Don’t forget that newer cars can log data as the human drives in situations where the AI cannot perform; it won’t take an AI long to use that data as a template for it’s own behaviors. In other words, show an AI how to do something once, and you never have to do it again. AND it will be able to extrapolate when new situations occur now that it has a database to work from.
Let’s look at this in the light of my previous post.
Let’s say that 100 people are using Waymo Autos. In certain situations the AI relinquishes control to the human. So the first narrow alley or dirt road turnoff that I, one of the humans, encounter the AI hands control to me. I take control, execute the maneuver and arrive at my destination.
The next time I go to that same destination, do you think the AI will be able to handle it, or do you think the AI will need me to take over?
The answer is that the AI will handle it, most likely using my own previous actions to base it’s decisions on. So I’ll only have to take over for most static situations once, and then the AI will handle it.
Now let’s suppose that you arrive at the same thing that caused my AI to hand control over to me that first time. Do you think your vehicle will hand control over to you, as mine did, or do you think it will access the Waymo database, find the solution to the problem already exists and then execute that series of commands to safely navigate the obstacle?
If the latter, each problem will only need to be encountered and overcome once by ANY vehicle before it will no longer be a problem for ALL vehicles. Imagine if teaching your 16 year old to come to a full stop at a Stop sign meant that every 16 year old everywhere learned it: that’s how AI can work.
But that’s specifically Tesla’s autopilot technology which mainly relies on cameras (also a radar and 2 ultrasonic sensors). As far as I know, this system is based purely on sensor input, not stored data & GPS. Other systems are more sophisticated. Waymo uses a LIDAR to supplement the cameras and radar detectors, and also rely heavily on stored data - the car would only go on routes that are in its database. Which means it’s not relying on the cameras to know where lanes end or merge, it already knows.
Do you not appreciate that one AlphaGo Zero costs about $25 million? These vehicles are being equipped with four of Google’s Tensor Processing Units.
Appreciating what powerful AI is capable of does not mean that anything labeled “AI” or “machine learning” is capable of it.
As for
No, the answer is clearly not. Even the same dirt road is not the same dirt road on different days in different conditions.
I will also predict with a high degree of confidence that while there will be some vehicle to vehicle (V2V) communication, likely including some shared cloud workspace for updated map data and conditions on the ground collected by units “in the wild”, the capacity to spread actual driving algorithms across the fleet will be highly constrained with many walls to prevent such from happening else catastrophic hacking or even non-malicious catastrophic failures be possible.
Functioning as a hive mind is too great a risk for hive collapse.
This is simply untrue. Off the top of my head, Volvo, Tesla, Ford, and Google have all made promises toward autonomous vehicle timelines that either disappeared, were supplanted by additional timelines, or for which the deliverable was significantly modified. And I’m sure I could find more if I went trolling through old news articles. The reality is this is proving much harder than they thought it would, and so timelines have been pushed back.
I just don’t see where the money is in this. I don’t know about Lyft, but Uber is losing money hand over fist - while suckering their drivers into providing the cars.
Also, there’s the regulatory environment that taxicabs used to exist in, that Uber and Lyft disrupted. Basically, if you have unlimited cabs out there, everybody exists at the break-even point. (And changing it to a ride-hailing app doesn’t change that.)
The way it worked, pre-Uber, was that localities would regulate the taxi business, handing out a limited number of ‘medallions’ that each authorized the holder to operate one taxi. This created enough of an artificial shortage of cabs so that operating a taxicab fleet was profitable.
Uber got around this by convincing the localities (or, where necessary, the courts) that they weren’t a taxicab provider, they were just a middleman for people who weren’t providing taxi service really. But of course they are, so now the drivers are getting along by turning the future longevity of their vehicles into today’s payday. Meanwhile the medallions of the legacy taxicab operators are worth a hell of a lot less, which plays into one of the many dramas in Michael Cohen’s life.
So what happens when the drivers go away? GM et al. won’t have the same legal argument that they’re not really providing taxicab service. If Uber and Lyft are disrupted in turn, or finally simply go belly-up, the legacy medallion-holders are still there, their medallions are worth beaucoup again, and it gets complicated.
Best case I see is that GM invests a shitload of money to buy ALL the medallions in a given locality, so that there will be no rival operators to say, “hey wait, you’re only allowed to put X cabs on the street,” because they’ll need a lot of cars out there to do what Uber is doing, but then unlike Uber, they’re owning and maintaining their own (expensive) fleet, though they don’t have to pay the drivers. They’ll add in those more traditional taxi fleet expenses, though - they’ll need places to garage the taxis during off hours, and a maintenance crew to clean the cabs and keep them running properly. Plus a more techy maintenance crew to make sure all the cameras/radar/lidar/etc. are working properly.
