When will self-driving cars be a real thing?

We discussed this at lunch yesterday, here are my predictions:

1. Singapore will have driverless or near driverless neighborhoods in less than five years.  But it will look more like mass transit than many aficionados are expecting.

2. The American courts and regulators will not pin too much liability on the car companies or software architects.  That said, the regulators will move slowly, and for some time will require a human driver stay at the wheel, even though this seems to be more dangerous.

3. Mapping the territory, reliably, will remain the key problem.  Until that is solved, driverless cars will be a form of mass transit — except without the mass — along predesignated routes.

4. A Chinese city will do it before America does, but Singapore first of all.

5. In less than two or three years, you will see some American car dealership advertising “driverless cars,” but in a gimmicky way.  You’ll still have to sit at the wheel and…drive them.  But they’ll park themselves and have super-duper cruise control and the like.

6. The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away.

Here is a related NYT article.  I thank Megan McArdle, Robin Hanson, Alex, and others for their contributions to this conversation.

Addendum: We also talked about whether “Virtual Reality” will be a revolutionary technology.  It will have its fans, but I don’t see it as a major breakthrough.  It makes too many people dizzy, and doesn’t really have a killer app; perhaps it will change sex however.


Don't need self-driving cars, just use mass transit and learn to ride a motorcycle.

Mass transit is rarely door to door and often entails long waits. The cities with the best mass transit worldwide have ameliorated but not eliminated these problems.

Motorcycles/mopeds/bicycles are awful in bad weather, hot cold and rainy. Expecting a significant number of Houstonians to voluntarily expose their bodies to the heat for more than 10 minutes in work clothes is delusional.

Americans could try walking.

Dare I suggest that it might explain the differences in prevalence of certain lifestyle diseases between Houston and a typical European city (stress, obesity).

Yes, but the climate in Houston compared to the typical European city....

As someone who has lived over a decade in both Houston and Madrid, I can attest to the veracity of the phrase "It's not the heat, it's the humidity". Walking in Madrid in a suit in the summer is bearable. It Houston, only between the months of November to March.

You say that like Houston didn't exist before cars and air conditioning. On the flip side, Europe has winter. They don't seem to let that stop them though.

For all intents and purposes, Houston really didn't exist before mass adoption of cars and a/c: http://www.houstontx.gov/planning/Demographics/docs_pdfs/Cy/coh_hist_pop.pdf

Even in Europe these walking partisans are ignoring how much people are switching to cars, especially on train routes or for commutes. German rail was famously complaining a decade ago already that its model was getting less sustainable because of the switch to car use.

So there are big differences in average use between the US and Europe. But the marginal trends are more similar than one would expect.

agreed, but walking in the S US - I live in FL - from May to Oct is not realistic.

try walking? Yeah, I'll just walk the 7 hour walk to and from my work. I could take mass transit, but then I'd only have to walk about 2 hours. Not to mention, this is America, I have freedom of movement, the freedom to control where I go and when. I'll never give up my car, nor will I ever buy an automated car.

Doctors don't call them "donorcycles" for nothing. Fatality rates per passenger-mile are much lower for cars than they are motorcycles.

Use bikes. Solves two public health problems (obesity and pollution), plus congestion problem. I really dont understand why bikes are so underestimated in this blog, you guys need to get out more...

Poor weather (too hot, too cold, too wet). Two wheels are fine for temperate climates, but not for the rest of the world.

People ski, skate, fish through the ice, snowshoe, etc. during northern winters. Why not ride a bike?

I biked to work a couple miles in a light snow a few times in the New Mexico Jemez; I was twenty-six and it was fun. Eight years later, I biked three miles to work in the winter in Livonia, Michigan (west of Detroit), and I did not enjoy it much; dark ride home with car traffic, slipping on ice a couple times.

Some of the top bike commuting cities are very cold like Minneapolis, Montreal, Copenhagen, Malmo etc... Also Portland and Seattle which are quite wet. Etc... And bike share data shows no drop off in biking in the summer time.

In other words - this just isn't true.

How often does your part of the world get snowstorms of a foot or more? How often is the high above 35 or 40 Centigrade? Bikes are fine as an occasional thing, but the USA really does receive a lot more extreme weather than Europe.

The UK trains break down when there's "leaves on the line," it's "too hot (high twenties)", "bright sunlight", and the famous "wrong type of snow".

A limitation of bikes is that people like their housing and employment markets to not be tethered together by an eight-mile bound. I had a five-mile bike commute in west LA for three years; it was great.

This is such a key point. I've had bike commutes and loved them. But when I don't have them, I need to be able to find employers a little further away.

Bicycles have been a mature technology for, you know, quite some time now. What is it that you expect will change that will suddenly drastically increase their uptake? Does it involve fairies?

Infrastructure. Bike usage is increasing dramatically right now in cities that are actively trying to encourage it, like Austin with its famously bad traffic. I mean, I'm watching that increase right now, but it's been going on for some time: http://www.citylab.com/commute/2014/05/rise-bicycling-smaller-and-midsize-us-cities/9059/

(Compare the Netherlands, which certainly gets cold, but has a strong cycling culture along with dedicated infrastructure and favorable laws: https://www.youtube.com/watch?v=q8h_DalTjV0)

I expect bike usage to continue to increase in areas such as a 5-mile radius of city centers. One difference between the US and Europe is that European (including UK) cities actually have "town centers" - they get confused when they come over here and can't find a town center.

Because we are so spread out and lack rail transportation in most places, cars are the only practical option many places. Bike usage likely won't increase in that sort of place.

Uber encourages bike usage. Think about the people giving up their cars, now that they can Uber on demand.

When I have a gig in the city, I look forward to biking to work. If the forecast calls for rain, I'll Uber.

I looked into getting rid of my car, but the numbers didn't work for me based on how many gigs I have outside of the city in a month. But for someone who live and works in the city, it's a viable option. Uber is too expensive to take to work every day -- you might as well just buy a car. A combination of Uber/biking is dirt cheap, and get you a car when needed.

Why should we rebuild our entire infrastructure around bike usage? Sounds like a regression. We had the ability to build cities like that, we chose cars instead, which seems to match the lifestyle and income of the average American family. This isn't a technological improvement, it's a cultural choice.

Cineminded, very good points. I've lived like that, more or less, for the past couple of years. It was harder to do about 10 years ago, but has gotten much easier.

Definite, if we eliminated the interstate highway system in favor of bicycles or something, that would be a technological regression! But adding protected bike lanes to cities doesn't eliminate cars - the cars can still drive. It just also encourages everyday cycling, which many people like to do. The more people cycling, the less bad traffic is for car commuters, which is why many cities have started supporting these protected bike lanes and other improvements for cyclists.

In general, I'm never in favor of going back in time. I hope we will see improvements in traffic and road usage through self-driving cars on demand and so forth - through further technological improvements, that is.

because this isn't denmark where everyone lives in a densely populated flat area. America is not Europe, it doesn't work outside some cities. There are 10's of millions of us who will never live in a city.

They're not called donorcycles in Europe. There are two reasons motorcycles have the fatality rates they do in the united states: the interstate system and MotoGP.

"They’re not called donorcycles in Europe." Maybe they should be. The fatality rate for motorcycles in Greece is three times the per capita rate of the US. The US rate is roughly 14 per million inhabitants. That's around the rate of France.


I have a cousin in Rome who is an orthopedic surgeon. He makes a very good living (metaphorically, as he is a salaried gov't worker) repairing the hands of scooter riders who whack them on cars, mirrors, pedestrians and other impediments.

@the critics of the suggestion you use public transportation and ride motorcycles/bikes:

1) wear a helmet. I've not fallen off a motorcycle ever. And off a bicycle once. And I ride rather fast. Then again, my reflexes are excellent, I have the body of an athlete, and I'm probably smarter than you.

2) OK, so you can't ride a bike or cycle: what's wrong with public transportation for getting to work? Your significant other can drop you off at the station, and/or ride the bike to the station? Worse case, drive an electric car to the mass transit station and park it there.

3) You don't like 1 or 2, yet you like self-driving cars, so you can do what, eat while you drive? Or maybe a self-piloting flying car so you can eat and sleep while you commute to work? You lazy lard ashe American! Then you wonder why the rest of the world hates you...

