Will truckers be automated? (from the comments)

Dan Hanson writes:

I wonder how many of the people making predictions about the future of truck drivers have ever ridden with one to see what they do?

One of the big failings of high-level analyses of future trends is that in general they either ignore or seriously underestimate the complexity of the job at a detailed level. Lots of jobs look simple or rote from a think tank or government office, but turn out to be quite complex when you dive into the details.

For example, truck drivers don’t just drive trucks. They also secure loads, including determining what to load first and last and how to tie it all down securely. They act as agents for the trunking company. They verify that what they are picking up is what is on the manifest. They are the early warning system for vehicle maintenance. They deal with the government and others at weighing stations. When sleeping in the cab, they act as security for the load. If the vehicle breaks down, they set up road flares and contact authorities. If the vehicle doesn’t handle correctly, the driver has to stop and analyze what’s wrong – blown tire, shifting load, whatever.

In addition, many truckers are sole proprietors who own their own trucks. This means they also do all the bookwork, preventative maintenance, taxes, etc. These people have local knowledge that is not easily transferable. They know the quirks of the routes, they have relationships with customers, they learn how best to navigate through certain areas, they understand how to optimize by splitting loads or arranging for return loads at their destination, etc. They also learn which customers pay promptly, which ones provide their loads in a way that’s easy to get on the truck, which ones generally have their paperwork in order, etc. Loading docks are not all equal. Some are very ad-hoc and require serious judgement to be able to manoever large trucks around them. Never underestimate the importance of local knowledge.

I’ve been working in automation for 20 years. When you see how hard it is to simply digitize a paper process inside a single plant (often a multi-year project), you start to roll your eyes at ivory tower claims of entire industries being totally transformed by automation in a few years. One thing I’ve learned is a fundamentally Hayekian insight: When it comes to large scale activities, nothing about change is easy, and top-down change generally fails. Just figuring out the requirements for computerizing a job is a laborious process full of potential errors. Many automation projects fail because the people at the high levels who plan them simply do not understand the needs of the people who have to live with the results.

Take factory automation. This is the simplest environment to automate, because factories are local, closed environments that can be modified to make things simpler. A lot of the activities that go on in a factory are extremely well defined and repetitive. Factory robots are readily available that can be trained to do just about anything physically a person can do. And yet, many factories have not automated simply because there are little details about how they work that are hard to define and automate, or because they aren’t organized enough in terms of information flow, paperwork, processes, etc. It can take a team of engineers many man years to just figure out exactly what a factory needs to do to make itself ready to be automated. Often that requires changes to the physical plant, digitization of manual processes, Statistical analysis of variance in output to determine where the process is not being defined correctly, etc.

A lot of pundits have a sense that automation is accelerating in replacing jobs. In fact, I predict it will slow down, because we have been picking the low hanging fruit first. That has given us an unrealistic idea of how hard it is to fully automate a job.


Back in the day, truckers were called (by some) the "white knights of the highways" because, when a car was stuck on the side of the sort of road that trucks traversed, in the East, in the West, in "Appalachia" and on both sides of the Rockies, on a typical day for trucks from back in the day, it was rare that more than a few trucks would pass by before some trucker pulled over and helped (fifty percent or so of the time - don't ask me how I know this, but I do, having spent thousands of hours in low-paid jobs where people talked about this sort of thing, back in the years when, for example, both Elvis Presley and Ronald Reagan were looking forward to their comebacks) - all that was needed was a fairly easy change of a tire (tires were easier to change back then, due to the more expensive - later cheapened by milling technology advances - nuts and bolts, such sad words to speak of in the same phrase, in so many ways, when you think about it, unless you are one of the blessed souls who was born to be a mechanic at heart - I was, too, believe it or not) or the even easier pouring of water into the engine, to bring the temperature down and to allow the "overheated" engine, with its telltale "steam", fading rapidly with the poured water from its initial furor, to once again operate (the engine, that is, not the steam) on the more human ((Maxwellian (just kidding)), on the simple,easy to understand even if you were not Johnny v. Neumann, principles of 18th century mechanics, as adapted for the "piston engines" that made so many people happy, mid-20th century (at their sound, at their power, at their usefulness, at the cool clean lovably metal way they looked before nostalgia, and so on) for so long. Too bad, though, that Elvis never got the decades of success, after his comeback (Vegas 74 ! - were you there?) that Ronald Reagan got, although Elvis, bless his heart, would have made better Supreme Court appointments (and if he had --- insert here the proper Chaucerian tag, Rotonda! Friendly! Learned and Augustus! Rehnquist, too - I remember!). Being drafted in a time when draftees were considered in the condescending way they were considered when Elvis was drafted makes for better and wiser politicians, later on.

in case you think that was an autobot comment, devoid of style, please note the 'v.' in Johnny v. Neumann, and the (slightly later in the comment) earnest desire that the Supreme Court appointees of Reagan had actually been appointed by the guy who sacrificed more for his country, Elvis.

Also, Dan Hanson's quote was great, and made perfect sense, which does not happen often.

Since this is an economics blog, and since the incentive equations are more familiar to many readers here than to me, I have nothing to say about why truckers are no longer admired as being generally helpful on North American highways. Maybe a trucker or two got sued by some ungrateful person for being a Samaritan and the rest of them decided, in a cascading series of preferences, to not be helpful anymore. Maybe people like Peter Drucker taught the powers that be how to make the truckers accountable for every minute of their work day. Maybe, some day in 1974, somebody - a modern Jacob, deciding to fight an angel; a modern Moses, deciding to repent, correctly, for decades, in MIdian for a burst of anger; or a modern Leah, with either forgiveness and love in her heart, on the one hand, or with a desire just to return to Paradise, as quickly as possible, on the other hand - had the chance to make this a slightly better world, going forward, or a slightly worse world. Maybe it was in Vegas, maybe it was not, but whoever made the decision (with the lime-green Chevy parked on the curb, and the offer of what - a cigar and a fantastically accurate conversation, that could be accepted or rejected, a Rolex, the first Rolex,that could be taken or spurned, or a phone number, the best phone number - God knows) may have made the wrong choice. And here we are, learning, slowly, about the different future potentials of truckers and robot truckers. (if you know any truckers the references to 1974 - within a decade or so, the best year, on average, to be a trucker ....)

"...why truckers are no longer admired as being generally helpful on North American highways." I think the worst influence on truck drivers has been the smart phone. Truckers used to be called the best drivers on the road. Today it is not uncommon to see a truck inexplicably swerve out of lane. If you can glimpse inside the cab you will often see a trucker holding a phone to his ear or looking down at... something. Freeway driving is the one area in which autonomous driving systems should make transportation of goods much safer for the rest of us. Moreover, I can't remember the last time I saw a trucker pull over to help another motorist, even another trucker.

There is no evidence to support the notion that an autonomous system will be safer than even a distracted truck driver and it seems unlikely there will be credible evidence available during at least the next 50 years or so.

With air moving over the engine at highway speeds it should be cool enough to operate. The coolant is just needed to idle at a stoplight. Anyone who can't change a tire deserves to sit on the side of the road, although many manufacturers don't know how to secure them in vehicles and it is best to take a dremel to the bolt as soon as it comes off the factory floor. If the problem is something like a fuel pump a trucker isn't going to be able to help.

Truckers will help in more rural areas, but on the Eisenhower system there is no point since there are so many wreckers.

The "White Knight" era of trucking was the time when interstate motor freight prices and routes were regulated by the federal government. During this era practically all interstate truckers were protected by union contracts and, in any case, trucks were not equipped with satellite transponders to tell their employers that the truck had stopped moving (let alone that the driver had exceeded the legal maximum hours of driving for the day).

Stopping to assist someone is a different proposition when your employer is paying for your time and doesn't realize that you're stopped and stretching the truth in a paper log is easy and widespread than it is if you're being paid by the mile and using an automated electronic logging system.

The industry is far more competitive now, it's harder to make a good living at it and, why would an economist be surprised that changed incentives have produced changed behaviors?

Truckers can't stop to help on the side of the road because of overly restrictive Hours of Service rules impsed by the Federal Motor Carrier Safety Administration (not by their employer). Under FMCSA's rules you are allowed to drive no more than 11 hours in a 14 hour duty period. That 14 hour period runs nonstop from the moment you first come on duty, so that any time you spend going to the bathroom, stopping to eat, getting fuel, waiting to pick up a load (or drop off a load), or stopping by the side of the road to help stranded motorist counts against their 14 hour maximum duty time. After the 14 hours have expired, the truck is required to immediately pull over to the side of the road and the driver must sleep or otherwise be off duty for the next 10 hours.

"There is no evidence to support the notion that an autonomous system will be safer than even a distracted truck driver and it seems unlikely there will be credible evidence available during at least the next 50 years or so."

Omigoodness, Dan. Stick to things you know!