They may eventually be able to make it work just in terms of the operating costs, but if this is supposed to be the payoff for a huge investment in developing AVs, I think they’re nuts. With respect to self-driving cars, they’re eventually going to have to get back to their core business: building cars and selling them to people like us. The ride-hailing business may be a good test bed to work out the kinks in their cars in order to get to the point where they can sell AVs the way they sell regular cars now, but I don’t see its being any more than that for them.
OK, so what you’re saying is that I should interpret language about money being made available to produce a product that isn’t in existence yet and whose workability is still speculative, exactly the same as I’d interpret similar language about an existing product that’s manufactured and sold on a routine basis.
Well, okay then.
If they put a fleet of self-driving cars out there in 2020 or sooner and don’t have to turn around and pull them back off the road because real life is complicated, then feel free to laugh in my face then. In the meantime, I find your faith in GM to be an awesome thing.
You make some excellent points about the market problem that ride-sharing services face but I think the notion that the route to success will be returning to a regulated taxi medallion model is not realistic. For one thing, autonomously piloted vehicles will expand to serve the vast majority of people who currently drive, at least in urban and suburban areas where population density and demand is high enough. As autonomously piloted vehicles become more mature and widely used it will be evident that they are far safer than manually piloted vehicles just by dint of the all-around sensory awareness and lack of distraction or fatigue, and insurance costs to hold a license to manually drive a car increase. The market for autonomously piloted vehicles is far larger than taxi/ride-hailing market today, and will likely grow to encompass the majority of people who would otherwise drive, including doing tasks like transporting children from and to extracurricular activities, the elderly and infirm to medical services, et cetera.
However, the real advantage of autonomously piloted vehicles is in the potential for efficiency in shared use. Privately owned vehicles are only used a small fraction of the time; even most long distance commuters use their vehicle for less than 10% of the day, and the vehicle sits idle in a parking lot or driveway while depriciating in both real and perceived value. Autonomous vehicles can be utilized any time they are not being serviced/fueled/charged, not only for commuting but off-peak usage such as delivery of purchased goods or supplementary logistics, and without the cost or need for availability of a human driver. The most practical business model for fleet autnomous vehicle use is a subscription model similar to how most people pay for cell phones, where you pay a monthly fee for a certain level of service (daily commuting plus some number of trips per week) and additional fees for ancillary service or premium options such as haulage, luxury features, et cetera. If costs are comparable or less than the total costs of ownership of a private vehicle you can expect there to be a large market for a provider who can invest in a fleet, and the subscription model removes a lot of the guesswork about the market and encourages user fealty versus a hypercompetitive pricing model.
Stranger
If you want to see a company taking what I see as a sensible approach to all this, take a look at Toyota. They built a stand-apart tech development company, but they are more heavily focused on advanced driver-assistance systems (ADAS) than they are producing magical robot cars.
One thing I’d disagree with in that article, though:
There are companies out there right now collecting LIDAR data to monetize it, and LIDAR is also slowly getting cheaper…there isn’t anything to say they couldn’t buy the data and/or add LIDAR to their cars once it gets cheap enough.
I think this is the real potential, though it’s very long term. As the US urbanizes, however, more people are going to be interested in not owning cars and instead relying on something shared. I’m not sure how big the scale of that is, but it feels big enough. [suburbanites and ruralites? not so much]
The biggest impediment in my mind was mentioned upstream, which is companies like Uber, who are burning through a ton of capital ($4.5B in 2017!) and suppressing rates. At some point that has to stop, however, either Uber prices up more reasonably or the capital runs dry.
Suburban commuters ride sharing with autonomously piloted vehicles actually makes a lot of sense, especially for a subscription model where the cost is upfront. It doesn’t have the downside of being in a commuter pool where you have to wait or depend upon other pool members to be on time but permits a cost sharing, and makes it practical for a household to have only one or even no private vehicles. One of the ancillary advantages is that it not only reduces traffic congestion once autonomous vehicles become common, but with vehicles not sitting idle parked all day, it also reduces the need for vast parking lots around businesses, which is not only a wasteful use of land but also contributes substantially to the “heat island” effect and reduces footage available for greenspaces. For those living and working in urban environments, it reduces costs associated with parking. Vehicles will still need a place to park for servicing and recharging/refueling, but these can be in offsite locations using real estate that is undesireable for development without consideration for accessibility to businesses and residences. I don’t think this is really all that long term; once the technology reaches a certain level of maturity and cost-effectiveness, adoption and modification of existing transportation infrastructure is likely to be as rapid as the ubiquity of smartphones. If that technology becomes ready for prime time in, say, twenty years, it’ll likely be pervasive in twenty-five to thirty.