Hates Americans so much that they're flocking here from all over the world....its ironically Europeans everyone finds smug and arrogant. The reason being...you. You can take your fake global warming save the earth with mass transit and electric cars and Shove it up your inferior self righteous asshole.

Motorcycles are more dangerous to the passenger than cars. Safety is one of the big benefits of self-driving cars.

What about mass transit Fat Bob? It's much safer than cars, possibly per passenger-mile than even self-driving cars (since non-self-driving cars running into a self-driving car is probably a fatality, given the flimsy nature of non-buses).

I agree. Self driving cars are economic warfare against oil nations. They are a BAD IDEA.

Re: VR & sex

William Gibson, 1982: https://books.google.com.au/books?id=qDrfRE6PspwC&pg=PA83&lpg=PA83&dq=count+zero+virtual+reality+sex&source=bl&ots=wrvyBX0a8v&sig=0LRxSttSw1I9fbZ75OFLL0N_ecA&hl=en&sa=X&ved=0ahUKEwimhPec8bzKAhVkMKYKHSz-C0YQ6AEIHDAA#v=onepage&q=count%20zero%20virtual%20reality%20sex&f=false

35 yrs later & its still just around the corner ;)

VR sex requires, if not hardware, at least firmware, to be manufactured and sold, and few American high tech companies have any interest in investing the billions needed for even a single low volume product. Apple and Bezos outsource manufacturing.

Only Elon Musk builds factories, and only in the US to keep everything that makes his factories deliver products no one else can under his control. (Nothing other than the Apple logo can't be found in less than two years in a Samsung product that is half the price. BMW and other Euro luxury cars are hinting they will match the Tesla Model S in a few years, and Elon will sell them parts, or give them design info, but it's the experience in actually making and delivering a high quality electric car that gives Tesla the edge. Ditto for SpaceX. Elon simply reversed engineered 60s NASA rockets and figured out how to use 21st century manufacturing which then supports rapid refinement of the old proven designs. SpaceX doesn't do rapid prototyping as much as rapid manufacturing at low cost to use production rockets as prototypes to launch stuff for revenue.)

The "hoverboards" where all knock offs of a design by a developer in the US who relied on Chinese contract manufacturing firms to do his pilot products while he worked out his patents. The Chinese workers saw what he was going for an made their own and soon dozens of Chinese factories were making knockoffs.

So, the US is number one in figuring out how to make money from sex, but it's difficult to prevent cheap Asian copies that really are cheap and sleazy.

Quality VR sex will require US manufacturer of the quality soft-, firm-, hardware package. Elon is married with kids and already very busy.

I agree with the relative timeframes you propose for adoption. , but I think "will look more like mass transit" undersells the potentially large reductions to auto fatalities that the technology will contribute in the near term. Even if its incremental over time and there still has to be a human driver behind the wheel. I don't expect to be picked up by a driverless Uber in 10 years, but I think we'll have plenty anti-swerve, crash avoidance, etc. tech in most new cars sold by then.

VR sex is already pretty good. She's at one end of a Skype connection, you're at the other. You know the rest.

Blackmail, credit card issues, privacy issues (her bf is recording you)..."from what I've heard".

Wow, you read a lot of background assumptions into Larry's statement.

'VR sex is already pretty good'

Sorry, mate, I go for the real thing; not in your target market.

Try http://wwrtc.blogspot.com/

VR sex, like sex robots and previously porn and masturbation, substitute for real human partners. This is useful to alleviate the scarcity of attractive willing partners, substitute for sexual violence, and prevent unwanted pregnancies.

7. Driverless cars will increase, not decrease, congestion, and by a large margin. Not just because cars might be dispatched on errands or just to go park themselves by owners who don't care how long that takes when they're somewhere else having fun or being productive, but because the only way such a car can be tolerably safe or immune to lawsuits is to have a very strong bias in favor of "When in doubt, any doubt at all, STOP!". Which is prone to cascading.

8. In spite of the bias in favor of stopping, the first generation of driverless (as opposed to driver-assist) cars won't be significantly safer than their manually-operated cousins. The first generation of pilotless planes were more than an order of magnitude worse, and that was a much cleaner operating environment. The crashes will be for different reasons, and in hindsight unbelievably stupid ones, but they will still be crashes.

The second generation of driverless cars will be safe, and the third might be congestion-free, but getting there from here is going to be a tall order.

7. The opposite is true. Driverless cars will have much faster reaction times than humans, which will make them much less clumsy in both stops and starts.

"Driverless cars will have much faster reaction times than humans"

Eventually. I believe what John Schilling is saying is that after three or four generations of self driving cars, they'll eventually get to be completely safe and congestion free. In the meantime, they'll make things worse before they'll make them better.

And I agree. Right now, driverless cars do not have faster reaction times than humans. There's just too much data to crunch in real time right now. The algorithms and real time systems used need to get a lot more efficient or we need to get better sensors. I have no doubt they will, but we're not even close to being there yet.

Reaction time is a small part of the issue. Braking time is also a small part of the issue. Judgement, is the biggest part of the issue. If the answer to the question, "Is it safe to proceed?" is "I am not entirely certain", the robocar will not proceed. And not even infinitely fast reactions will make the robocar entirely certain it is safe to proceed, because e.g. braking time is still finite.

Humans are much better than robots at judgement. Including judgements of the form, "I am not certain this is safe but I should do it anyway". This may lead to accidents, but it also keeps the roads moving. With human drivers, traffic is obstructed whenever there is actually an accident. With robot drivers, traffic will be obstructed whenever a stupid paranoid robot thinks there might be an accident. If there's still a human driver with manual controls, the obstruction will be at least be brief.

The best argument for driverless cars is to sit at an intersection and watch what all the meatbags behind the wheel right now are doing.

"The best argument for driverless cars is to sit at an intersection and watch what all the meatbags behind the wheel right now are doing."

Indeed. The driverless car is far more reliable than one of the teenager's that routinely passes me on the right (when I'm in the middle and the left lane is free) going 15 mph over the speed limit, texting on their phone. This happens at least once per day.

They could be made smaller (http://www.commutercars.com/) and drive closer together.

"But they’ll park themselves"

Is parking really going to be solved first?

Automated parallel parking is an easier problem that has already been solved.

The think the mapping problem is overrated. The best way to precision map is to keep very precise data on where every car goes. Think about google maps traffic. It's really good. And the way it works is that it is collecting location data from the smartphones in all those cars and inferring density and speed and even forecasting future congestion pretty accurately.

So, unless you are going someplace that literally no one has even gone with their smartphone gps on, then you are going to have an adequate map for automatic driving. There are little foot-trails through the woods near wear I live and google has already accurately mapped those too using the same technique because people are carrying the surveillance sensors around with them all the time.

It sounds like there would be a causality problem with mapping routes based on data generated by driverless cars following routes mapped by driverless cars following routes mapped by driverless cars following routes . . .

Are you saying it's a bootstrapping problem? Google has already had manned cars map most of America.

I would donate my time to the for-profit Google if it would help get my commute mapped.

It was my understanding that the maps are 3 dimensional renderings of the environment - not simply the path - that were read but not written by the passenger vehicles.

Ok, so the drivers teach the software what the road looks like and so swerving gets incorporated to avoid pot holes, but when the pot holes get fixed, the self driving cars keep swerving, with most driven cars swerving as well as they pay close attention to the driverless car ahead, to avoid rear ending it. Meanwhile, the new potholes get hit repeatedly because few driven cars are providing the needed over voting to change to swerve.

Let's say a manufacturer adds road roughness sensors to quickly adopt swerving around pot holes. Do you think they will make that data available to all other manufacturers?

What about burnt out signal lights?

A just read a conservative tirade against a Obama administration proposal for radio spectrum to signal between vehicles and road features. Like signal lights, warning signs, etc. Their argument was the private sector should be left to do everything visually.

At any point after typing the first sentence of this comment did you go back and realize "this is utter nonsense, I should just delete the whole thing"?

I know progress pessimism (stagnation, whatever) is Tyler's overarching framework for everything, but these estimates seem way too conservative. And the only real gains come from everyone having driverless cars? Someone will have to explain that one to me.

I happen to be a Software Engineer. We will not have Strong AI in the near future so we will not have driverless cars in the near future.

Why do you need AI for driverless cars? Isn't loads of data plus a series of decision rules good enough?