In less than 2 decades, a $1000 computer will be capable of over 100 times the calculations per second of a human brain. Autonomous trucks will likely see by LiDAR, radar, sonar, conventional cameras, and mapping systems. The vehicle will likely know its position and the coordinates of the road at all times to less than a foot. If vehicle-to-vehicle communication is standard, which it almost certainly will be, it will know exactly what's happening to potentially hundreds of vehicles around it. It will have a reaction time to get from accelerating to braking of less than 0.1 second.

No human in the world will be capable of driving a truck better than an autonomous system in less than two decades...let alone a "distracted driver".

Autonomous trucks transforming trucking in the next 20 years? I'll put it in the queue, right behind flying cars. I50 years ago we were told we'd have flying cars by the turn of the millennium. Where are our flying cars? Did ANYBODY commenting here fly their car to work this morning?

"Autonomous trucks transforming trucking in the next 20 years? I’ll put it in the queue, right behind flying cars. I50 years ago we were told we’d have flying cars by the turn of the millennium. Where are our flying cars? Did ANYBODY commenting here fly their car to work this morning?"

Where I worked in 1984 there was a guy who consistently flew his ultralight aircraft in to work.

But more seriously, the reason I say that autonomous trucks are virtually guaranteed to transform trucking in the next 20 years, is because I read fairly extensively on the progress of computers and autonomous driving systems. Look at:

1) The exponential growth in the number of calculations per second per $1000 of processor power,
2) The exponential growth in flash memory per $100 spent,
3) The dramatic improvements in LIDAR, in terms of reduced cost and increased performance,
4) The rapid improvements in machine learning, and in particular applications such as pattern/object recognition.

Or simply look at some videos of the initial DARPA challenge (the first one, not even the first DARPA urban challenge). Then look at videos of Teslas preventing crashes:


But this isn't just my opinion. I would say that 90% of the people who study trends in automated vehicles think they will transform ground transportation in less than 20 years.

Well, Mark,

I'd make a bet with you on the matter, if the logistical considerations of 2 random folks on the 'Net making a bet on something 20 years out were less daunting. Here's to hoping that one or both of us recall this in 20 years.



Hi John,

I made a reply to this comment, but somehow it showed up as its own comment above.


Just to give a feel for how rapidly computers are improving, and what that means for two decades into the future, let's look at how much flash memory could be purchased for $100 in about 20 years. All these numbers are "YMMV." (Your mileage may vary.)

Currently, about 400 gigabytes of flash memory can be purchased for $100. Flash memory capacity per dollar has been increasing by about a factor of 10 every 4 years. (Or less....again, "YMMV.") So 20 more years at historical rates of progress would be 5 more factors of 10, or a factor of 100,000. So in 20 years, it would be possible to purchase about 400 gigabytes X 100,000 = 40 *petabytes* of flash memory for $100.

To give a feel for what 40 petabytes is like, 4K video is about 8 gigabytes per hour. So 40 petabytes would be 40,000,000 gigabytes divided by 8 gigabytes per hour = 5 MILLION hours of high definition video. That's almost 600 years of high definition video. A truck with even 10 high definition cameras would be able to easily recall the view of every inch it traveled for its entire lifetime. (And not only that, it would be able to *learn* what other trucks saw on the same road essentially instantaneously.)

Oops. I misread a graph. It appears from the following graph that costs of flash memory are "only" declining by a factor of 10 every 6 years, rather than every 4 years as I previously wrote. If we round up to every 6.67 years, there is "only" a factor of 1000 increase, rather than a factor of 100,000 increase, in the number of gigabytes of memory per $100. So rather than 40 petabytes per $100 in 20 years, it would be "only" 400 terabytes per $100 in 20 years. So rather than 5 million hours of 4K video, there would "only" be 50,000 hours of 4K video stored per $100.

Fading remnants of a previously high-trust society.

"In less than 2 decades, a $1000 computer will be capable of over 100 times the calculations per second of a human brain."

The human brain operates at 30 Hz (beta waves). The computer on which this is being typed operates at 3 Billion Hz. This computer is capable of 100 million times the calculations per second of a human brain. And it still can't tell the difference between your grandmother and a giraffe.

To believe that there is even a tenuous connection between brains and computers requires a very special kind of stupid, the kind that requires years of study at our finest universities.

"The human brain operates at 30 Hz (beta waves). The computer on which this is being typed operates at 3 Billion Hz. This computer is capable of 100 million times the calculations per second of a human brain."

You need to do some reading. Estimates of the number of operations per second that the human brain can perform vary considerably (to put it mildly), but Ray Kurzweil's estimates of around 20 quadrillion (10 x 10^16) calculations per second is near the middle of the range of estimates I've seen:


Quoting from that webpage:

"My estimate of brain capacity is 100 billion neurons times an average 1,000 connections per neuron (with the calculations taking place primarily in the connections) times 200 calculations per second. Although these estimates are conservatively high, one can find higher and lower estimates."

"The human brain operates at 30 Hz (beta waves). The computer on which this is being typed operates at 3 Billion Hz. This computer is capable of 100 million times the calculations per second of a human brain."

That's nonsense. Like I just wrote, you need to do some more reading. No one who knows anything about computers and brains would back your statements about the number of calculations per second of the computer you're working on versus the number of calculations per second of a human brain.

"To believe that there is even a tenuous connection between brains and computers requires a very special kind of stupid, the kind that requires years of study at our finest universities."

Heh, heh, heh! The reincarnation of Archie Bunker!

It's almost certainly too late and futile to offer this advice, but: Grow up. And talk to or read people who know about brains and computers before you start slinging insults.

A few things everyone should carry in their car are: a) a car tire pump, not the same thing as a bicycle tire pump, b) a tire pressure gauge, c) a spare full-size wheel, not the mini-wheels so common these days, and most importantly d) a real cross-shaped lug wrench, not the mini-wrenches which are sold with the car. You cannot reliably expect to remove the lug bolts with those mini-wrenches. If you've got all those things, you can recover from a blowout and get back on the road, assuming you know how to use them. (I never cease to be amazed how young putative adults today do not know how to do basic things like using a plunger to solve a plumbing blockage or change a washer in a dripping faucet.)

"c) a spare full-size wheel, not the mini-wheels so common these days"

This depends on the car. With Subarus and (I assume) other AWD cars using tires with different diameters for any non-trivial distance can damage the drivetrain, so swapping in a full-size spare has no advantage. Either way you want to visit a tire shop ASAP to patch the flat tire, buy a new set, buy a single tire shaved down to match the wear of your current ones. Also carrying a full-size spare has the disadvantage that tires are big and would take up a significant portion of the car's cargo capacity.

Generally speaking, my spare tires are the best of the previous set of four when I buy new tires. I've never heard of tires being shaved down.

This isn't true. Current specs allow the 4x4 drivetrains to operate with different size tires over arbitrarily long distances which is why OEMs are doing this. Tires are also very large using an spare on an 18" rim will sacrifice to much cargo space.

Dan, on AWD vehicles if it's not true the dashboard certainly tells you it is. I've got an Acura AWD and because I neglected to rotate the tires when I knew I should, I ended up where I had to replace two tires. When I took two smaller wheels off an old Acura to get to the tire shop the dashboard very quickly started lighting up with transmission and drivetrain warnings because the front and rear wheels were traveling different numbers of rotation to go the same distance. I had to replace the other two tires with the other two smaller Acura tires to get rid of this and then drove the two bad tires over and got them replaced and drove them home and replaced all the smaller tires with the original new wheels. I don't know if the cars are actually strong enough to withstand the issue but with something like a drive train I'm not inclined to let lots of transmission and drivetrain warning signs remain on.

Nick M, your comment about the AWD with different diameters is an important one. Nevertheless, a full-sized spare is still better than the puny ones to get to the place where you get the originally sized tire repaired because you still have better driving, and the size differential of the puny one is even greater and has more potential impact on the AWD. I had to think about this a great deal recently because I lived the experience when I needed to replace two tires on my AWD car (my fault that it happened). One thing though - I think AWD can do a decent job of adjusting if only one of the tires is off. My old FWD had a full-sized spare. In both cases they bump the trunk bottom up a bit but not in any disruptive way.

When I got my daughter's car the first two things I did were to get a real spare tire for her and a real bottle jack, an air gauge, a real lug nut wrench, and an air pump, and then while I was at it got the bottle jack for my own car since I didn't have it.

"Also carrying a full-size spare has the disadvantage that tires are big and would take up a significant portion of the car’s cargo capacity."

Yes, in a poorly designed car. An E36 BMW M3, a not overly large car with big tires/wheels, for example, had a full size spare on a matching alloy wheel neatly tucked under the floor of the trunk which could hold four passengers stuff without undue strain.

The reason manufacturers have gone to minispares, "mobility kits", or "runflats" is nothing more than they are cheaper than putting in a full size spare.

I'm not sure about this advice. Tires are more reliable now. I've had a couple develop slow leaks but I've never had a blowout, nor really a flat (I detected the slow leaks before the pressure got too low). So those tools may be the equivalent of preparing your anti-aircraft defenses to stop a zeppelin attack.