For rural communities shared use autonomous vehicles make less fiscal sense and would probably have to be subsidized with public money or incentives, but the advantages for public safety and convenience may be a legitimate trade. The more problematic issue is having vehicles sufficiently distributed that wait times for a ride are not unduly long. The adoption of fully autonomous vehicles may take much longer, both because there is less of an incentive for ride sharing services and because fewer people in rural areas are financially able to adopt a new technology and may be culturally unwilling to give up control to an autonomous piloting system even if it is demonstrably safer.
I’m not betting any money on Uber (or Lyft) in particular, though. For one, the company does not appear to be well-managed or have an actual coherent vision. For another, adoption will, as someone else pointed out, undercut their current contract-driver model and raise their expenses without immediately generating more revenue. And frankly, there is probably someone waiting in the wings now waiting to undercut Uber and Lyft that they aren’t even aware of. By the time the technology becomes practical they’ll be an established player in their existing business and likely won’t have the flexibility to make that kind of pivoting change.
Stranger
To my layman’s eye, it looks an awful lot like a municipal transit system, minus the drivers.
I agree that businesses are likely to try and move this to a subscription service (it is the trend right now) but I’m not sure how widespread or how popular such a program could be:
I have a very short drive to work, usually 30 minutes or less. Then another 30 minutes going home. For those two half hour periods, I do not want to be sitting in someone else’s filth. How many other people will use this subscription car when I am not using it, and are they clean, non-ill people? I’m not sure that I’d like a subscription car service any more than I’d like a “clothes-sharing” program that didn’t include pre-use (MY use) cleaning. And I don’t see how it can be economical to have every auto return for cleaning in between every use. That’s one speed bump to get over, for instance.
I do see that driving is much, much less of a priority for younger generations and I also see that these generations have much different notions of both privacy and ownership than I do, so I can see that my concerns might not be broadly shared.
Well, it would be that. It just would be privately owned and operated, even by multiple businesses and people.
Except you aren’t locked into an new infrastructure that costs hundreds of millions or billions of dollars to deploy and that can only service people on a set timetable and within walking distance of a depot. One of the failings of commuter rail in many cities like St. Louis or Charlotte is that it just doesn’t offer enough flexibility to be useful to the majority of commuters. Even in New York City, Boston, or San Francisco with high population densities plenty of people still drive despite the inconvenience because public transit doesn’t suit their needs, and as cities evolve it is often difficult to expand transit routes to support them.
Stranger
I thought he was talking about a public bus system.
Honestly, I can’t find a lot of these statements that have been walked back or ignored. Google said in 2012 it would have self-driving cars on the market in 2018. It won’t but it seems to be close to offering a commercial self-driving car service this year. It’s already offering a limited experimental self-driving car service in Arizona. GM said it would be the first to produce self-driving prototypes on an assembly line. It did. In 2014, Audi said it would offer a Level 3 self-driving car with its 2018 Audi A8, which includes "Traffic Jam Pilot."It did too.
Most of the aggressive targets are for 2021. Audi has promised that its Traffic Jam Pilot will work up to highway speeds. BMW, Ford, and Volvo have promised self-driving cars on the market or in fleet service available to the public by the end of 2021. Daimler said that self-driving cars will be on the market by 2021, without saying it will be their cars.
A few companies are walking back their predictions. NuTonomy (now owned by Aptiv) said in mid-2016 it would have a self-driving taxi service available in Singapore by 2018. Now, they are saying it won’t be until 2021, so their self-driving taxi is now further away than it was two years ago. They are the exception though in walking back their predictions so strongly. Tesla has also completely blown its deadlines, but it does that with all its self-imposed deadlines, including those having absolutely nothing to do with self-driving cars. I can conclude from this only that Tesla doesn’t have its shit together.
No, I’m saying that that you are obtusely interpreting GM’s fundamentally clear statement about its self-driving car goals and pretending it doesn’t say what it does.
Self-driving cars will increase traffic congestion rather than reducing it. Self-driving cars will be able to drive more densely packed into lanes and with fewer traffic-causing accidents than ordinary cars. This would tend to reduce congestion if those effects weren’t, in my humble opinion, going to be completely swamped by self-driving cars’ ability to increase traffic.