Because driverless technology is a mapping problem. If you read about the technology you'll see that the mapping problem will be practically solved only thru large advances extraordinary machine learning.

There's a reason why Google's cars do almost all their miles on a "track" even if it's an urban street. It's the only routes they've been able map so far.

That reason is liability. In Koch-paradise, without regulations or effective tort, you'd see those cars in the road within 3 years.

Don't forget, these cars don't have to be perfect; they just have to be better than humans. That is a startlingly easy task to achieve and we're there.

I'm not sure why people think it's a mapping problem. There are so so many algorithms that can deal with maps/graphs effectively. Again, it doesn't need to be perfect; just better than (the worst) humans, which is crazy easy. Also, the decentralized nature of learning will mean what mistakes occur will not be oft-repeated.

"Don't forget, these cars don't have to be perfect; they just have to be better than humans."

As a political (and hence, regulatory) matter, this couldn't be further from the truth.

"It’s the only routes they’ve been able map so far."

The city/suburban dwellers here ignore the roads in rural areas. There are tens of thousands of miles of winding, two lane roads in rural areas in the US. Many are unpaved township roads.

People live there too. They vote and have representatives too.

When we can map every little road in the US, then we might have driverless cars.


Just like we only got broadband once everyone in America had it? Silly argument. Many things happen without everyone being onboard.


Care to elaborate? I'm not sure I follow.

@AI Engineer

"Don’t forget, these cars don’t have to be perfect; they just have to be better than humans. That is a startlingly easy task to achieve and we’re there."

What do you think about replacing voters with AI under the same logic?... "They just have to be better than humans..."
Every week I meet a few cranks who believe the craziest conspiracy theories. Not that I think my opinions merit any more worth; in democracy we're all equal.

Re: rural roads - just for the heck of it just now I did a Google maps search of my dad's tiny hometown in rural Georgia (population 565). Google has street view even there.

They're not decision rules. They're learned behaviors in Neural Networks. Maybe with a few rules on top, but mostly the cars learn to drive the way we learn to drive. By watching, then being coached, then learning from experience.

I'm not sure why people think it's a mapping problem or that the AI is strong enough... Neither problems exist as far as I'm aware. If they do exist, they're not strong enough to stop the trend.

I would assume the map accounts for multiple features in the neural network in addition to the sensor data, both in learning and prediction. No?

"Mapping the territory, reliably, will remain the key problem"?

Perhaps one could call a vehicle that's dependent on precision mapping an "automated vehicle," somewhat like a train that is confined to its rail (but without the rail), and which runs on a protected right-of-way.

But that's not even close to a "self-driving car," as "self-driving" implies the ability to deal with the unexpected without just entering into a freeze-in-place state. A driver (human or otherwise) that can't deal creatively with unexpected events is not a competent driver.

It's a coordination issue. If all the cars are driverless we will see huge improvements in traffic flow.

"6. The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away."

"I know progress pessimism (stagnation, whatever) is Tyler’s overarching framework for everything, but these estimates seem way too conservative. "

I'm not sure what leads you to think that. The average age of an automobile in the US is 11.4 years and the trend is upward. If you assume that every vehicle sold in the country was fully automated with in the next 8 years, then it would be 20 years before half the vehicles on the road were fully automated.

I can't imagine virtual reality being all that much more revolutionary than Nintendo or Play Station 4. Sure, gamers will like it, but the sensory experience is far from vital to a good gaming experience.

Last year I got to test an Oculus Rift that had a scuba diving simulation (including headphones). Within 60 seconds I felt like I was in another world. Very relaxing.

We'll have driverless cars when you can answer this question:

Suppose your driverless car is tooling down the road, and a dog runs into it. There's no time to stop, and there are children lining the sides of the road. Swerving to avoid the dog may endanger children. So what should the computer do? Just drive over the dog? It's a moral choice, and a very human one. When a human makes a poor decision when facing a snap decision like that, we tend to cut the poor person a lot of slack no matter what he or she decides, because we understand. But we are certainly not prepared to extend that courtesy to a car company.

Now replace the dog with a child. The car can either swerve and maybe kill more than one child (but maybe not), or it can just drive over the kid. Tell me a scenario under which the car company won't get its ass sued off if the car either decides to run the kid over or it swerves and plows through a crowd of children.

Now replace the kid with a soccer ball. How is the car supposed to tell between a child, a soccer ball, and a dog? Pattern recognition is nowhere near fast enough or good enough to be able to tell them apart in the split-second that would be required.

There are a whole host of problems like this that we will have to solve. Moral choices, tradeoffs between two bad outcomes, and other problems in which the car will be involved in hurting or killing people and destroying property.

I would also not like to be the computer engineer that coded the 'avoid dog at all costs' algorithm if it causes the car to kill a person. Or the guy who coded the 'drive over the dog if there are any other risks' algorithm and the car kills someone's family pet to avoid a mild fender bender.

It's even worse than that. Consider how drivers and pedestrians in Boston constantly play chicken with each other, trying to force the other guy to yield. Computer-driven cars, as you say, have to have caution built in to them. The scenarios that you describe with dangerous potential accidents arising and forcing the car to slow down are bad enough, but now add human agency instead of mere accidents and we've got armies of Boston pedestrians and drivers taking advantage of the computers' need to slow down or halt when faced with a hazardous situation.

mkt42: The only good point made. Picking up on social cues is the hardest thing.

Dan: Discerning between a soccer ball and a kid? Really; you think that's hard?

Think along the margins. How about a dog vs a kid bent over to pick up a toy? Add in some odd or insufficient lighting and a few stray reflections, and a decision that has to be made in milliseconds. Now how confident do you feel?

But the larger point is that there are a lot of decisions that have a moral component to them. We allow people moral agency, including cutting them slack when they make a bad decision under duress, with incomplete information, or when a snap judgement is required. I don't believe we are remotely ready to give computers that sort of leeway.

Let's see what happens when an automobile mows down a child when it could have swerved onto the curb and just destroyed some property because the software made a poor choice. Let's see what happens when an automated car intentionally runs over a dog because it made a complex decision that hard braking was too dangerous in traffic. The computer may even have been right from a logical standpoint, but is that going to placate the dog owner or her lawyers?

I think the history of the Segway should give people pause. Remember the predictions? It was going to be so popular that entire cities were supposedly going to be modified to allow for Segway use. It was going to change the world. But what happened? A few early sidewalk accidents caused a rash of bannings in cities across America. Social pressure stopped people from tryng to ride them to the office. And the next thing you know, the Segway became a niche product.

It doesn't even have to be a safety issue. Google glass was killed off by the 'ick' factor and the sense that people who wore Google glass were pretentious douchebags. And suddenly all those thousand-dollar google glasses were collecting dust in drawers instead of being the ubiquitous helper Google thought they'd be.

Do not underestimate the social and moral dimensions of driverless cars. We have become a society of hysterics. Look at the nonsense surrounding the 'drone' scare. Accidents involving driverless cars are inevitable. Even if they cut fatalities by 90%, that still means thousands of accidents and deaths per year. Each one of those will be fodder for anti-autonomous car hysteria.

BTW, I am a computer engineer as well, and I work in factory automation.

I've heard this before and find it overblown. Yes, driverless cars will be cautious and therefore exploitable, but reasonable limits will be programmed into their exploitability - seen cynically as rules for when it's OK to hit human jerks. Courts don't generally bend over backwards to protect the rights of reckless pedestrians (especially when their recklessness is on video), and the urge to exploit the caution of cars will be balanced by the few deaths where courts ruled that the asshole had it coming. I'm saying that you are pointing out a problem, but it's a problem that gets solved in maybe three precedent-setting court cases, not a problem that holds back an industry.

The answer in almost all of these silly hypothetical situations is: stop the car. Swerving substantially increases stopping distances and makes the car much less predictable to all other vehicles and pedestrians in the area.

The big advantage of a self-driving car is that it has 360 degree vision and sees these dogs and children and will have slowed to anticipate sudden movements before the situation would have arisen.

If the car cannot see dogs & children because of visual obstacles, it's going too fast, whether driven by human or robot. Which of course means that self driving cars will be going a lot slower than non-self driving cars in residential areas. That speed reduction is going to be a large part of the predicted 90% drop in automobile fatalities.