What I have experienced, twice: batteries that seemed okay when I drove the car somewhere and parked it. And then just a few hours later were too dead to start the car, so I had to get a jump start.

So I do continue to carry jumper cables. The tire pump might be a good idea though, e.g. if I have a slow leak and am gone from the car for several days.

I stop to help people fix blowouts amd flats all the time. Most are the result of poor folks with bald tires, or of hitting debris on the road.

I'm a firm believer in knowing how to do all this yourself, but to be honest in most parts of the country AAA or equivalent is cheap and only a cell call away. It's not a required skill.

I end up using the pump mostly to fill up people's spares. Spares do leak slowly, and people rarely think to check them.

I once had a tire which seemed like it was tearing the car apart while travelling from Berkeley to Sunnyvale. I got off the freeway and pulled over in darkest Oakland. Fortunately it was just a strip of tread from the cheap recap tires my dad used to buy hitting the wheel well. I had a knife and cut it off, and I was so happy that temporarily cured the problem. I was so grateful that when I got back on the freeway I saw a car pulled over on the side of the road, I stopped to help them. It was a black family which were some of the most clueless people I've ever met. Fortunately, they did have a spare tire. I pumped it up, jacked up the car, and replaced the flat. While I was doing this, a well-dressed guy in a Cadillac stopped to ask if I needed any help, but I told him I was handling it. I learned a lot about people that day.

We stopped to change a woman's tire one time because we had just run over the same nails or whatever right up the road and had just finished changing our own tire two minutes ago -- so we felt like we were in a groove and doing it again wouldn't be too bad.

You've been extremely lucky. I keep my tires in shape and over the last 200-300k miles I've lost tires 5 or 6 times and not because of neglect except in one specific instance. Tires ARE better but the tools that are being suggested are minimal and really a car ought to have a few more. And you're right about jumper cables - extra long ones that are really tough so that the day you really need them they haven't stopped working because you (or I) forgot about them and never really double-checked them from time to time.

" I keep my tires in shape and over the last 200-300k miles I’ve lost tires 5 or 6 times and not because of neglect except in one specific instance."

Crap, where are you driving? What states? I've got over 200K in the current Toyota and haven't had to change a tire once. Tires are better than they used to be. (And I'm more apt to replace them at then end of their life, instead of stretching the life span an extra 25%.)

I do carry a set of extra jumper cables. I have long ones with a heavy gauge. The cheap thin cheap jumper cables often have too high a resistance to jump a dead battery. I end up using the jumper cables twice a year or so. Usually on someone else's car, but occasionally on mine.

The trend is that new cars don't come with even limited-use ("donut") spare tires but only with a can of fix-a-flat. Which means there isn't even a well in the vehicle in which to store a tire if you decide to buy yourself a spare. Although some cars do come equipped with run-flat tires.

And, really, what percentage of motorists will both attempt and be successful at changing a tire anyway?

As for a liquid-cooled engine remaining cool enough to use so long as one maintains highway speed the answer is, no, it won't. An engine that has lost all its coolant, or had a water-pump failure, will overheat even at highway speeds within a few minutes.

The engine will overheat because it wasn't designed so that airflow is directed toward the parts that most need cooling (e.g., the cylinder heads) and because it was built to tighter tolerances than would have been used in an air-cooled engine on the assumption that its operating temperature would be well-controlled.

I had a Subaru overheat on my because the mechanic left an air bubble in it when he changed the head gasket. I just got it back to the dealership. At highway speeds in 95 degree PA weather I was driving on I-80 with the heater on full-blast. When I got to the dealership the cap popped off when we opened the hood.

Checking your spare is a lot tougher now - minivans keep them UNDER the car - you have to remove seats and stuff to get at it. I only went flat once with my minivan - on the side of the New York Thruway - I called AAA just because I didn't trust the other drivers not to smear me (front left tire) - I told the guy I just called him so he could put his truck and lights in the way of the insane drivers. I saw guys move to the right (where I was) so they could SPEED UP. Insane.

To go back to the point of the post - I could see some trucking routes automated - like from one Amazon warehouse to another - where there is staff at either end who load it and it is within the same company. But for most things I don't think truckers will be automated out. Although I do think that autopilots on the trucks will make certain things easier for truckers.


The spare tire has been replaced by the mobile phone. Blame CAFE standards.

I assume you mean a spare full sized wheel with an inflated tire mounted on it. Which is a nice thought if you don't mind giving up what may be about half of your cargo space. But whatever spare you carry, you need to check it every time you change the oil (Jiffy Lube, strangely, doesn't do this) to make sure it is still properly inflated and not suffering from dry rot.

You don't need to check the spare tire every oil change. You do need to flip it over every year or so.

Needs a *drops mike* at the end.

I’ve thought the same about automating drilling rigs and frac crews. It’s a long road. This is why you have these superstar companies making huge investments in IT/Automation that slowly builds their advantage while their competitors can’t even figure out what their employees are actually doing.

It comes down to the return on investment. Automated drilling lowered costs not simply on labor but allowed different drilling strategies that saved substantial amounts of money in other ways. And created a new way of extracting oil and natural gas.

What are the gains from automated trucking at what cost?

The question isn't automated trucking. It is cheaper moving of goods. Sometimes the solution is to move it half the distance.

It's like automating Justice Kennedy. Just try to get a computer to write: "At the heart of liberty is the right to define one's own concept of existence, of meaning, of the universe, and of the mystery of human life."

I can't even get my son to write like that for his AP Poetic Civics class, not to mention his laptop to write it on its own.

Kennedy Quote: “At the heart of liberty is the right to define one’s own concept of existence, of meaning, of the universe, and of the mystery of human life.”

Ah, but notice the twisted context of that remark. "At the heart of liberty" is a court dictate that mothers can kill their baby when inconvenient but no "heart of liberty" in permitted that would allow that same baby's father to get out of 18-26 years of child support for a baby that might have been no more than a one-night stand. The same can be said of that child's later education. Neither the mother nor the father has a court-dictated right to guide their child's education. If you're poor, you are often stuck with dismal public schools.

I prefer Justice Blackmun's honesty in the opening paragraph of Roe v. Wade, where he noted that the decision had "eugenic" and "racial overtones." Rendered even more blunt, it was about aborting babies with Downs and doing everything possible to bring down black birthrates. And never forget that during the 1960s, the chief proponent of abortion legalization, was Dr. Alan Guttmacher, who was not only president of Planned Parenthood-World Population, he was the former vice-president of the American Eugenics Association.

Abortion legalization isn't about liberty. It's about getting rid of entire classes of people that progressive hate.

Good discussion, thanks Dan.

That's pretty valid, of course...I suspect most topics one actually knows the nitty-gritty of are like that. Self-driving trucks aren't going to put truckers out of work, they may increase the amount of goods one operator can handle, maybe one person becomes responsible for a platoon of trucks, there may be limited situations where they can be eliminated, but yeah there are just too many tasks to automate.

Farm equipment has had limited self-driving capability available since I think at least the 90s, but again the idea of driverless tractors whirring about outside of very very specific situations is just not going to happen anytime soon. I am reminded of something I heard about a nearby farm several years back, where they bought a new gigantic high-tech planter and the operator, thanks to the auto-steering, had little to do but read the manual for it, thanks to which he learned it was not working properly because the dealer had put it together completely wrong. Until the robots can do THAT, these machines still need human babysitters. Maybe that sounds...boring?...but even without self-driving it's not far from the reality today, running big expensive equipment is mostly easy and fun or easy and boring, skill and experience only matter when something breaks.

"Until the robots can do THAT, these machines still need human babysitters. Maybe that sounds…boring?…but even without self-driving it’s not far from the reality today, running big expensive equipment is mostly easy and fun or easy and boring, skill and experience only matter when something breaks."

I'm a factory automation engineer. On the most advanced line, the primary job of the operator is to watch for alarms and refill raw materials as needed. As lines get automated, you still need operators, just fewer.

For example, there was one facility I worked at that made nearly the same product on each line. The first two lines were built in the early 1980's and required 16 operators each. The next two lines were built in the early 1990's and required 8 operators each. The last set of 2 lines were built in the late 1990's and required 2 operators. The output of the last set of lines was twice the output of the original 2 lines.

I drove for a while.

Much driving @ major long haul firms (Werner, swift, England) is drop and swap. You drop your trailer and pick up a full one, so there's no loading supervision at all and the paperwork is a quick exchange at the yard kiosk. When there is loading, usually the loaders know better what to do than the driver bcz they load the same cargo on every load. Aside from that long haul driving is such horrible work that the driver turnover ensures that a substantial proportion of drivers have almost no knowledge or experience.

Automated trucking will happen but, rather than just replace drivers it will reshape the industry. Humans will do local movements, delivering trailers to yards near major highways, where they'll be picked up by automated long haul rigs.

Pretty much - there is a lot of long distance hauling that will be as easily replaced by automation.