Self-driving cars reduce the opportunity cost of spending lots of time in traffic. Once people can work, watch movies, sleep, or be drunk while traveling alone in their cars, many more people will spend more time in their cars doing those things. People who can start their work day in their car might decide that a one or two hour commute from the distant suburbs makes more sense than living close to work. People who can’t drive at all today, like the very elderly, the blind, and the very young, might start to be ferried around by self-driving cars. People who can’t afford to own cars today and who can rarely afford to take cabs or car services might be able to use a low-cost self-driving car service much more frequently.
Once cars don’t need a driver to go somewhere, more cars will be driving around without people in them. If I didn’t have to pay for parking, I might take my car to work rather than rely on public transit. If my car can drive itself, I can ride it to work and send it home to park for free. If that’s the paradigm, not only do I replace six or seven miles of transit commuting with car commuting, my car clogs the roads twice as much. If my car can run errands for me while I’m at work, like picking up my and dropping off my dry cleaning or shopping, I might send it to the cheap dry cleaner across town rather than walking to the local dry cleaner and have it pick up specialty sausage from the suburban store instead of settling for the local butcher’s stuff.
Finally, today’s ride-sharing services are already causing public transit usage to decline. (Uber, Lyft Ride Hailing Services Hurt Mass Transit Ridership | Fortune). I have read that 60% of the cost of today’s ride-sharing services is paying the driver. If self-driving ride-sharing services are cheaper than today’s services, public transit use will continue to drop and many places may stop investing meaningfully in making them viable competitors to car sharing services. This effect may be mitigated if self-driving buses make it cheaper to expand public bus routes or if they attract private entrants into the bus market.
The issues around suburbanites are 1) in large, sprawling metropolises (like, say, Dallas), rush-hour commutes aren’t from or to a centralized location, so capacity is going to be an issue (whereas a Boston or SF might be okay?), but more importantly 2) people who’ve bought houses in suburbs aren’t indicating a very high interest in ridesharing right now, or in relying on an autonomous vehicle. I think that’s a culture thing that may have to age out of the population, but existing car culture will be difficult to shake up.
I’ll respond to these as I can. Here’s Volvo in 2014 - this was very cool, they were going to give 100 families AVs to drive around! (not being sarcastic here, i was very excited by this and envied the Swedes)
Unfortunately, it didn’t pan out, and in 2017 they announced it was being moved to 2021. (everyone does like moving things to 2020+)
How many people who currently own vehicles do you see using car sharing services to the point that they wouldn’t own their own car anymore?
I don’t know about other people’s habits, but this is how I use my car and why using a shared car would be an inconvenience for me:
-
I use it to drive to work and back, about a 45 - 60 minute commute. A shared car would be inconvenient because I am often on call and don’t know when I will need the car until I need it. I also like to keep certain work items in my car, I can’t do that with a shared car.
-
I use it to transport my children. I have two child seats in my car, they are set up for each child, with the head rest height, shoulder strap height specifically tailored to the child. If I used a share car to transport my children I would either have to install their seats into the share car every time I want to use it and then uninstall them when finished (what do I do with the seats while we are out at the shops for the afternoon?), or the car would have to already have seats installed in which case I’d still have to adjust the shoulder straps and head rest. Now the pool of cars available is limited to only those that have the specific seat arrangement I need, one baby seat and one toddler seat.
-
I go mountain biking. This involves putting my bike on a bike rack on the car, loading the car up with helmet, gloves, water, shoes, etc. Then driving to the trails where I will ride for several hours. While I’m riding the car has my non riding stuff in it (shoes, jacket, sunglasses) plus any bike equipment that doesn’t go with me on the ride, e.g., a floor pump. I like to go riding early in the morning and have the car loaded up the night before. Obviously a share car wouldn’t work.
-
I go on holidays. The car is loaded up the night before for an early start. The car must also have child seats. Again a share car wouldn’t work.
Even if only one of those situations applied to me, I’d still have to own a car.
Obviously a car sharing service wouldn’t work for me enough that I could dispense with owning a car. I suspect that the vast majority of people who currently own a car would still have to own a car because a share car wouldn’t cover all of their personal travel needs.
A car sharing service could nicely replace current public transport options, but I don’t see it significantly reducing private car ownership. People who don’t need a car for the types of things I’ve described already get by without a car.
In 2012 Sergey Brin got California to support testing in the state, which is great, but proclaimed that robot cars would be available to the general public in five years. Maybe undefined enough to give it a pass, but in retrospect woefully optimistic.
Just two years ago Ford said they’d have fully autonomous vehicles on the road by 2021…
…just two years later, they’re setting more realistic expectations:
https://www.sfgate.com/business/article/Ford-CEO-Self-driving-taxis-won-t-replace-the-11859110.php