But will you care if you're busy watching a video or commenting on the internet that the speed is slower?

or asleep.

Eh, this is pretty incredibly overrated, since no matter which algorithm is performed tens of thousands of lives will be saved compared to human drivers. Human drivers have to perform these decisions all the time, and other ones, and definitely do worse-- and yet the auto insurance prices are affordable.

It seems at least as plausible that insurance prices and outrage force human drivers off the road.

Good point. I wonder if it will ultimately be the insurance companies that decide how those algorithms are written.

When driverless cars are a minority but totally reliable they will drive even more slowly than they can, but then insurance companies will start raising rates for humans -- especially humans with even minor infractions. At that point bad drivers will start dropping out and then as rates rise, the remaining human drivers will have to drive more cautiously to avoid huge bills before finally giving in to nearly 95% driverless with the few remaining being those willing to pay enormous fees for the pleasure of driving themselves with huge liability risks for the humans not the driverless vehicles. At which point driverless cars can raise average speed due to not having to worry about humans.

No. Rates will be lower for safer non-human drivers. But in a competitive market, prices cannot increase for human drivers unless they start to get in MORE accidents, which would not happen.

Good point.

Humans rear end driverless cars more often because they get frustrated by the safe driving of the driverless cars and tailgate and thus end up rear ending them.

I think you are missing the point of his argument. Insurance is affordable because as long as a human makes an understandable choice for a human, ie swerves to avoid something, hits something else instead, the insurance company pays for damages but not liability.

With a computer algorithm if the jury decides it doesn't like the risk balancing the being done, maybe decides to hold the company that wrote the software liable for several million dollars for every loss of life. And the likely outcome is that no matter how prudent your algorithm is, when it does result in loss of life, the jury will always find the software company liable.

I think it's a pessimistic outlook but not completely unreasonable.

One possible approach is to establish a self-driving car injury compensation fund, similar to what HHS has done for vaccinations. The tech/car companies pay into this fund and a special master administers the claims and compensates victims. The rationale for such a system, in my mind, is consistent to our vaccine policy: the encourage research, adoption, and get the lawyers out of the way by resolving the liability questions.

Dan, You have some good points.

Look for college courses entitled: The Theology of Driverless Cars, The Moral Philosophy of Automata, and Coding Morality 101.

It will be interesting to see how we write algorithms for moral choice.

Driverless cars will adhere to the Golden Mean: don't hit too many dogs, but don't hit too few.

No one gives a shit.

I mean, I'm sure professors standing in their classrooms and liking to be quoted on driverless-car ethics will have Something To Do.

But this is a sideshow issue. DL may insist on having an answer, but he doesn't get a veto.

A less dire, but equally hard-to-program, scenario is the merging issue when there's an unexpected lane blockage (emergency vehicle, accident, etc.). I see it all the time, all traffic is stopped, I'm in the left lane, I wave the right-lane motorist into my lane, he does so, gives me a thank-you wave, then we both inch along. I don't see how the driverless car will manage that social interaction. How will it know I'm waving it in? Will it just stop forever until it sees a clear left lane? Will I ever get my thank-you wave?

Of course if we're both driverless, the computers can do the communicating.

Zipper merge is pretty common. The car can tell if space is being left for it.

Right now, if you signal a lane change in a merge situation about 20% of the drivers will leave a full car space for you to move into. Another 70% will open a little space, and then allow enough space once you start moving into the lane. The remaining 10% aren't going to allow anyone in, and the merger had better call off his merge at the last moment when that becomes clear.

I think merging will become a lot slower if self-driving cars don't push a little, but that's a fairly "aggressive" move for an automated system.

Reaction times for computers are at least an order of magnitude faster than yours. Let's assume that your merging at no more than 80 mph. Now, assume the same merge situation but you (and the other vehicles) are driving at 8 mph. How hard do you think that would be? How often would you hit another vehicle at that speed?

Wait, aren't we allowed to weigh the overall number of likely lives saved against any single unintended incident?

As of now a drunk or tired person can just decide to get in the car and run head-on into a car with a family and kill them all.

What's so difficult about this? Of course it will choose the dog if for no other reason than liability. Compare it to a human: how many times has this happened to you personally? And if it hasn't, have you thought about this scenario enough to know with certainty how you will act? I doubt it. The driverless car benefits from the matrix of thousands of cars' experience rather than just one driver.

Also re: the Boston pedestrian problem -- the people playing chicken will learn very quickly whether and how to run out in front of driverless cars. It wouldn't surprise me if a web site of hacks for trolling driverless cars pops up.

Based on what I've read, driverless cars move like a grandmother is driving it, not an angry commuter.

Could the driverless car automatically send a video recording of the people playing chicken? Software could match up their faces and send them fines.

Path dependencey makes this though because Coase's theorem would say if the self driving cars are dangerous avoid them do not let children near roads put up barriers. Trans cannot stop and everyone knows it.

It would allow the elderly and the drunk to continue driving, correcting their mistakes in real time. If they stomp on the accelerator in front of a shop the car will know they really didn't want to do that

I wonder what it will do to sales of sports cars. If people get used to the car driving itself, is anyone going to want Porsches, Corvettes, Lamborghinis?

We already know the answer to that. How many people have their chauffeur drive the around in a sports car?

The cars will be differentiated based upon size and price, nothing else will really matter. The enormous, and tremendously, wasteful business of car style will go away.

As for virtual reality: Have you tried it? With the latest generation of hardware? Because it seems to me there are currently two kinds of people: Those who have tried it, and think it's going to be a huge deal, and those who haven't and don't think it's anything more than a niche product.

The HTC Vive provides an astounding creation of a completely artificial world that you can walk around in and interact with. You honestly believe you are in that space. The sense of 'presence' in the virtual world is almost total.

Obviously, gaming and porn are going to be big industries for this, but the applications range from travel to social media to education/training, and I'm guessing whole new forms of entertainment we will discover as we figure out what to do with all this capability.

Nikon now makes an 'action camera' that takes full 360 degree video. You can record an experience, and then play it back in VR and feel like you are right there again. But even better - you can look around and see everything. For example, you could record 360 degree video as you walk through a park or a museum, then later play it back and watch what was going on around you.

Imagine entire new industries of 'experience designers' who take 360 degree cameras into caves, skydiving, on roller coasters, or put them on satellites and rovers on other planets. Anyone with a VR headset could then experience being there. With streaming video, we could create live experiences that people can experience as they are happening, as if they were there.

Imagine a 360 camera on the 50 yard line of the superbowl. Put on your VR headset and headphones, and YOU are sitting on the 50 yard line watching the game. How much will sports fans pay for that?

Tech forecasters are saying that VR hardware shipments will reach almost $3 billion dollars by 2020, and including the content market VR could be a $20 billion industry within four years. If that forecast is wrong, my guess is that it's wrong on the low side.

Agree. Went to a computer conference last year and viewed some of the new goggles and devices. Beyond entertainment, look at more virtual real estate listings, hotel accommodations, car interior views, etc, and then imagine how this will affect online purchasing versus in store purchasing.

I toyed with the idea of starting a VR company to create relaxing and meditative experiences. I got to try a scuba driving simulation last year and it was transformative. I could even see using it to help fall asleep. And imaging the scenarios you could create for VRers who are on LSD or even just marijuana.

Yes. Assuming technical progress in computation continues as it has been for at least a few more years, VR is going to be really really amazing. Its addictive potential when compared to current smartphones will be like heroin to menthol cough-drops. Time to reread Infinite Jest.

I think "with the latest generation of hardware" is the key point. I suspect many VR naysayers are like me, tried out a set of goggles ten or twenty years ago, and are trying to do the equivalent of predicting iPhone adoption based on familiarity with the Newton.

Of course, it took until the iPhone 6s that my wife felt that there was finally (finally!) an _almost_ acceptable successor to her beloved Newton. But then she was probably the only person on the planet for whom the Newton would consistently and accurately translate their handwriting.

"Imagine entire new industries of ‘experience designers’ who take 360 degree cameras into caves, skydiving, on roller coasters, or put them on satellites and rovers on other planets. Anyone with a VR headset could then experience being there. With streaming video, we could create live experiences that people can experience as they are happening, as if they were there."