There is also a fair amount of short distance hauling that is likely to be replaced too, at least in manufacturing settings. Daimler has a plant that makes transmissions in Gaggenau, which are then transported to Rastatt and Wörth, for example. A regular route which is suited to automation, much as many of the local haulers/suppliers could use automated cabs (the Seifert and Dachser trucks feeding Rastatt over the B3 are a perfect example - timing included, as the drivers are clearly following a JIT loading dock schedule at the plant).

Maybe long distance trucking on interstates will be automated but then a human pilot will take over for the last 40 miles into the big city, like how ships traditionally picked up a local pilot to guide them into a tricky harbor. All else being equal, being a local truck driver is a better job than being a long distance truck driver because you go home to your family every night.

Okay, you just gave me a vision of a pickup truck pulling up next to a big rig traveling at speed and a driver jumping from the bed of the pickup into the open cab of the rig. That is they way they do it on the high seas. I always watch when the pilot transfers, and I am always amazed at how well they manage it. It must give OSHA people nightmares.


That's a fantastic image - I'm picturing the pickup with 'Pilot' emblazoned down the side.

On the plus side, no swells to deal with.

Heh. Or maybe it'll be a Knight Rider style thing where a guy drives a sports car out and it docks with a moving truck!

Do you have any idea of how many urban areas a typical long haul goes through? The GEOGRAPHIC environment, whether out on I-90 in the middle of Montana or in downtown Boston isn't the challenge. It's the DYNAMIC environment. So when our automated long haul truck is pulling a load of produce from Stockton, CA to Boston, MA, it will be going through multiple chaotic dynamic environments on the way. THAT'S when you need the "butt in the seat". You don't need a butt in the seat to "take the load" (of yogurt from Idaho) from a truck stop on I-15 north end of Cedar City, UT to the last 30 miles to the Wal-Mart DC. That's relatively chaos free. You need the butt when you're going through the Ogden/SLC/Provo area at rush hour.

Not to slam on you specifically, but this is exactly the sort of ignorance that the article is railing against.

It's not the distance. It's not the geography. It's the chaos. THAT'S why you need a human driver. Paradoxically, it's also one of the arguments AGAINST human drivers, because WE are the source of most of the chaos. (Agent Smith is pseudo-amused at the irony.)

"It’s not the distance. It’s not the geography. It’s the chaos. THAT’S why you need a human driver."

I think you have the implicit assumption that the highways are chaotic. The conditions are chaotic to you, because you are a human and you operate at certain speed. Research indicates that the median human required 2.5 seconds to respond. An alert driver reacts in around 1 second.

A standard PLC has a scan cycle of 100 milliseconds, if it's not mission critical to be fast. IE a cheap system will operate 10 times faster than an alert human.

Imagine you were driving the "Ogden/SLC/Provo area at rush hour" at 1/10th the speed to which you are accustomed. Say 6 miles an hour. How chaotic do you think it would be if you were driving 6 miles an hour?

"Imagine you were driving the “Ogden/SLC/Provo area at rush hour” at 1/10th the speed to which you are accustomed. Say 6 miles an hour. How chaotic do you think it would be if you were driving 6 miles an hour?"

That's very good. And add the possibility that you could remember your exact position and what things looked like every time you'd driven that area.

My father and I were just talking about when he last came from CT to NC, and somehow missed the GW Bridge out of Manhattan, and got stuck in Manhattan (without a cell phone or gps to give turn-by-turn instructions). He'd driven that route several times before, and couldn't believe he'd missed the GW Bridge. But the problem is that the human brain just isn't set up to remember exactly how everything looked after a period of years. An autonomous vehicle (with sufficient memory) can remember *exactly* what its position was even decades after it last traveled a road. Or exactly what its position *should* be, based on other vehicles traveling that same road. Contrast that to humans: "And make sure you don't get caught in the right-hand lane coming off the Alexander Hamilton bridge..."

Good points, but I think they will be overcome eventually ....

My big take away is how I "roll my eyes" when people say that government can become more efficient. We can try standardizing government, but with policy always in flux said standardization will last no more than a few years. Lots of government functions look rote from a think tank or private sector blowhard, but it ain't so.

Excellent post. Detailed, "on the ground" analysis by people who actually know about a topic is far far more useful and insightful than top level analysis. Unfortunately, economists don't seem to understand this, thus their insistence on applying shitty models to complex problems outside their field that they know little to nothing about.

How many hours were spent doing non-driving activities vs. driving in the average workday?

This comment reminds me of the guy in Office Space who ends up helplessly screaming at the downsizing consultants, "I have people skills!"

It's trying way too hard. Sure, the truckers do all of these things. But is there any reason to think that the self-driving AI won't be just as good, if not far better, at confirming that the load is secure, verifying the manifest, monitoring vehicle maintenance, and interfacing with the weigh station? Does anyone really think a computer can't split loads and optimize returns? Even accepting the dubious premise that the poor driver is somehow providing effective security by sleeping in the cab in front of the load, the whole need for that sleep-time security disappears when the vehicle is self-driving and doesn't have to sleep. Truckers who are self-employed and have to do bookwork, taxes, and customer relationships will have a lot more time to devote to that end of the operations if they don't also have to drive the truck. Loading docks that are ad hoc will find themselves either standardizing or paying extra for failing to do so, like every other aspect of an industrial, computerized world.

I've been a litigator for 25 years. At one time, we saw our profession as so paper-based and quirky and fact-dependent that it would always be safe from automation. Sure enough, OCR text recognition, keyword searches, and predictive algorithms completely changed the economics of document review. No one is immune from automation.

Tom T, Dan is right and you are wrong. Stop watching Office Space, it is a midwit show.

No, Dan is wrong, because most truckers don't actually do the things that he claims they do. Most of those processes are already automated, in that they follow a scripted formula with pretty much zero thinking on the driver's part (flat bed loads excepted).

Once I picked up a load at a copper plant. The load consisted of six rolls of sheet copper. **EVERY** load picked up there is exactly the same. There's a diagram posted next to the dock of how to load the material and hammer down the braces. Three rolls over the king pin and three over the rear duals. No driver knows or needs to know this formula and you could probably automate the entire process. Today drivers sign the paper work but they hardly know what they're signing. The whole process could be done with a bar code on the cab and an automated scale at the exit kiosk.

Ever watch the switching in a rail yard? It's almost entirely automated. There's no engineer in the locomotive most of the time. Sometimes you'll see the "engineer" following the locomotive around with the control system. Other times I suppose they're in the tower.

To the comment about rail yard switching: for the most part, it is not automated. The days of the semi-automated hump yards are passing. When you see flat switching, you are seeing an engineer operating the locomotive, not to mention brakemen making and breaking connections in many cases.
Just because the engineer is not in the cab does not mean a live human is operating it; one is, and it's not an automated process. He or she is on the ground with a remote control, but the operator is indeed human.
There are a few highly-specialized automated yard operations that still have large proportions of human supervision, but it is not common.
To the notion of automated truck driving: autonomous vehicle operation sounds fine, and mostly is/will be. Yet, consider the Amtrak tragedy of the other day: according to reports, the line was under Positive Train Control, which is an automated system claimed to make such accidents impossible. While there is much more to the situation than these bare facts, the reality is that automation was not capable of preventing the tragedy in an of itself.

LOL. Office Space is not a TV show, midwit.

I assume he midwittingly confused it with "The Office".

I've been a transactional lawyer for 33 years and have also watched legal automation (not to mention foreign sourcing of research). I agree with the commenters above who say that one truck driver will be able to do more than before, and also those who suggested something that I was thinking - that some local truckers may be like tugboat operators in areas with complex currents and boat traffic. Given how successful containerized freight has been I'm surprised that various groups haven't been able to get antitrust or other necessary legal clearance to standardize more of trucking than they do already, but "driverless" trucks with the need for standardizing them may affect this if the standardization effort is broad enough.

But more to your last point, people often think they can't be replaced until they are.

It's nice to read an article on this topic by someone who has done at least some research. I worked on a loading dock for most of a decade and heard a lot of drivers' stories. There are so many variables to take into account in a driver's average day that I suspect there's no way a machine can accomplish it all. To mention just one: a driver leaves the terminal and gets to his first delivery only to find that that customer has decided, for whatever reason, that they can't accept delivery that day. Please come back tomorrow. So for the rest of that day the driver has to unload the (hopefully palletized) first shipment, set it aside (if he's lucky there's a loading dock that's large enough to do this), unload the next shipment at the next stop, and reload the first shipment, close the trailer and move on to the next stop. There have been times when a driver has to repeat this process ten times in a given day. The waste of time and effort is phenomenal and never appears on a report.

Contact terms will change to accommodate the new technology. Self-driving trucking will eventually be enough cheaper (consider only the productivity boost to long-haul trucking from eliminating sleep breaks) that customers will either give up the flexibility to refuse shipments in the way you describe or pay significantly extra to keep it.

