Great post! I overall agree, but wanted to add two pro-Waymo safety points:
One, Waymos are plausibly complementary to public transit rather than a substitute. We have some causal evidence that Ubers complement public transit, though I don't really know how settled the literature on this is: https://www.sciencedirect.com/science/article/abs/pii/S0094119018300731. It'd be fair to extrapolate this to Waymos (well, at least the only time I rode a Waymo it was to get to BART =P).
Second, the existence of Waymos will probably make removing the license of a dangerous driver much less costly for that driver and hence more socially acceptable -- especially once Waymos fall below the price of an Uber. Granted, this effect should be muted for now, given that they're still (much) more expensive.
I think Waymo will follow the airline trajectory, becoming safer over time. Technology is perfectible and humans not so much. But I'm more interested in AV mass transit, such as vans that run a route or go door-to-destination. I think that form of transit will be ubiquitous and very safe, with more substantial vehicles and the need for fewer on the roads. Some will be feeders to bus-sized AVs, which offer even more advantages in terms of vehicle reduction and passenger safety. An AV transit system can be far more flexible at eventually lower costs than today's very restricted systems, which are also quite safe. Just getting more intoxicated people off of the roads will have a salubrious impact.
Thanks for writing this up! I've written a bit about Waymo's safety record at Understanding AI and this all seems quite plausible. I especially appreciate the point about car quality, which doesn't get mentioned much but is plausibly a decent part of Waymo's safety advantage.
One small note: another way you can divvy up fatalities is by fatal crashes/number of cars involved in a crash. In that case Waymo has 0.33+0.14 = 0.47 fatal crash/# of cars because it was involved in a three car crash and a seven crash crash. This is a _little_ bit less than the human number -- Phil Koopman estimates that humans would be at 0.6-0.7 at about the same mileage.
But that's really just a nitpick: you're totally right that significance hasn't happened yet.
I mostly agree, though there is some evidence that they induce additional demand, which could offset this (since they are less safe than public transit). But, I’m generally pro Waymo!
Really awesome article and thank you for spending the time to layout your arguments and your thinking. It’s so nice to read both a well argued position written with honesty about the gaps and nuance. Thank you!
Half way through the article I was thinking about the replacement of alternative transport like ride sharing and then you nicely brought it in as well.
From the perspective of public-transport-heavy cities like Singapore and Tokyo, I don’t see AV adding much value. Most modern trains have been autonomous or semi autonomous for a long time. I find the obsession with AV to be more an affliction of American and similar we-can’t-build-anything thinking so let’s just make cars into public transportation.
I also don’t like how many arguments for the benefits of AVs assume a future where 100% of cars are AVs. Yet that is an incredibly naive and simplistic view of how the world actually works. Even if today, by draconian mandate, all new cars were forced to be AVs, it would take 20 years or longer for every car on the road to become an AV. And it will be horrifically uneven with old and new AVs mixed in with old non-AVs. But that future is simply impossible. For some of the same reasons that public transportation isn’t funded in the USA, no group will agree to give up their cars. People do enjoy driving among many other reasons.
So like you have argued, we are much more likely to have a future where there are AVs in certain places, ride share, self driving and everything inbetween.
If any company, politician or citizen of any city wants to seriously address transport safety, the answer is very clear - invest more in public transportation, and strict enforcement of safe driving laws and cycling/pedestrian safety engineering. AVs solving all transport safety is another distracted wet-dream of techbros. Yes it’s another possibility and option to add to the transport mix. No it’s not going to solve any problems that we have today - it’s just another alternative with some pros/cons.
Great post! I overall agree, but wanted to add two pro-Waymo safety points:
One, Waymos are plausibly complementary to public transit rather than a substitute. We have some causal evidence that Ubers complement public transit, though I don't really know how settled the literature on this is: https://www.sciencedirect.com/science/article/abs/pii/S0094119018300731. It'd be fair to extrapolate this to Waymos (well, at least the only time I rode a Waymo it was to get to BART =P).