But they're not there. This is a further development of the newest and most significant form of mental illness, confusing artifice with reality, constructing an artificial environment that corresponds to desires. The process of this disease may well have begun with air conditioning, allowing people to live most of their lives in a controlled ambient temperature. Extensive efforts have been made to neutralize odors that are found to be objectionable. Television, but not radio and movies, were the next development, particularly insidious because people watching television, a football game for instance, have the impression that they are actually seeing the event. They're not. They're looking at a manipulated electronic image.

Evidently there is a combined effort to totally avoid any discomfort while at the same time having unusual and exciting experiences with no discomfort or risk. Is that what life is really all about? If you leave your climate-controlled home, get in an air-conditioned limo, arrive at a modern airport, fly in a jumbo jet to Kenya, ride in an air-conditioned Land Rover while listening to Adele on the CD player, enter a new safari resort and sit on the insect-free deck watching the wildebeests at the waterhole while sipping a whiskey soda, have you really been to Africa? Would making a similar trip via virtual reality goggles be in any way meaningful? Life is getting more and more weird.

Bill Gates can go on his yacht and anchor it in the Bahamas.

I can do that, too, with virtual reality.

I can also arm my yacht with torpedos and patrol the rings of Saturn in it.

Bill Gates can't do that. Expect more fantasy and less reality in VR.

My thoughts exactly. The really exciting stuff hasn't even been dreamed up yet. Imagine people predicting what the web would be like in 1989 when Tim Berners-Lee invented it. I started an Internet company in 1996 and spent a lot of time educating smart investors why the web and email were important.

You're not on a yacht anchored in the Bahamas, you're in your basement. You don't have a yacht, it doesn't have any torpedoes and it's nowhere near Saturn. It's all bullshit. If you thrilled by it but you have a sad, sad existence.

I see your point, but doesn't your claim come down to all fiction is pathetic?

> How is the car supposed to tell between a child, a soccer ball, and a dog?

It's not. The car should attempt to stop as fast as reasonably possible, but without changing a lane:

1. Any object on the street can be dangerous to the car inhabitans, including a soccer ball.
2. Swerving increases the risk of side crashes, which are generally more dangerous for people in the car than the frontal crashes.
3. The car can deploy airbags in advance, if necessary, unlike the current technology which deploys them after a physical contact.

That shouldn't actually be hard on a residential street, since the driverless car will obey the speed limit rules (general 30 mph, 20 in a school zone, less if ice conditions or if there are children near the street) and have better reflexes than the human driver.

On a highway that's a less viable option. But it's not like "child running unexpectedly into a middle of a busy highway" is a scenario with a likely happy ending with current, non-driverless cars.

Why would swerving increase the risk of side crashes for an AI? Wouldn't they have a 360 degree understanding of their environment?

20 years is a long time in tech. 20 years ago I was first introduced to the internet at 56k. Self-driving cars are not the internet. Once they are economical, they are brilliant. I would say they would be ubiquitous in Northern Europe in 10-12 years. The reasons:

1) Musk says the technical problems are solved. I have heard the mapping issues but that was 18 months ago, they have very few cars collecting the data, and the improved processing of sensory data makes accurate mapping less important.
2) Regulators will listen to Google, Apple, Tesla, Ford, Mercedes, Volvo and so on. If it was Big Tech vs Detroit you would expect a lag, but it's not. And success in one location will accelerate uptake in others.
3) Very few people will buy a non self-driving car once they are on the streets. The value of second-hand non self-driving cars will dip rapidly as people choose robo-taxis instead. Once this begins to be reported people will try to offload their old cars as they know the value trajectory will be down. This is an accelerating process.
4) Most of the adoption will be robo-taxis bought in bulk by the likes of Uber. You are unlikely to see a city with a few cars pootling about, there will be hundreds all at once. The companies won't be able to produce them fast enough.
5) Cities may well favour the self-driving cars as they may be smaller (one person cars), not need parking, be cleaner and so forth. Again, once TV cameras start broadcasting images from the first cities that have them and the benefits, there will be a rush.
6) Insurance for regular cars will increase.

If they are prohibitively expensive that could slow adoption as the robo-taxis might cost to much, but since Lidar is expected to become very cheap in the coming years that seems unlikely and there will be fierce competition too. I am also sceptical that accidents will make a difference. Lots of people think mobile phones give us cancer (evidence is very poor), but they all have one. The benefit is too great.

Americans a) love to drive and b) have fewer reasons to believe that the government will protect them from hypothetical corporate and/or spy state over-reach.

Hence, I suggest that adoption is likely to be a lot faster outside of the USA.

(Insurance for regular cars will not increase, as risk and payouts will not increase. )

I would add that driving is far cheaper in the US than in much of Europe, which may have an effect.

You are right about insurance. In fact would not risk be reduced as there would be fewer bad drivers on the road?

As always, you are insane.

The typical driving experience in a net negative. This is why people have chauffeurs are a thing. America has no 'love of driving' that will make adoption substantially less rapid.

The day we have cars that work well enough that those who run the programs are happy with the risk is the day that almost everyone who can buys one.

Then why do Americans go on roadtrips and long Sunday drives?

Speak for yourself, Alain.

What percentage of Americans regularly go on leisure drives just for the sake of driving? What percentage of Americans driving miles do you think is voluntary leisure driving? I don't even know why I'm asking if you don't even understand why Americans go on roadtrips. You might as well argue that Americans love buying gasoline... they do it all the time!

There's no reason the cars can't be hybrid. Self-driving in your daily commute, but put it in manual when you feel like it.

"Musk says the technical problems are solved. "

Musk says a lot of things. Some are even true.

Based on past practice, we should conclude the technical problems are not solved.

"I would say they would be ubiquitous in Northern Europe in 10-12 years. "

The average age of a car on the road in the US is 12 years old. It's probably similar in Northern Europe. It's hard to imagine even 10% of vehicles on the road being automated in 12 years. Hell, I'd be surprised if 10% of vehicles being sold are automated 10 years from now.

> 6. The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away.

Well, "everyone" is not an exact statement (there may still exist people who collect 'ancient manual-driving cars' 100 years in the future). But I could bet on "majority of new cars in 20 years in Scandinavia or US Northeast".

But I could bet on “majority of new cars in 20 years in Scandinavia or US Northeast”.

Really, you think that the first places will be where it snows? I would definitely bet on more temperate climes first, like the US West Coast.

Where it is easier--not harder--for a person to manually drive? I believe the opposite.

The point isn't where it's easier for a person to manually drive in an absolute sense, the point is where it's easier "for a self-driving car to drive than manual driving." Manual driving in the snow is harder than in good weather, but self-driving cars perform even worse in bad weather compared to their normal performance.

>Really, you think that the first places will be where it snows?

For some, a kneejerk response that all things pure and forward-thinking can be found first in Scandanavia is automatic. As you can see here, they often do not think it through.

If you ask such people why they choose to reserve their highest praise for the whitest countries on Earth, they will become very self-conscious.

Since one of the biggest problems is mapping, I was mostly going to name a couple of rich and compact places that came to mind, like Denmark or, yes, northeast US.

After all, economically, hiring a cheap chauffeur will continue to be a viable option in China as well as Singapore, so I'm less confident there will be 1/2 of new cars penetration in 20 years, which are the terms I named.

In the blizzard one would have to drive a self-driving car manually. But those don't happen every day in Scandinavia, you know :)

Culturally, I think Sweden has a target of 0 road deaths, and in general rich societies will be able to pay more for safety features in a car, including being driverless.

One country to add to the original list could be Japan.

When do commercial trucking companies start using driverless trucks?

Probably 5 years or so. The challenges of interstate driving are much less than in stop and go traffic and the advantages of driverless big rigs should be fewer accidents, lower insurance costs, and elimination of driver pay and time limits compared to a driverless 24 duty cycle.

Truck transports costs should fall by 30 or 40%, the cheap 1 day JIT range should expand from 500 to over 1000 miles. Not a logistical revolution, but maybe half of one, and will increase cost pressure on rail.

Right now almost 3 million people, almost all men, make their living driving big rigs, and they might all be out of a job in 10 years or less. Will the rest of the economy be able to find new roles for them at anything near what they are making now?

>Probably 5 years or so.

This is always the right answer.

All the technology is available today, and is already in use in limited applications. The 5 years part is clearing regulatory and litigation hurdles.