Just saw this, you said it better.

+1, great comment.

Basically, do want the $100 discount on the shipping load you get twice per day? Then you have to have people on-site at the loading/unloading dock to deal with issues.

(I'm making up the $100 number, but I'm assuming that the savings will be at least $200 per shipment and the shipping company won't give a discount of more than half the total savings.)

It's conceivable that, to avoid this problem and this exact reason, customers who accept goods from automated trucks will no longer be allowed to cancel delivery once the goods arrive.

What about damage? I own a drinks company, about one pallet of bottles in 25 I get delivered is damaged and I don't accept it. Same for the product I ship, maybe one shipment every six weeks is damaged in transit and not accepted by the customer.

Obviously the industry norms would evolve to accommodate those kinds of issues. Maybe you only pay for the undamaged goods. The shipper is responsible for getting you the goods intact, just like now.

4% damaged between the production line and the bottling plant?

It seems a lot doesn't it? Mostly it's the 1 litre bottles, which come on a half height pallet. You'd be surprised how often the transporters think they can drop a half ton pallet on top of 4 layers of empty bottles.

You're correct for local LTL companies (eg, Old Dominion, R+L, YRC, ABF, Estes). LTL carriers will move anything of any size or shape as long as it fits in the truck. I got a bundle of 8"x20ft pipe delivered once by LTL. It was quite a job to get it off the truck. Can't automate that.

But long haul drivers frequently pick up pre-loaded trucks at major production centers, where companies have their own yard hostlers to move empty trailers to the docks and back to yard.

These points apply to a great (although lesser) extent to autonomous cars. The drivers do of course drive and this is a non-trivial task. They load and unload luggage, they often can give useful advice about the locality, they can notice if the passenger is having a heart attack or seems to be on his way to carrying out a crime, or being a victim of one, or even perpetrating one in the vehicle, and that the passenger isn't damaging the vehicle. You can call a cab to help you change a flat tire or to help you start your car if you are not so skilled at this, or to bring you a canister of gasoline if you run out. All this doesn't mean that there will never be a useful role for autonomous vehicles, it just means that the labor cost of the driver is the upper bound on the savings you can achieve with them. Since this labor cost is not really very high compared to the services they perform and the capital investment they economize on, it means that these vehicles are not going to revolutionize transportation any time soon.

Oddly enough I've read your list of useful tasks beyond actually driving and come to exactly the opposite conclusion.

"They load and unload luggage" Not worth that much to me.

"they often can give useful advice about the locality" I've got apps on my smart phone for that.

"they can notice if the passenger is having a heart attack " Well, most people manage to drive themselves without that being a day to day concern, so...

"seems to be on his way to carrying out a crime, or being a victim of one, or even perpetrating one in the vehicle," What? Sarcasm, I hope.

" You can call a cab to help you change a flat tire or to help you start your car " I've never even heard of this happening. Anyway, I have AAA. They specifically sell this service on a yearly basis.

Wouldn’t this argument be similar to saying dockworkers and sailors were irreplaceable a century ago because different goods with different sizes and destinations all have to be stowed in a ship a certain way?

Then someone invented containers and supertankers now cross the seas with just a dozen crew members.

What stops an automated truck from transporting a securely sealed container or several mini-containers from A to B?

And as for all the arguments about ensuring payment, the best routes etc, this is already being surpassed by tech.

Sailors is a good analogy.

The job hasn't stopped existing, and the ones who sail the Staten Island ferry do much the same job their grandfathers did. But the number of man-hours involved in getting a ton of furniture from one continent to another has fallen through the floor.

Are those super-advanced cranes which load container ships robots? Is the autopilot which keeps course in open ocean a robot?

"Is the autopilot which keeps course in open ocean a robot?"

Of course it is. So are NC and CNC machine tools, so are automatic elevators and thousands of other devices.

Today's journalistic conceit that 'robots' are something entirely new is unhelpful to understanding the trends and social/economic impact of technology. Reading some of the magazine articles from the early 1950s thru 1960s about the emergence of 'cybernation' and the massive unemployment that will surely follow provides a useful perspective.

The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger by Marc Levinson is all about how much had to be rethought and changed (and how much new infrastructure had to be built) before "sea-land" shipping in closed containers revolutionized the business. I suspect the same will be true for autonomous trucks.

(I loved the book so much I made the Port of Long Beach part of an LA vacation.)

That is the book that I always recommend when this issue comes up. And I emphasize that the industry change was generated from private individuals/companies, i.e., from the ground up, not imposed willy-nilly by government diktat. IIRC, the only (first?) government in put was to organize the various shipping industry component leaders to standardize container sizes to ease stacking & locking. And that was pretty late in the game. I also emphasize that each industry component had to decide what its "tipping point" was to convert from break bulk to container: the ship designers & shipyards; the container manufacturers; the port facilities (cranes and dredging deeper) and longshoremen unions; the railroads w/cars to accept containers; the OTR trucking companies to do the same; writing tracking software; etc.

In truth, it's a fascinating book, and it's a fascinating story of organic change in an industry.

How long did that change take?

The model to look at is train with wheels. Instead of tracks, we have a white line and each rail car has simple steering adjustment. The engineer sits in the diesel electric locomotive, up front. Just like a train.

We change only one thing, replace the steel rail with white paint, then we program a sophisticated bot to understand the straight line. The human does all the rest, just like a train.

Until it snows and you cant see the lines and due to that the actual driving lanes in the winter are not the same as they are in the summer (usually straddling the fog line).

There is a shortage of truck drivers because young people have more attractive alternatives, and it's only going to get worse as older drivers progressively retire. This will make automation a necessity.

same with airplane pilots btw

The shortage of both is due to regulations, not a shortage of interested individuals. This is most obvious with pilots, but the same dynamic applies with drivers to a lesser extent. On the pilot side, the barrier to entry is huge. Do you have any idea how much it costs to get qualified as a commercial aircraft pilot? And the FAA recently made it MORE expensive. For the drivers, the regulatory burden is greatest once they're in the job.

There's no "shortage" of truck drivers anymore than there ever is a shortage of labor. There's a "shortage" based upon a certain level of pay. That being said, automation is always dependent on what can be automated for a given economic cost. IE raising minimum wage levels will lead to fast food restaurants pushing the use of tablets and phones for ordering.

"A lot of pundits have a sense that automation is accelerating in replacing jobs. In fact, I predict it will slow down, because we have been picking the low hanging fruit first. "

I strongly agree -- but then I would, I've been banging that drum for a while. One good example is the Roomba. It was introduced back in 2002. In that time it has gone from a small, disc-shaped, battery-powered autonomous vacuum to a small, disc-shaped, battery-powered autonomous vacuum. Does it now go up and down stairs? Empty its bin into the trash? Do the dusting as well? Move furniture and put it back? No, no, no and definitely not. It does exactly what it did going on 20 years ago. The Roomba plucked the lowest of the low-hanging fruit in housecleaning and that's as far as it gotten (and may be as far as the device -- or company -- ever get). All the logical next steps are *vastly* harder (as in orders of magnitude -- imagine what it would take to handle stairs or dusting).

Similar to the Roomba are unrealized attempts to introduce autonomous lawn mowers which haven't come to dominate suburban lawns. Of course, there's the possibility that the man on the cul-de-sac actually enjoys the atavistic ritual of tending to the plant life around him.

But even if Roomba-style robotic lawnmowers do become commonplace, that'll be another case of picking the lowest-hanging fruit. Those 'smart' mowers won't be trimming hedges or tending the flower beds or doing edge trimming. I live in a neighborhood where the vast majority have lawn services (nobody's addicted to lawn care rituals), but I've yet to see a robotic mower. Probably because the lawn services do so much more than just mow.

There's also the fact that low-cost immigrant labor has not prompted anyone to make the capital investments that would be required to develop it. Cotton was once thought to be a crop that was impossible to pick with machinery - you would always need stoop labor, or the cotton ust wouldn't get produced. Then the blacks started leaving the South after WWII, and lo and behold, someone figured out how to create a machine to pick cotton...

Apparently, people steal robot lawnmowers. If you have to watch the robot, you aren't really gaining any free time.

I had nine Scoobas (before Hammacher-Schlemer cut my lifetime warranty off and refunded my money). They work great, but, obviously, reliability is an issue. Water and robots is still an issue.

I don't think that's a good example. There is very little demand for personal autonomous vacuums. Vacuuming is a cheap task with little benefit gained from automating outside freeing up 15 minutes of someone's time. If housecleaning cost $2,000 a week and involved multiple union hired hands you'd absolutely see a change in Roomba usage and abilities over the last 20 years.

There are some parts of trucking that can be automated soon. Large volume transports between logistic hubs. First in a role of glorified 'trailer truck'. The first truck has a driver, optionally second and third truck follow within few kilometers behind the first. The driver can still fill his other roles. Logistic connections within same logistics company can be also automated easily. Long distance drives without mandatory rest periods can also work. Small road repair crews and monitoring and problem solving services with telepresence may serve thousands of trucks.

