Waymo is in a weird place right now. They're now operating an honest-to-goodness commercial driverless taxi service. No safety drivers. No rider non-disclosure agreements. A pretty big service area (~50 square miles). But it's growing very very slowly. arstechnica.com/cars/2020/12/t…
Three years ago, I thought that if Waymo "solved" the self-driving problem first, as seemed likely, its big challenge would be scaling up quickly enough to grab territory before other companies came to market. I was wrong. arstechnica.com/cars/2017/10/w…
Waymo has driverless cars that can operating in most situations in the Phoenix suburbs. But for some reason they don't seem to be trying very hard to scale up. They haven't provided a clear answer about why not.
One possibility is that their cars are too expensive, but I don't think that's right. Costs for high-tech gear is mostly about scale, so if the service was operationally profitable the capital costs should take care of themselves.
Another possibility is that the back-end costs of operating a Waymo car is higher than the revenue. Perhaps because they have a lot of well-paid people overseeing every ride.
Waymo says that no one is steering the vehicles in real-time, but they might have someone watching the ride in real time ready to hit a kill switch or something if they see a problem. As long as this back-end support costs more than a driver, then the service won't be profitable.
A final possibility is that they're just risk-averse. They want to maintain their near-perfect safety record and they're not yet 100 percent confident its cars won't make a serious mistake.
Such caution is admirable but I don't know what the alternative is to putting cars on the road to see what happens. Maybe they feel there's more they can learn from more simulations or by exhaustively studying a small number of real-world rides.
One other possibility people suggested that I don't think is right: maybe it's really hard to expand their geofence. I see two problems with this theory: (1) I don't think they're close to saturating their current territory
(2) They've barely expanded their geofence at all in the two years since they launched Waymo One. Going from Chandler AZ to Boston might be hard but expanding into Scottsdale should be pretty easy.
Two years ago I wrote a piece comparing Waymo to Xerox PARC, the research center that invented the modern PC in the 1970s and then failed to commercialize it effectively. I think it might prove precient. arstechnica.com/cars/2019/02/g…
Update: Waymo tells me that they're now doing hundreds of rides a week. I'm pretty sure that's more than what they told me at launch in October. So maybe they're scaling up faster now.
One other possibility: maybe Waymo knows how to expand Waymo One to other parts of the Phoenix metro area but decided the economics of a suburban taxi service are so bad that it's not worth the bother until the cars can also operate in dense urban areas.
I am skeptical of this theory because even operating at a modest loss should provide them with a lot of valuable experience and data. But people tell me they are heavily testing in SF right now so maybe they decided to focus there.

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Timothy B. Lee

Timothy B. Lee Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @binarybits

29 Nov
The fact that three different companies have apparently made COVID vaccines in ~8 months makes me wonder if there's room to be a lot more ambitious about other technology projects. Like maybe we should follow the UK and ban internal combustion engines in 2030.
A lot of corporate decision-making is driven by risk aversion about future market conditions. What if your car company goes 100 percent electric and it turns out customers don't want electric cars? When that uncertainty is removed an industry can move pretty fast.
The de facto US ban on incandescent light bulbs a decade seems like an under-appreciated model. It seems to have significantly accelerated light bulb technology, and the transition happened so smoothly that most consumers barely noticed.
Read 4 tweets
27 Nov
Robert Caro's first LBJ biography includes a passage that explains how rural electrification transformed the lives of farm families, especially women. It makes a powerful case for Robert Gordon's thesis that innovations of the last 50 years pale in comparison to what came before.
The arrival of electrification relieved farm families of several categories of back-breaking labor: washing clothes by hand, milking cows by hand, canning, hauling wood to (and tending) woodstoves. Refrigeration drastically reduced milk spoilage. Plus of course electric lighting.
Big-screen TVs and smartphones are nice but they just aren't transformational the way washing machines and electric lights were to our great grandparents.
Read 5 tweets
25 Nov
Nobody refers to Twitter as a "micro-blogging" platform any more but I think it's under-appreciated how much Twitter today fills the same niche that early blogging did.
A lot of early blog posts block-quoted a paragraph of text and then offered 1-3 paragraphs of analysis. Now we screenshot a paragraph from an article and offer 1-3 tweets of analysis.
Early bloggers spent a lot of time responding to other bloggers. Bloggers today (especially professionals) don't do that much because they're trying to maximize the readership of each post. Instead, we do short, blog-style responses here on Twitter.
Read 4 tweets
19 Nov
Nvidia has an amazing new technology that essentially uses deepfake AI technology to reduce the bandwidth needs of video calling by 10x. Full explanation of how it works here: arstechnica.com/gadgets/2020/1…
The software sends a single frame of video. Then for subsequent frames it just sends data on the positions of the subject's eyes, nose, mouth, etc—much less data than a whole frame. The receiving computer then uses a neural network to re-create the subject's face.
Our comments have a lot of hand-wringing about how this "doesn't show reality," but I think this is based on a philosophically untenable conception of reality. A conventional video isn't "reality" it's a pixel-by-pixel approximation of reality. Even more so with compression.
Read 5 tweets
30 Jul
People seem to think this is a compelling argument against antitrust enforcement but it's really not. Anyone familiar with economic history knows that new high-tech industries tend to have a lot of competitors in their early years before settling down.
There were dozens of oil companies in the 1860s, dozens of car companies in the 1900s, lots of small-scale experimentation with radio in the 1920s, etc. Then Standard Oil, Ford/GM/Chrsler, and NBC/ABC/CBS emerged and became dominant for decades.
It's relatively easy for new companies to emerge when the industry is still young and growing. New customers who don't yet have established brand loyalties. Untapped innovations for a new company to discover and exploit. It gets harder as the industry matures.
Read 10 tweets
21 Jun
I made a scatterplot comparing the change in coronavirus cases over the last two weeks to cooling degree days—a proxy for air conditioning use. (Thanks @jeremy_gibbs!) The hottest states have suffered the worst outbreaks.
I think this effect explains a lot of the partisan divergence in this chart. The hottest states are all Republican. Still, at any given temperature Democratic states seem to doing modestly better, on average, than Republican states.
Lest you think the recent increase in southern states is an artifact of testing: a scatterplot of the positive coronavirus test rate shows a similar relationship. The hottest states are seeing coronavirus infections outstrip testing capacity.
Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!