I've long said that this driverless tractor-trailer case is the low-hanging fruit. The interstates are the most predictable, best maintained, best marked roads around. Get driverless trucks making those long-haul routes. Approaching their more urbanized destination, pull off into a holding area. In the short term a human driver would then take the truck through the more difficult urban area to the distribution center or the actual destination. Longer term, still a human driver doing the tough part, but remotely from a central location using the truck's sensors.

This is also a case for getting driverless cars through tough spots. Don't dump control back on the human passengers, who are now not paying attention and less practiced as well. Instead the car asks for help from a pool of human drivers at a remote location, a super-OnStar that could operate as a subscription service. The remote driver gets the car through the blockage and then hands back control to the car. And once one car gets through an unforeseen situation, then all the following cars can apply the same solution.

The low-hanging fruit-- and where it's already happening-- is on tractors on farms, which can be programmed to take pretty automatic routes without worrying about other vehicles.

"When do commercial trucking companies start using driverless trucks?"

Barring legal and regulatory issues the trucking companies will be the very first to convert. And the average age of a large commercial truck is much less than the average age of a vehicle. So, adoption will be much more rapid.

Singapore has to put those autonomous buses to work before Spring 2016 to be first https://bestmile.com/2015/12/17/sion-presentation/

Safety and how people interacts with autonomous vehicles are important issues. But I love the Swiss honesty:

"Shuttles can transport 9 people. Their price is around 183K CHF (1 CHF = 1USD). It's quite large amount, but the shuttles allow to save money since a driver is not needed. This kind of vehicles are amortized quickly. They allow to have a shuttle line with operating costs 40% less than a standard line."


Self-driving cars "will look more like mass transit". I trust Cowen sees the irony in that comment. People are opposed to "public transport", preferring "private transport". But there's no such thing as "private transport": drivers of privately owned cars must have a public license, must use public roadways, must comply with public traffic laws, and must reimburse the unfortunate individuals who the driver injures as the result of negligence. But Cowen's point is correct: once people realize that cars are nothing more than a less efficient version of public transit, self-driving cars "will look more like mass transit". But don't tell anybody, because people will reject those self-driving cars if they believe they are a more efficient version of public transit.

. But there’s no such thing as “private transport” ...

Your argument here seems as exaggerated as saying "there's no such thing as "public transport" since passengers of publicly owned mass transit must spend their own private money on fares."

People don't use public transit because they don't like chance encounters with poor, dirty, mentally-ill strangers. That's one stigma driverless cars will not have. In fact, that's their selling point even though you'd have to waterboard Tyler to get him to admit it.

#5: Tesla already promotes what you describe, right down to the self-parking and super cruise control.

The Tesla self park isn't close to being as skilled as a human driver yet. The super cruise control is definitely a work in progress. For example, if you're in the right lane and there's an interruption in the solid line to your right, usually for an exit but here in Chicago there are lots of little areas to pull over on the Kennedy or Lake Shore Drive, the car does a little fake to the right that freaks you out a bit. Also, when you're passing on the left, the auto driver gets uncomfortably close to other car (this should be easily solved). I've been a big can't-wait-for-self-driving car geek but am less optimistic after seeing Teslas perform.

It's been possible to operate airplanes without pilots for years but pilots still have jobs. Driver-less cars will be more of a certainty when autonomous lawn mowers become normal.

That's BS, chuck. Commercial airliners cannot be operated without pilots.

Compare the percentage of airplane ticket sales that go towards paying the pilots' wages compared to compared to the percentage of taxi fare sales that go towards paying taxi drivers.

When car AI fails, the car can just stop, and the passengers can get out... We won't see plane AI for quite some time because the costs for failure are much higher. I agree with Chuck, though, that we could probably have planes that fly themselves today. It's easier to navigate skies than roads.

FWIW, I personally can't remember the last major plane crash that was caused by something other than human error/intention. Especially in recent years, the people flying the plane are the biggest liability towards the people in the back.

"When car AI fails, the car can just stop"

Car driving 75 can just stop?

...Yes? Not to brag, but I've done it.

Also it goes without saying that it can much more easily than a plane going 500 at 30,000 feet.

Many of the engineering problems have been resolved, but bad weather will be a difficult point for driverless cars. I'd expect them to catch on both in the temperate climate areas as well as in the very hot areas of the US first (as rain will be easier than snow), but the Midwest and Northeast will remain a challenge at first and will come later.

My daughter is getting an M.S. in medical sociology, looking into the ways VR is used in rehab therapy (think "gait"). But her larger point is that people who underestimate VR's potential underestimate the role gaming plays in the lives of young people. She has articles from earlier decades about overhyping laptops or even the internet. The incorrect predictions focus on the technology rather than on the effect it has on socialization. Wouldn't want to make the same mistake on VR.

I think you forget the role of 'looking cool' to young people.

It is too easy, right now, to make fun of someone with a giant box on their head. Especially since it is expensive and the experience is not easily shared with other people. Note that PC gaming never achieved widespread popularity; only consoles managed to achieve that.

When I first saw a man using one of those hands-free tiny-earpiece cell phones, talking loudly straight into thin air, I thought he was schizophrenic. Then I realized what was happening, and I just thought he looked bizarre and foolish. I never thought such weirdness would catch on, but here we are.

But those things didn't catch on because using one signaled Armenian selling minicopters at a downscale mall kisok.

I still dont see a lot of earpieces in use. Weirdly, the most common user seems to be black men, based on my own observation.

PC never achieved widespread popularity apart from being the global main source of game revenue in 2014? The issue with PC is that is divided in 3 submarkets: 1) PC/MAC games (sold in a box), 2) MMO (multimillion online user games), 3) casual web games. However, the 3 of them are served through a PC screen. Market share PC 40% Vs consoles 29%.


Ps. If you want to talk about gaming, please consider 45% of global market is found in Asia-Pacific area.

Ok, I was being loose with my words. They're popular, but not in a mainstream, something-I'd-talk-about-on-a-first-date kind of way.

I guess I mean that if you saw a guy in a movie playing a PC game, you'd think 'loser' but if he was playing X-Box, you'd think 'cool'. It's not fair, but it is true. To use an analogy from television, they're Adult Swim where consoles are ESPN.

You're certainly right about Asia, where the PC games market (and strategy games market) is much stronger and much more mainstream. I wasn't really thinking about that.


Apparently pc game sales will soon outstrip console game sales. Interesting and unexpected.

One possibility I heard recently: require autonomy in congestion areas. So you can drive into, say, London, but once there you must switch to full autonomy.

To me it doesn't have to be full autonomy, just restricted speed, power and acceleration. A human could still be in control, but the automobile would be rendered much less dangerous. This is a much simpler task.

The reason for full autonomy would be parking. Your car would go park itself and come back for you when needed.

The reason for full autonomy would be the greater efficiencies that could be realized coordinating with other fully autonomous vehicles.

As a stepping stone to full automation, I'd like to see GPS-directed cruise control so that once a car is within a high density zone as determined by the municipality, the car is forced to act more like a golf cart: severe speed (30km/hr max), acceleration and power restrictions. This could be done now with existing technology. The biggest problem with cars is they make the public travel ways hazardous for pedestrians and bikes. If they are severely limited, they'd have a profile more like horse carriages than horseless carriages. The pace of city life would be much improved.

I'd rather they speed up and get the hell out of my way. Your way sounds like 10x the congestion.

I don’t see it as a major breakthrough... perhaps it will change sex however
Considering that for the last billion years or so, the main drivers of all life have been to get energy and to reproduce, these two phrases are incompatible.

I can't believe I actually have to ask this to what I assume is a 50 year old male, but you don't think VR sex will facilitate reproduction do you? Inventing an even more nerdy way to masturbate seems like the definition of trivial.

Why would that be trivial? That's like suggesting the movement towards streaming HD digital nudity changed nothing, which does not seem to be the case, and even shows up in declining rapes!

"the main drivers of all life have been to get energy and to reproduce "

It's Tyler. His concentration is heavily biased in favor of getting energy (eating).

The earliest Marginal Revolution posts on driverless cars were written in 2011. I predict that in five years, driverless cars will continue to be five years away, but in twenty years they will be powered by electricity generated by fusion energy.

Fusion energy was twenty years away twenty years ago. But if it happens, great.