What is hardest to automate is the daily deliveries to restaurants and retail. Drivers's job is often pulling trolleys long distances, flexibly bending traffic rules when stopping to unload in the city.
Some delivery drivers shelve the products they deliver, reducing the demand for shelving in the retail shops.

'The first truck has a driver, optionally second and third truck follow within few kilometers behind the first.'

Not precisely - the Mercedes thinking on this is that keeping all those trucks much tighter than possible with drivers (the trucks are networked wirelessly) means significant fuel economy savings. Though again, this is more related to shipping between hubs/ports than typical deliveries to varied locations.

"Not precisely – the Mercedes thinking on this is that keeping all those trucks much tighter than possible with drivers (the trucks are networked wirelessly) means significant fuel economy savings."

While not strictly required, the fuel savings alone from a close following train of automated trucks could well pay for the system, even discounting the labor savings.

Much of the discussion seems to be an ‘all or none’ provision. A future of robot OR human drivers. As a few others have pointed to, the future will be both. It will also be a gradual shift that affects every aspect of the supply chain. The AI detractors seem to point to one aspect or another of that chain as the poison pill for AI adoption. Those aspects are what made the current system inefficient and require human supervision (ie...temporary human job security.)

Once there is sufficient cost savings upside, I believe companies can and will strive to process around those issues. As others have suggested, humans will own the ‘last mile’ of transit with the bulk distances run autonomously. I also think the first adopters will be companies like UPS that own/control every link of that supply chain. They will serve as the model and motivation for a massive consolidation trend in logistics.

+1, most of the automated systems I've worked with did not replace 100% of an operators tasks. Instead they replaced some percentage of an operators load and the operators scope was increased.

I think in this, as in much of future automation, the trick will be to somehow take a job which can be 90% automated, and restructure it into multiple jobs, some of which can be 100% automated and some of which need people. The comment by Nick Nolan (+reply) hints at that. This may call for initially-strange, out-of-the-box thinking to get right. See also https://blog.foretellix.com/2016/04/23/the-rise-of-mostly-autonomous-systems/

I know this has been linked to on MR in the past, but it's dead on point here:

Machines as Children, Humans as Intestinal Flora

Like parents, we have to let them have the fun while we child-proof the environment (sanitize their inputs) and clean up after them (do whatever they are too clumsy to do and clean up any messes they create).

-- and --

Without humans inhabiting their guts, technological systems cannot process much of the arbitrariness of the world.


Could we automate the long-haul portion of trucking? Possibly. But even there, we have the problems of bad weather/low visibility and detours and accidents & breakdowns (autonomous long-haul trucks will still have tire blowouts and accidents even if they're not legally at fault). And for long-distance shipping, wouldn't we be better off putting as much as possible on the railroads anyway?

That's a really good link, and I never encountered it before. Thanks.

"One of the main new jobs will be “operators of mostly-autonomous systems”"

That's a pretty good description of nearly every line operator at the average highly automated plant in the US.

Trains do long haul really well. Trucks do local delivery really well. Work on streamlining intermodal.

I was wondering if I could please get the group's take on the impact self-driving tractors will have on agriculture. The tech is also getting very close, as GPS has a sizable presence on the farm already. And automizing the "loading" labor can be easily mechanized (hoppers for seeds, payload tanks for herbicides/fertilizers.) What does this mean for an already depopulated rural America (24/7 tilling and a skeleton staff of maintenance men in a large hangar)?

I think in agriculture we will simply continue to see the trend that we've seen for centuries: ever-increasing automation and ever-decreasing numbers of person-hours required.

There are already machines that use GPS to plant the seeds in more precise neater rows than a human can do, as well as automated harvesting and loading machines. That revolution/evolution has been going on for decades and will continue. Some harvesting is harder to mechanize, or requires the hybridization of sturdier fruit that can withstand being bumped around by mechanical pickers. Solution: newer hybrids that are both sturdy and taste, and/or mechanical pickers that can handle the fruit more delicately. Again, this has been going on already and will continue to happen.

Right now in agriculture the limit of technology is breakdowns and land variability. Machines that run in dirt break down often. Far more often than most would believe. A planter will have an issue leading to the operator leaving the cab several times a day. Bearings lock up, spray nozzles clog, steel parts fatigue and simply fall off, trash gets into seed and plugs it up. Detection of these events can easily be automated. Repairing them is the issue. A good operator knows how to fix it and can get running often in minutes. They have spare parts and tools on hand and know how. Not having someone sitting in the cab means potentially hours of lost time. And when weather is variable and will make or break a season hours count.

The machine can drive better, but right now the operator is still necessary. And honestly, the cost of adding sensors to every moving part so that it can detect every possible point of failure on a machine that has hundreds of bearings, hundreds of bushings, and random plow and ripper points that wear or break off will make an implement prohibitively expensive for the foreseeable future. As long as there are high school dropouts, farm labor won't be too expensive.

Also in muddy conditions current GPS autosteer falls because simply turning the front wheels no longer turns the machine. It keeps going straight. You have to steer using the independent rear brakes. Eventually that can be implemented, but I've not seen it yet.

Reminds me of when GM was going to automate to higher quality. Turned out that humans were much better at some jobs then robots. Plus a poor design and price pressures on suppliers (leading to them cutting corners on quality) couldn't be overcome by more robots.

I don't claim that robots didn't improve some processes but humans are able to make on the fly subtle adjustments that were very difficult to program into robots. The learning curve for robots can be much steeper then advocates assume.

I once went into a UAW plant that had a robotic parts picker as the first thing that would greet you inside the plant. I was asked what I thought of it. I said I thought two half-drunk Irishmen could do a better job. But I guess it is a nice friendly FU to the UAW workers every day.

But I guess it is a nice friendly FU to the UAW workers every day.
Not so much an "FU" as a reminder....

Sic transit Gloria?

Why do trains run into each other? Is this a symptom of under-automation, or incomplete automation, or a restructuring into multiple jobs, or technology misapplied [the engineer playing with his phone!]? Railway control seems vastly simpler than highways, but suffers from a range of the same challenges. Snow? Suicides? Mechanicals? Congestion? Amtrak failing to integrate with the freight system?

I saw this discussion in three forums before we got an intelligent comment. Thanks, Dan Hanson!

The point is not that all of that "truck related" stuff is going to replaced immediately; it is that the driving part is going to replaced, and soon. And then the next part. And the next. And each step of the way, people who are currently good at "trucking" will have to change, and adapt, and rapidly, or else, and America 2018 suggests people don't do so well when they're forced to adjust to a rapidly evolving workplace.

+1 from the trenches of trying to infuse AI into at least some simple business functions, for which the data is readily available. Full disclosure required, I work for Microsoft.

Recent impressive(!) successes in "AI" are all in closed worlds with finite state description, like chess. Those will all go to machine learning in a few decades. Low-level perception (object detection and recognition) will also be done.

On open worlds, we basically have no idea because acting in open worlds requires a causal and semantic understanding. We don't have a handle on how to do that. My best guess is that semantics and natural preferences can only emerge when intelligence is embodied, when there is way for mistakes to actually "hurt". I don't know that a value function representation, however complex, can substitute for that.

Driving is in in itself an open world. Self-driving cars will be solved by turning them into trains, an effectively closed world. Infrastructure functionally equivalent to rails (e.g. guiding wires + inter-car communication) will be put in when major streets are re-paved, and we'll take the open-world automation risk on the slow side stretches. This might cost trillions, but the incentive is at least as large.

Most posts on automation assume an all-or-nothing proposal, that the machine either consumes the job entirely, or not at all.
But most of what we are actually witnessing is a partial consumption, where the machine or software de-skills the job.

Consider cashiers. At one time say in the 1950's, operating a cash register was a complex task that required a fair amount of skill. As the machines grew more complex, the amount of skill needed decreased. Now we are at the point where the "skill" is merely swiping an object across the scanner and one lonely employee stands idly by, waiting to assist the confused.

Driverless trucks will still have truckers; they just will be relegated to the simple low skill tasks, while the machine and software do the high skill stuff.

Machine learning and software is to white collar professionals in the 21st century, what motive power was to blue collar workers in the 20th.
Most of what professionals like doctors and lawyers and engineers do is remarkably adaptable to software; A large part of our job consists of analyzing data, performing logical assessments and searching for patterns, tasks that already are be absorbed by software.

Ironically, to the extent that doctoring, lawyering, and engineering can be taught in schools, the jobs can be automated.

I was thinking about this last time I ordered something and it was shipped fedex across the country. I was tracking it on the website, and the package literally never stopped moving in 3 days. Clearly the truckers have a system worked out so that cargo gets transferred to new hands when the driver needs to sleep. This is probably mostly an internal fedex thing, but I imagine private truckers could coordinate something similar, by swapping trailers with trusted partners.