Second, the existence of Waymos will probably make removing the license of a dangerous driver much less costly for that driver and hence more socially acceptable -- especially once Waymos fall below the price of an Uber. Granted, this effect should be muted for now, given that they're still (much) more expensive.
I think Waymo will follow the airline trajectory, becoming safer over time. Technology is perfectible and humans not so much. But I'm more interested in AV mass transit, such as vans that run a route or go door-to-destination. I think that form of transit will be ubiquitous and very safe, with more substantial vehicles and the need for fewer on the roads. Some will be feeders to bus-sized AVs, which offer even more advantages in terms of vehicle reduction and passenger safety. An AV transit system can be far more flexible at eventually lower costs than today's very restricted systems, which are also quite safe. Just getting more intoxicated people off of the roads will have a salubrious impact.
Thanks for writing this up! I've written a bit about Waymo's safety record at Understanding AI and this all seems quite plausible. I especially appreciate the point about car quality, which doesn't get mentioned much but is plausibly a decent part of Waymo's safety advantage.
One small note: another way you can divvy up fatalities is by fatal crashes/number of cars involved in a crash. In that case Waymo has 0.33+0.14 = 0.47 fatal crash/# of cars because it was involved in a three car crash and a seven crash crash. This is a _little_ bit less than the human number -- Phil Koopman estimates that humans would be at 0.6-0.7 at about the same mileage.
But that's really just a nitpick: you're totally right that significance hasn't happened yet.
Thanks! That’s a useful way to do the fatality attribution. I did see one effort to look at how modern cars do against Waymos: https://storage.googleapis.com/waymo-uploads/files/documents/safety/Comparison%20of%20Waymo%20and%20Human-Driven%20Vehicles%20at%2025M%20miles.pdf . I don’t think it’s amazing methodologically, as it includes higher speed roads Waymos don’t drive on and makes some other odd comparisons, but at face value, it suggests that using modern cars explains ~20% of Waymo’s safety advantage.
Even if they only made the rides they replaced 1.05 times safer, there would still be a moral imperative to roll them out faster.
I mostly agree, though there is some evidence that they induce additional demand, which could offset this (since they are less safe than public transit). But, I’m generally pro Waymo!
Really awesome article and thank you for spending the time to layout your arguments and your thinking. It’s so nice to read both a well argued position written with honesty about the gaps and nuance. Thank you!
Half way through the article I was thinking about the replacement of alternative transport like ride sharing and then you nicely brought it in as well.
From the perspective of public-transport-heavy cities like Singapore and Tokyo, I don’t see AV adding much value. Most modern trains have been autonomous or semi autonomous for a long time. I find the obsession with AV to be more an affliction of American and similar we-can’t-build-anything thinking so let’s just make cars into public transportation.
I also don’t like how many arguments for the benefits of AVs assume a future where 100% of cars are AVs. Yet that is an incredibly naive and simplistic view of how the world actually works. Even if today, by draconian mandate, all new cars were forced to be AVs, it would take 20 years or longer for every car on the road to become an AV. And it will be horrifically uneven with old and new AVs mixed in with old non-AVs. But that future is simply impossible. For some of the same reasons that public transportation isn’t funded in the USA, no group will agree to give up their cars. People do enjoy driving among many other reasons.
So like you have argued, we are much more likely to have a future where there are AVs in certain places, ride share, self driving and everything inbetween.
If any company, politician or citizen of any city wants to seriously address transport safety, the answer is very clear - invest more in public transportation, and strict enforcement of safe driving laws and cycling/pedestrian safety engineering. AVs solving all transport safety is another distracted wet-dream of techbros. Yes it’s another possibility and option to add to the transport mix. No it’s not going to solve any problems that we have today - it’s just another alternative with some pros/cons.
“Will cities underinvest in public transit because Waymos and other AVs are successful?”
Cities largely have divested from policing public transit which is likely a greater threat than Waymo taxis. At least in the short term.