"I predict that in five years, driverless cars will continue to be five years away,"

The safe bet.

I personally would predict that driverless cars will continue to be 5 years away for the rest of this century at least.

Tyler's prediction is that they'll be significant in 20 years and dominant in 50 years. I concur. A lot of tech is over hyped and has major technical and economic hurdles to overcome. Driverless cars are past the major technical hurdles and they look to be economic. So, they are at the point where you would expect to see limited adoption within the next 5 years. I'd assume that partially autonomous cars would be 1% of new car sales in 2021. (By partially autonomous, I mean still needs a driver in the seat, but the driver can engage the autopilot as soon as they are on major streets and the car will drive until it gets to it's destination or is entering a minor street.

If we were less atomized, we'd probably conclude it's easier to retrofit cities for human scale and public transit. But since we're a nation of strangers, we get these sperg fantasies of driverless cars so you can stay plugged in to your electronic gewgaw on your way to work.

The problem is pure volume: there are too many cars and too many roads. Replacing cars with robots carrying one person each, creeping along at 20 mph and stopping for shadows, will make no difference. How about we learn to cure cancer or repair neural tissue first? Or just pick the low-hanging fruit: OTR trucking is a boring job that disrupts Circadian cycles. Set up OTR-only lanes for the interstates and get tired, stressed humans out of that 70,000 pound projectile going 65 mph. They can climb back in at the terminal and make the Last Mile delivery.

By the time the technology is adequate to the task, some time next century, I imagine the model of urban, commercial core surrounded by concentric rings of residential suburbs will undergo a lot of change.

How about we learn to cure cancer or repair neural tissue first?

Why cant we do both? Do you suppose the people curing cancer are identical to the people building driverless cars? Do you suppose that doubling the number of people curing cancer will halve the amount of time it takes to come up with a solution?

Re: why can't we do both.
Trade-offs exist. Every second Elon Musk spends on self-driving cars is a second Elon Musk isn't spending on his space-ship program that can avert human extinction during a Black Swan event.

It's not like he's bolting the spacecraft together himself. He has people to help, and while you could argue that it would go faster if he paid more attention/devoted more capital, I'm betting the marginal returns aren't very high at all.

What I mean to say is that the limiting factor in human advancement is seldom the attention of smart people/rich people.

In fact, I feel stupid even writing the singular 'limiting factor' when regarding advancements. There is almost never a single limiting factor, especially since time, capacity, and attention aren't fungible.

"How about we learn to cure cancer or repair neural tissue first?"

The rhetorical purpose of that question was to get us thinking about the foolishness of expecting long-standing, hard, already much-researched problems to be worked out over the next decade or two. It was not addressing trade-offs of research in transportation or computer science versus research in medicine and biology.

"How about we learn to cure cancer or repair neural tissue first?"

Why are you wasting time blogging? Shouldn't you be researching the cure for cancer?

"Singapore will have driverless or near driverless neighborhoods in less than five years. But it will look more like mass transit than many aficionados are expecting."

I would not characterize one-north as a "neighborhood," it's more like a glorified office park.

everyone talks about the next industrial (economic) revolution from this, but i wonder ...i suspect those benefits from the driverless cars are mostly for old, sick people or people with disability. they will be free from travelling. if that is the case, will that increase our average national productivity in the same way as in the industrial revolution.

Driverless commercial trucks. Driverless taxis. Driverless package delivery vehicles.

At some point, you won't go to the grocery store. You'll place an order on-line and 30 minutes later an automated delivery vehicle will honk it's horn (or text you) that it's outside waiting for your to unload. I still wouldn't classify driverless cars as the equivalent of the industrial revolution. But it will still be significant.

That will be useful, as US society gets more crowded and unpredictable. It's pretty bizarre how much energy technically savvy people are putting into withdrawal from public spaces. I think it's a primeval instinct kicking in.

Asimov's Spacers.

Seems to me a lot more like an upgrade than a revolution.

my prediction: American kids age 8 and under won't even bother getting driver licenses, except for the ones who need to drive for work (utility truck or some other specialized vehicle). In 20 years, driving will be for race tracks and offroading, and it will be mostly a hobby.

Too soon. 8 year olds will be 16 year olds in 8 years. And they'll get used cheap vehicles to drive. Now, the kids just being born are probably going to be much less likely to ever have a drivers license.

The important use case for virtual reality is telepresence. Particularly industrial telepresence. Possible impacts are
(1) Decreasing cost of all construction projects, if combined with semiautonomous robots.
(2) Expert advice available to non-expert workers.

Uber and its ilk are already getting into more bus-like service. Uber Pool for example. I think that the reason busses are large is because of labor costs. We will have fleets of tiny busses that platoon together, when we didn't have to pay drivers. And set pickup points solve the last hundred meter problem for pickups

Almost every word you have written is wrong.

>Mapping the territory, reliably, will remain the key problem

No, it is in fact the easiest problem. Terrain changes very slowly, if it all. Much greater problems include weather and construction and pedestrians and bicyclists and cops who have suddenly decided to stop traffic.

>you will see some American car dealership advertising “driverless cars,” but in a gimmicky way.

True, but this will be (and already is) dwarfed by bored bloggers insisting that driverless cars are an imminent life-changing thing, so the dealership ads will pale in comparison.

>The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away.

... and they will remain that far off until long after you are dead. But at least you can keep saying this, and you'll never be proven wrong.

>“Virtual Reality” ... I don’t see it as a major breakthrough.

Funny that, because VR already is a major breakthrough, but you'd rather talk about driverless cars that are many many years away.

There are already endless, highly-realistic videogame universes that suck up zillions of human-hours of every year, worldwide. The fact that they exist on a 4-foot flat screen, and do not wrap around your head in a helmet, is immaterial. Zillions more hours are spent "playing sports" by moving your body all alone in a windowless room. You can also discover, and navigate to, a specific restaurant 2000 miles away by simply talking to a machine in the palm of your hand. The fact that you don't even consider this to be VR/AI any more tells you just how much of a "breakthrough" it already is.

"… and they will remain that far off until long after you are dead. But at least you can keep saying this, and you’ll never be proven wrong."

""everything that can be invented has been invented." -Charles H. Duell was the Commissioner of US patent office in 1899

Tyler: "The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away."

The Other Jim: "… and they will remain that far off until long after you are dead."

Would you like to put some money on that? I have $50 that says that more than 50 percent of the passenger-miles driven in the U.S. will be by fully autonomous vehicles by 2036.

I would reverse #6. The big gains will come when no one owns a driverless car, but everyone uses them. The economics of driverless cars are most efficient in an Uber-service type scenario: a fleet of on-demand driverless cars. It's a waste for driverless cars to sit in someone's driveway - there's no reason why they shouldn't be available on demand 24/7, with someone else taking care of the maintenance/storage/etc.

I'd suggest driverless cars for greenfield retirement developments.

The roads can be fully mapped. Slower speeds are okay. Build out from that experience.

"The big gains will come when no one owns a driverless car, but everyone uses them."

Well said.

And huge gains will also come when driverless delivery vehicles flip retail from "People go to store" to "Store (really a warehouse, not a retail establishment) delivers to people."

No parking lots. Little or no air conditioning and heating, and virtually no lighting, in the warehouse. Aisles less than 4 feet wide, and shelves stacked floor-to-ceiling. The need for packaging is essentially eliminated. (The packaging appears online in virtually reality...there's no need to have a similar package for the product delivered to the customer.) No need to wait in line for checkout.

The point about #6 that interests me is whether we let this shift happen naturally, or at some point we say "given the penetration of driverless vehicles, and the safety and efficiency gains from having their inter-connected/networked systems be in control, human drivers are now not allowed." Because in a largely automated world, isn't there the potential for a "rogue" (from the system's perspective) human driver to bugger up the whole thing, or at the very least add a massive amount of complexity and uncertainty, and thus warrants prevention?

And yes, I know it's just a film, but this concept always brings to mind a scene in Minority Report, where Cruise takes over control of one automated car amongst many, and chaos ensues.

One aspect of driving that may be difficult to automate is interaction with manual drivers, particularly communicating via body language (driver plus the vehicle as an extension of the driver) and from a Game Theory perspective where by the very predictable behaviour may be exploited by human drivers.