Its done through the doubles and triples you see on the interstates. They can drop one of the trailers for another driver to pickup and move on with the remaining trailers to other locations. But even with this process those drivers are pushed to stay in a very tight schedule (which is why those guys are the ones you see driving like the devil is on their heels).

It’s a well known rule of thumb that it’s better to clean up and standardize your work processes before you automate them. It’s extremely difficult to understand the as-is process, and design the “improved” to-be process in one step. And it’s often quite amazing how much improvement comes from evo,vine to the to-be process, ever before any automation is done.

An interesting question is whether any of the trucking/logistics firms are making systematic efforts to improve their operations prior to automation (for example by separating the long haul and local driving tasks).

I don't know if the trucking companies are but surely the big companies in related markets (FedEx, UPS, Walmart, Amazon) have been furiously studying trucking logistics for years. How much of their research and day-to-day experience carries over to the independent trucking industry, I don't know.

Excellent point. Notably, it is often hard to see where a process has its performance ceiling. Think product demand forecasting: how accurate CAN you be, and how much would that save in inventory holding cost? Thus the need to improve that process is not obvious until you try. That often means doing exploratory data science.

One pretty typical data science outcome is that the data scientist tries to wade through the data which captures the process, and finds poor data quality, sparse coverage, accumulated legacy kludges, etc. Visualizations end up created that point out the problems and improve visibility. We still haven't built a better forecaster, but now the data is cleaner and the business owners see more of the process.

More often than not, the greatest value of "the AI push" is to serve as a motivator to drain the data swamp.

"It’s a well known rule of thumb that it’s better to clean up and standardize your work processes before you automate them."

Yes, and thumbs up.

An interesting question is whether any of the trucking/logistics firms are making systematic efforts to improve their operations prior to automation (for example by separating the long haul and local driving tasks).

Pretty much all of them are. Trucking/logisitics is an extremely competitive business. One of the problems they are running up against though is that maximizing operational efficiency is seriously counter-productive to recruiting and retention. Basically, there's a certain personality type that makes for good OTR drivers. It's one that doesn't like being micromanaged. So the drivers will welcome a routing tool that identifies optimal fuel stops, but start telling the driver where they MUST stop, and you'll start seeing the better drivers (and some of the lousy ones as well) say "screw this" and leave. It's made worse by the ever present conflict between field people out there on the ground and the REMF's sitting in their comfy offices without a clue. (Note: Characterization provided from the driver's perspective.)

aww nuts, missed my closing tag. Sorry 'bout the overabundance of bold.

This can also be read as why disruption is easier than incremental improvement. Don't worry about "what a factory needs to do to make itself ready to be automated" just make sure the inputs and outputs are the same and rewrite all the rules in between.

I think this perspective is skewed towards a particular type of trucking, sole operators. A significant portion of transit within large trucking companies takes place within their network. The driver is picking up a trailer at service center A and dropping it off or swapping at service center B. This arena negates almost all of the author's objections. No loading, no securing, no verifying, no customer interaction, no cash on delivery. Often these trailers are sealed the entire time they are the responsibility of a driver.

In my mind, shuttle routes where a single driver goes from A to B and back to A in a single shift (at night) are the low hanging fruit of automated trucking. Far fewer variables than are being suggested here.

One of the easier things to automate should be middle managers -- who accept front line managers' reports, and summarize them for upper management.

It might be happening, but if so it's pretty slow. Still, the data collection and analyzing tools are getting better and easier to use.

More middle managers will see a squeeze and more companies will have higher percentage reductions of middle managers than truck driver employment reduction in the next 5 years.

So you automate driving on the trucks. How many automated trucks could one "driver" manage in a convoy at once? The length of semi trailers is limited because you don't want thousand foot long vehicles on highways, but 15 trucks in a row?

80' X 15 = 1,200 feet.

Someone may have mentioned this, but if you live in the west, California in particular, you are probably aware of chain controls on I80, US50, The Grapevine, Conway Summit on US395. When controls are implemented by the CHP there are guys who turnout to put chains on for motorists, but often 18 wheelers are chained up by their drivers. I'm sure this function and most of the others can eventually be dealt with on behalf of autonomous trucks, but it's just one more example of an idea not being as simple as may sound.

I suspect the nearest automation in trucking will be driverless convoys. The lead truck will have a driver but the next truck (and maybe 2-5 behind) will be programmed simply to follow the truck in front. They will be able to draft incredibly close behind each other saving gas but a real person will oversee the complicated stuff. Basically a train on the highway. Hard to pass though.

I'm an automation guy too. I also know truck drivers. I can tell you Dan Hanson is 100% correct on this.

I think a lot of office work will be incrementally automated in the coming 20 years. The last 40 years was the automation of factories and blue collar work. The next few decades will be the automation of mostly office and white collar work.

There will also be lots of job opportunities for good maintenance/repair people with good electrical (control panels, etc., not just standard commercial) and hydraulic skills. Do you guys have any idea how hard it is to find someone who can work with hydraulics? And, no, electrical is not completely replacing hydraulics any time soon.

"The last 40 years was the automation of factories and blue collar work. The next few decades will be the automation of mostly office and white collar work."

An awful lot of this has already happened. Mainframe computers in the 1950s and 1960s replaced the vast armies of clerks doing (for example) insurance company billing. Word processing in the 1970s thru 1990s made secretaries an endangered species. CAD systems largely replaced the work of draftsmen and tracers in engineering offices. Programming itself has been substantially automated via compilers, standardized database systems, prepackaged software, etc.

I would also say that we're in a tech bubble. Lots of visionary dreams are bandied about during tech bubbles, and go away once the tech bubble deflates.

The ideas that aren't economical go away, the ideas that are economical will (barring legal impediment or collusion) eventually be implemented.

The idea that you can automate a truck should take a place in line behind automating grocery store scanning. It turns out that telling the difference between fruits, or bread products, is more difficult than weight plus picture can accommodate: every line with scanners has menus to match up these products with prices. They have been working on this problem since the early 1990s, and they still have the problem.

The most automated and sophisticated method of transportation are commercial airplanes.
Guess what, they ALL still have TWO pilots despite the fact that if all goes well, the pilots are not needed at all.
So why have pilots?
Because sh^t happens , things malfunction and things ALWAYS arise that the aero engineers did not envision or thought had a zero percent chance of occurrence.
You know, like wining the Powerball Lottery where the odds of winning are 1 in 100 million, which of course, is why there is never, ever a winner (please tell all previous winners that they really did not win anything and please return all the money they did not win).

And air traffic is a joke compared to road traffic.

All trucks and cars will become more automated; no doubt about it, but it will be a long time before drivers are considered unnecessary.

Theres a problem I see with making that comparison though. In trucking, or any road based transport for that matter, there are opportunities (though not always immediately) for vehicles to pull off the road and stop to prevent danger due to a malfunctioning or struggling vehicle taking up road space. In the air, there isnt an ability to just stop and inspect. The danger/impact of problems in the air is much more severe , despite far less traffic and not having to deal with changing surface conditions. I predict that passenger air will be the holdout for manned transport out of trucking, rail, sea, and air...

I couldn't agree more. In my early working life, I worked as an industrial electrician's apprentice on the construction of industrial facilities, and I saw how even the simplest material handling processes took a tremendous amount of engineering effort and tinkering to automate. I will always vividly remember the glorious chaos that happened the day a meticulously engineered automated sawmill I worked on was turned on for the first time. The new log line used a variety of technologies that were still fairly novel, including laser scanners and a new generation of programmable logic controllers running that latest ladder logic software. The expectations were that this would be technological triumph. Instead, logs were driven vertically through gang saw housings, planks were sent cascading off chain conveyors, and piled up like jackstraws, and sawdust piled up chest deep around machines within minutes while the waste conveyors remained stubbornly empty. To make matters worse, we quickly discovered that logs behaved very differently depending on how wet or cold they were. The entire mess had to be painstakingly revised piece by piece until it worked properly, and the final process required more human operators than the engineers had originally planned for.

I learned a few lessons from this:

1) Automating even simple process involving physical materials is much tougher than it looks
2) Physics is a bitch, and mass plus inertia is deadly
3) Nothing is as simple as you think it is
4) Sometimes, removing a human from a loop is just too expensive to be worthwhile

All valuable lessons and true.

"Sometimes, removing a human from a loop is just too expensive to be worthwhile"

This is the most important lesson. Successful automation isn't about removing the operator. It's about automating well defined tasks. At the end of the day you might have 10% of the operators remaining, but they'll still be there.

You are 100% correct that unforseen issues are extremely commonplace. As other commenters mentioned, however, we still seem to have an all or nothing approach to automation. Systems are always refined and improved as time goes on, and despite often rocky beginnings, the kinks are ironed out, unless of course, they are so severe there is no economically feasible way to do so. The issue as to the workforce and employment at large isnt the "all human workers will be replaced", but its the issue of having significant numbers of humans replaced, and with no expansion elswhere. Commenters have mentioned multiple times about real world examples of systems (usually manufacturing) increasing productivity while drastically decreasing needed manpower. At some point, this pool of displaced workers cannot fill the empty job spaces, either due to too much employee competition or lack of qualification.

