of driving with no at-fault accidents and even a low number of not-at-fault ones. It is now past time for them to deploy a real service. In addition, this throws down the gauntlet at all other companies to be transparent with data.

Read about it in Waymo Data Shows Superhuman Safety Record. They Should Deploy Today

Comments

Your report on Waymo's newly released performance data is really helpful. Can I ask your source for the human driver collision stats?

Come from a variety of reports from NHTSA, IIHA and Naturalistic driving studies, and Waymo's own research. There are 3T miles driven per year in the USA, about 6 million accidents reported to police, about 3 million injuries (from a smaller number of accidents) and 12 million accidents known to insurance companies. Naturalistic studies and Waymo's experience suggest that there are many more never-reported accidents, probably in the range of 30 million but that's harder to pin down as you can only do naturalistic sampling.

This article isn't really journalism-- you didn't try to interview the other parties involved in the crashes, and your concluding paragraph pressures *in bold* every other major player to publish a safety article.

Could you please disclose your holdings?

But they have been asked, my myself and others to disclose these numbers and they have declined. My holdings are my own business, and they are not large enough to affect my opinion. But no stock I hold is large enough to affect that. Forbes does require disclosure of potential though, so I disclose it. I like Waymo because they are the best, and this is a very widely held view. I invest in things because I think they are good, I don't say they are good because I invested in them. If you don't believe me, don't read me.

What do they gain from deploying now vs. waiting until they've built something than can be profitably deployed in many more locations first?

If market conditions were better, maybe they'd be able to deploy now and then IPO to get the funds to expand worldwide. But if that's their route they're probably better off waiting at least a few months. Hopefully in a few months we'll have a new President and a COVID vaccine.

In the meantime, they keep doing everything else, of course. But what do they gain by deploying a service without an in-car safety driver? (They save a little money representing an insignificant portion of their costs, and they risk having a catastrophic incident that destroys their reputation. Anything else?)

Deploying is expensive. Even with Google's money you don't do it "several locations at once." You also learn from deployment. And you gain a foothold in the towns you deploy in.

You don't understand the question "why now?" Why should Waymo deploy now, and not later?

They can gain essentially the same foothold in the market whether they have safety drivers in the car or not. (Probably not much. It's easy to switch taxi services.)

If Google doesn't have the money to deploy in multiple locations now, how are they going to get the money to do so later?

Either they've got a viable business plan, or they don't. I suspect they don't, but you seem to think they do.

Let's say it costs them $100 million to deploy to a new city. And let's say they can make $20 million a year per city. So, they can stay in one city for five years, and then use the profits from that city to deploy to a second one. Ten years from now they can be in four cities. Twenty years from now they can be in 16 cities. Or they can borrow $500 million and deploy to six cities. Feel free to change the numbers, but it's the same principle. Either they have a business plan that can pay a decent rate of return on their initial investment, or they don't.

(As I said, now is probably not the best time to raise capital. But in a few months it might be an okay time. If Waymo really has what you claim it has, we might see a Waymo IPO in 2021. Facebook raised $16 billion. How many cities can Waymo deploy to with $16 billion?)

I respect your focus on the fact that this technology can save lives. If Waymo has perfected it well enough to deploy, they should do so in as many cities as possible. Save as many lives as possible.

But maybe they're still in the development phase.

Deploying requires skilled humans. Yes, initial mapping takes effort. You need to learn the unique rules of the road in each town, test the car against them, confirm safety (not as much as you did for Chandler but still a fair bit) and make nice with the local government. You don't have the people to do everywhere at once, so you do it one at a time, and the teams grow and move to the next city -- and as they grow you can have 2, 3, 4 cities in development and testing at a time. And you are getting better at it.

Money helps a lot, but there is just certain amount of time that is needed that you can't buy with money -- unless you can build AGI, I guess.

All of the skills you've mentioned are ones that it'd be easy to find in locals. In fact, the two of the efforts that require the most expertise are the ones where you most certainly would need to utilize the skills of people with knowledge of the state and local laws: Learning the unique rules of the road and making nice with the local governments.

Doing all of that with a single team, one town at a time (or even four at a time) would be idiotic.

But hey, maybe that's what Waymo is doing. Maybe 20 years from now they'll be in a whopping 20 cities.

Imagine if Uber had one team that went from town to town spending months convincing local governments to let them deploy there! If Waymo can't figure out how to parallelize deployment, they're in way worse shape than I think they are, and I don't think very highly of them.

Sometimes I wonder if you might intentionally be saying things like this to try to throw off Waymo's competitors.

Hiring people for mission critical safety jobs is not trivial and takes time. Hiring people for that part time is even harder, since you will let many of them go when done. It's not flipping burgers.

Of the jobs you listed, that's about the only one. But it's also one where people can easily manage teams in multiple locations at once. Will it take time to build those teams? Sure. Will it cost money? Yep. But that's the very essence of what Waymo needs to do if they've truly built the car you claim they've built.

Again, how could it be any other way? You really think Waymo can work on one city at a time and not quickly be surpassed by its competitors? C'mon, deploying to new cities is the essence of their business once they've built the basic car.

Then so is everybody else, unless the somebody else has breakthroughs the others can't do. Some hope for those breakthroughs, but they show no sign of being close to them.

As I said, one city at a time is the start, and you speed up as you go because you are increasing the number of ground teams you have that can move on to the next city. And you get smarter, better and faster. This is normal for products like this. Even Uber, which can open up a town with just 2 people and a new country with 4, still had limits on how fast it could grow in towns.

I'm not sure what you mean about limits and breakthroughs. If by limits you mean the artificially imposed constraint of only deploying in one city at a time, Waymo's direct competitors won't have that limitation. (Neither will Waymo, of course. There is nothing that prevents them from working in multiple cities in parallel. Their choices make this expensive, and maybe working in multiple cities means they are spending money in multiple cities that theyll never fully recover, but it is possible.)

I'm also not sure what you're saying when you say "one city at a time is the start." Clearly Waymo is working on multiple cities already. The question we were initially discussing (I thought) was whether it makes any sense for them to continue to not use in-car safety drivers in one of those cities once they are able to install protections from COVID spread.

That said, even a few cities at a time is not the right approach, in my opinion. I much prefer the Tesla approach of creating a general solution that's not so dependent on location. Even at four cities at a time it'd take centuries for Waymo to catch up to Tesla.

The Tesla approach is obviously good if you can make it work. Tesla is the only player to think this will work and get there first. Time will tell which was the right bet. (Tesla is making several contrarian bets. Tesla is known for contrarian bets paying off, but that's been against car OEMs. Not against companies that are their match, or superior, in fields like software development, mapping, neural networks, processor development etc.)

I guess time will make it more obvious, but I'd say we've already seen that Tesla's strategy works.

interview helps

www dot zdnet dot com/google-amp/article/ai-critic-gary-marcus-metas-lecun-is-finally-coming-around-to-the-things-i-said-years-ago/

Add new comment