These problems seem much more relevant in an urban environment (roundabouts especially). One solution may be to use drone style remote driving to take autonomous vehicles to highways at which point the computer takes over and the remote driver waits to take control of another vehicle. If this is the pattern then there would likely be a large increase in urban highway building.

Perhaps other cars will be required to have transceivers to communicate their position, speed etc. to make fully autonomous cars work on highways.

Physics / econmics background here, although I'm not sure that matters. Yet the vast majority of economists seem to be in Tyler's "nothing has changed much over the past 40 years camp and nothing is likely to for the next 40 years" camp. Tyler said that is when he expects The Great Stagnation to end: in 20 flippin' 50.

My predictions: 1995 -- virtual reality will be in use by 2015 but not widespread. It will be absolutely everywhere in 2025.

2004 - I saw a Discovery Channel show on driverless cars and a Google guy thought we would see them from 2020. Now, unlike economists, Google people are born with the exponential increasing computer power curve in their brains. His prediction seemed reasonable, so I added 5 years to take into consideration my home state of Wisconsin where there are a lot of guys who would refuse to be told they could no longer drive. Michigan seems similar. They can hold out for 5 years....

Tyler's 20+ year prediction is nuts - no surprise there. As for VR, dizziness has been a major stumbling block, but it looks like that has been solved.

In the 1990s, I watched some NOVA type show on VR where a woman had come out of a tank of water, crying. She had been in a scuba simulation and was overwhelmed by how beautiful it was.

Tyler: "6. The big gains come from everyone having driverless cars and that is more than twenty years away, but well under fifty years away."

Todd: "Tyler’s 20+ year prediction is nuts – no surprise there."

One significant problem with Tyler's prediction is that he seems to assume that "everyone" will have cars. Driverless cars will lead almost simultaneously to transportation as a utility. Both Uber and Lyft are eagerly anticipating driverless cars. It simply does not make any sense economically to own a car that sits idle for 20+ hours a day, if a driverless car can pick one up on short notice, in a vehicle that is ideally suited to each trip (rather than a compromise vehicle that may be inappropriately large for a majority of the miles driven).

My prediction is that within 30 years, 95+ percent of the vehicle-miles traveled in the U.S. will be by fully autonomous vehicles, owned by fleet owners (e.g., Uber, Lyft...or at least owners of more than 20 vehicles).

P.S. Among the many monumental changes resulting from autonomous vehicles will be the obliteration of brick-and-mortar retail:

Brick and mortar retail will be obliterated

Hey Mark, It was funny that I read your post in reverse by scrolling up from the bottom, then thought "this has gotta be Mark Bahner!"

Thanks for entering into the fray!

I still think you are too conservative with your timelines. At least we physics and comp sci guys are usually on the same wavelength. Tyler is to busy eating around the world to get what is happening.

Hi Todd,

My prediction was for 95+ percent of vehicle miles traveled 30 years from now to be fully autonomous. I could see how that time could be shrunk...maybe to 20 years. But, say, 15 or less? That seems pretty unlikely to me.

One aspect that seems to me to add uncertainty to the situation is that I'm very sure that fully autonomous vehicles will lead to transportation as a service, rather than personal ownership of cars. It seems to me that adds uncertainty in how the government and the courts will react to transportation as a service. For example, could an owner of a fleet of autonomous vehicles, or the manufacturer of that vehicle, be sued for hundreds of millions of dollars for a single death? Right now, we have drunk drivers killing people right and left. But the drunk drivers don't have deep pockets, and therefore can't be successfully sued for hundreds of millions of dollars. But owners of autonomous vehicle fleets and manufacturers have potentially very deep pockets. It seems to me that if the potential liability is infinite, that will really discourage fleet ownership and even autonomous vehicle manufacturing.

What do you think is the expected date that fully autonomous vehicles will be more than 50% of the vehicle miles traveled? And 95% of the vehicle miles traveled?

Best wishes,

P.S. I'm not a computer science or physics guy. I'm an environmental engineer who used to be a mechanical engineer. I'm interested in autonomous vehicles because I can see many likely society-changing aspects, such as:

1) No more private ownership, so the number of cars can go way down, and be much more closely tailored to the actual trips taken. For instance, I expect many more single-seat vehicles.

2) "Flipping" of retail, from people going to stores, to goods delivered from warehouses. This will be much more energy-efficient and materials-efficient (as dozens of stores are replaced by a single warehouse, and all round trips by single vehicles are replaced by one delivery truck running on a variable route).

3) Much greater integration of electric vehicles with the grid, because fleet owners will be much more certain of the schedules for charging vehicles.

4) The potential for an army (100,000+) of electric vehicles coming to disaster areas to provide emergency power. Similarly, fully autonomous snowplows could drive across multiple states to provide emergency snow removal. (This would be especially valuable for southern states where snow is rare, so the snow removal equipment is minimal.) And utility company trucks could do the same (while the operators of the trucks flew in to the disaster area, to save time). A couple years ago, my parents in CT lost power for a week or more. I went out and talked to the guys who were restoring their power. They were from Montana or some such state. So they drove all the way from there to CT to help. It would have been a lot better for them (and the people paying for them) if they could have flown in and flown out.

5) The potential for fully autonomous vertical takeoff and landing passenger and cargo planes. (A bit off-topic, but something I think will be big.)

I predict

There will never be

Artificial or Even Natural

Intelligence behind the wheel.

Would people get on a pilotless plane?

I would get on a pilotless plane, if I knew that pilotless cargo planes safety records as good as human pilots.

I'm sure it will be solved, but I'm wondering about the transition from automatic to driver control in some of the situations mentioned. If the automated system hands off a driver with a sandwich in one hand and a soda in the other, headphones on - what happens then?

My background is ship management. I've seen Rolls Royce and a few others starting to experiment with unmanned vessels. I see an interesting set of challenges there:
1 - For transoceanic voyages the navigation & collision avoidance is pretty simple, with some exceptions for large fishing fleets you encounter here and there. The problem is the reliability of machinery including that for temperature controlled cargo. With the vast majority of crew coming from East Asia it will be a long time before the expense of increased redundancy and higher quality fuel (to keep the machinery running properly) makes sense. That's before getting into the regulatory issues.

2 - For interharbor work like ferries, tugs, etc. the machinery issues are not so large because they would be at the dock regularly, could be easily serviced, and most run on much cleaner fuels already. The problems here are much more like those facing vehicles - complicated collision avoidance made even worse by sailboats, kayaks, other recreational boaters, the occasional swimmer, debris in the water, and so on.

I'm sure I'll live to see some automated vessels in action but I'm not certain at all what the first practical, viable application will be.

"I’m sure it will be solved, but I’m wondering about the transition from automatic to driver control in some of the situations mentioned. If the automated system hands off a driver with a sandwich in one hand and a soda in the other, headphones on – what happens then?"

This, to me, is why the trend in automobiles is towards full autonomy, not human backup of semi-autonomous systems. The more you take humans out of the loop, the less likely they are to be able to get back into the loop quickly and effectively. It's one thing for a computer to do all the freeway driving, and turn over to a human after exiting the freeway. It's crazy to expect a human who hasn't been paying attention for the last hour to react quickly and accurately to a situation that occurs during freeway driving. I think it's much better for the automated car to simply try to get to the shoulder and stop...or try to minimize the collision energy if a collision can't be avoided.

Predictions on where the Chinese city will be a populated one or one of the ghost cities?

Not sure if there is more on the mapping issue suggested -- a quick search got me to http://www.livescience.com/50841-future-of-driverless-cars.html and I honestly wonder if we're making it a bigger deal than it should be. There are definitely some challenges but some seem to be due to the approaches taken so far. I suspect we'll see other ways of skinning the cat here that avoid the problem (which is certainly one way of solving a problem)

> In less than two or three years, you will see some American car dealership advertising “driverless cars,” but in a gimmicky way. You’ll still have to sit at the wheel and…drive them. But they’ll park themselves and have super-duper cruise control and the like.

How is this different from what Tesla is shipping today? (highway autopilot/super-duper cruise control and "summon" mode assuming you have a straight driveway)?

The problem with the self driving cars will show up if it becomes an issue where its required everyone has one to make it work. There are millions of car enthusiasts would rebel and refuse. Not even mentioning the problems of the idea of this being a vast giant country with freedom of movement.

Comments for this post are closed