One area that automation will take over easily is regular route trucking. This is on a fixed route with a daily frequency (normally called LTL (lighter than truckload)).
The other class of trucking is irregular route trucking with delivery times that are unknown until the load is picked up. Because of routing which is not on the interstate system most of the time, the algorithm would be too complex to consider. The driver is responsible for knowledge of every state and town that he/she drives through.

The author sets up a bit of a strawman: Truck automation is really hard because ALL the things that a trucker currently does cannot easily be automated.

That is the wrong way to look at it. The right was is to ask is the current mix of tasks that constitute trucking performed most optimally by a single human trucker or by a mix of automation and humans doing other tasks?

For example, consider the truck driver's diagnostic capabilities. Sensors throughout a truck can track everything from tire pressure to the integrity of the memory of onboard computers. A computer can process much more sensor data and make more accurate calculations about likely failure than a human can. And should a vehicle fail, there are numerous ways in which it could notify others of the hazard. The simplest would be to have a non-driver human occupant that rides along precisely for this circumstance. (To say nothing of the fact that if most vehicles are automated and all are equipped with appropriate communications technology, there would be no need for physical flares, because virtual ones will appear in a automated vehicle's guidance and - for the human operated one's - head's up displays)

Similarly, loading and unloading can be done by humans whose specific job is to do just that. Cameras and other sensors in the truck can observe and upload the state of a loaded vehicle to the shipping company's central servers. There, detailed and complex analysis can be done to determine the correctness and efficiency of the load. Again, these sensors would "see" far more than a human can. What human could detect slightly elevated temperatures coming from a crate of lithium batteries, or low levels of radiation?

Computers too already have far more detailed and up-to-date information of road and traffic conditions than any human driver, and this will only continue to improve.

Finally, the book-keeping and scheduling tasks mentioned are most likely already being performed by the trucker's laptop or tablet.

Cameras and other sensors in the truck can observe and upload the state of a loaded vehicle to the shipping company’s central servers.

snort!! gufffawwwwwwww!!!! HAHAHAHAHAHAHAHAAHAH!!!! Here's a hint as to the challenges to your pipe dream.

Most of the trucks are not, in fact, trucks. They are tractors. You have to have the cameras and sensors in the TRAILERS. The scope of the problem can be illustrated by this:

In railroading, there is a dream of simultaenous braking. This is only a dream thus far. At it's core, railroad braking technology hasn't advanced since George Westinghouse invented the air brake in the 1870s(?). Air brakes keep the brakes OPEN, i.e. NOT applied as long as the system is pressurized. This way, if there's a break in the system somewhere, the brakes AUTOMATICALLY come on. It is a darn good system, evidenced by the fact that it's still in use more than a century later, and is also the braking system of choice for tractor-trailer rigs.

BUT, there's a serious shortcoming to it when you're dealing with a train a half mile or more long. Lag. When the engineer applies the brakes up there in the locomotive, it takes a while (we're talking multiple seconds) before the end of the train gets a clue. So while the front of the train is braking, the rear of the train is merrily rolling along. Easy to solve, right? Just use electronics to activate all the brakes at the same time. Except for one problem.

EVERY car on the train will have to be equipped with effectively the SAME electronics. Now, this can work for small carriers who don't interchange, i.e. Amtrak and commuter rail and subways, and also sorta maybe for unit trains. But for normal freight traffic? 'Tis a monumental challenge. Remember EVERY car has to have it. SAME system. And it has to work flawlessly, because if 3 cars in the middle of the train don't get the message, the train can be RIPPED APART. (FYI: That would be bad.)

Now, compare that situation to the situation in trucking, where there are about 15 times (conservative estimate) as many trailers (aka "cars") and a bajillion times as many owners. Then throw in the fact that this is a HIGH TECH solution, which means before even 20% of the fleet is equipped the tech will have changed enough to render the first 10% obsolete... rinse, repeat.

Now, understand, I'm not saying that it CAN'T be done, simply that it will be far, far more expensive and challenging than most people realize. The only advantage that trucking has over railroads in this sort of tech adoption is that capital equipment replacement time is far shorter in trucking. And there's more very small entities to serve as guinea pigs, er, early adopters. From a technological standpoint, it is near trivial by today's standards. (Something similar is already underway with reefers, who report back to the mothership their temperature readings.) It's the compatibility, reliability and adoption that are the real world challenges. These latter three challenges is why you don't see widespread adoption of backup cameras on tractor-trailer rigs.

What will be automated soon, very soon, is long haul truck driving. Robotic trucks will depart truck stops and drive themselves to a truck stop near their final destination. Truck drivers will drive the truck between the truck stops and places where loads are picked up and dropped off.

Bureaucracies always fail, eventually. They always, ALWAYS, without fail, become systems where the rules/procedures are the only thing that matters. Building houses? Actually finishing a house does not matter, only that the correct number of TPS forms are filled out. In cornflower blue. Building software? It does not matter that it does not run, only that the procedures were followed, that enough meetings were had. Teaching children? Literacy don't matter, only that union people have paychecks and pensions. The seeds of the destruction of everything, are planted in the rules people make up, that sounded like a good idea at the time.

With automated trucking, getting loads picked up or delivered don't matter; only that fewer people are allowed to drive, and government has more control. Less freedom. More control (by the central bureaucracy of course). It is as simple as that. With tyrants, it is never about what they say it is. Safety, efficiency, just smoke to distract from the real goals. Every time somebody says "if it saves just one life", or "won't somebody please think of the children".

I drive a tractor-trailer about 20% of the time on my job. I can see robotic driving, but mostly loaded trucks being sent over-the-road - in the beginning it would probably be like trucks leave a holding area, entering a turnpike or other large road, and the robot would end the trip at another holding area near the load's destination, as John Moore says, above.

I think too it would take a relatively long time to get the bugs worked out of the system. Despite the fallibility of human drivers, the first time a robot driver killed somebody; heck - even the first time a robotic truck had a system crash and blocked traffic until a human driver or system repair person could be dispatched - there would be a great outcry from the public.

I am well aware that people are rarely swayed by numbers and stats and rather via emotions, so I assume that the idea of Robot truck is still scarier to the public than falliable man driven-truck. That being said, dont you think that as automation becomes more mainstream, the outcry against a robot truck crashing would be lessened? Especially if compared with the numbers of human crashes caused by driver error

Hi John,

"I’d make a bet with you on the matter, if the logistical considerations of 2 random folks on the ‘Net making a bet on something 20 years out were less daunting. Here’s to hoping that one or both of us recall this in 20 years."

There's the "Long Bets" website:


It costs $50 to make a prediction, and $200 from each side is the minimum bet...I think. Both the winner and the loser's money go to the charity designated by the winner.


If you're interested, we could try to come up with a prediction about which we could both agree was a good prediction on which to bet. Perhaps something like, "I predict that by 2038, more than 80 percent of the interstate trucking miles traveled will be by autonomous vehicles"....?

If you're not interested in that, I could do a post on my blog, and do $20 in twenty years. That way, even if either of us couldn't collect because we lost touch, it wouldn't particularly matter.


I don't see how those truck trains are going to work on highways. How many cars or other trucks can drive between the units of the truck train? Other vehicles are going to want to pass. Other vehicles are going to want to get over to let someone else pass or to get to an exit. Vehicles are going to want to enter the highway in a weave lane. If someone is grabbing a few hundred feet of road, they are a nuisance and a likely a hazard for everyone else. Truck trains have to be designed to work with other traffic.


I've been following autonomous vehicles since the mid-1980s. I'm willing to believe they can be made to work someday, most likely on multi-lane interstate highways in relatively uncongested areas. I think we are farther away than we think, even on this more modest goal. It isn't about cycles per second. We already have more than enough computing. It's about algorithm development. Most machine learning systems have limited semantic knowledge. They are just input/output machines, so they are rarely robust. Look at all those minor adversarial sign modifications that totally confuse them. We are going to have to develop algorithms with better introspection, so we can improve them, and to compensate for algorithmic weaknesses, we are going to have to do better multiple sensor integration.

The business model is going to have to change though. Right now, truckers own their own tractors and operate as "independent contractors". Trucking companies are going to have to own their own fleets. That moves the responsibility for maintenance, insurance and fueling. Right now, trucking companies can lean on truckers and cut their margins to improve their corporate profits. With self driving trucks, they will wind up having to scrimp on maintenance and the like themselves, without the current liability barriers. I'm guessing that the accounting will be quite different too, and the breakeven points will not be where one might expect.

I do believe that some things will change in trucking in the near future because of vehicle automation. For example maybe trucks become just partially autonomous and drive on the highway without driver assistance. Or in auto transport like my company people could start just programming their cars to drive to a new location instead of shipping it. These things I believe are coming in the next 15 years.

Comments for this post are closed