This will probably result in very few jobs. The number will decline as the technology gets better, too. The USPS's first machine sorting system had tens of thousands of people keying in the ZIP codes as envelopes went by. Then that was automated for machine-addressed mail with specific fonts. Then for all typed and printed ZIP codes. Then for all typed and printed full addresses. Then for clearer handwritten addresses. Now there's just one national center where a few people look at images of illegible addresses.[1]
Right, but in industries that are not replaced (like mail), you're forgetting about the effect increased efficiency will have on demand. A counterexample would be when looms increased employment in textiles. Less people were needed per item, but it was so much cheaper to make textiles that the increased demand ended up employing more people in textile mills than ever.
This is fascinating. Can you please point me somewhere I can learn more about automation increasing employment? When I google it I just get a bunch of highly political pieces about robots stealing jobs.
There's a classic book from almost a century ago, "Chapters on Machinery and Labor", which lists the three usual outcomes from mechanization. It's still relevant.
The "good case" was typesetting. The introduction of the Linotype enormously increased the volume of material printed and created more printing and typesetting jobs.
The "medium case" was bottle making. Hand-made bottles were expensive, taking a trained group of about five people to blow a bottle. Machine-made made bottles were cheaper, and required far fewer people to make. Cheaper bottles increased demand for bottles, but, overall, employment in bottle making was about the same or less. Also, much of the labor was now low-skill machine tending - putting in sand, taking out bottles. Only a small number of bottle-making machine experts were needed.
The "bad case" was the stone planer. Brick buildings used to have stone lintels over doors and windows, and there was an industry of big guys with chisels hammering out those things. The stone planer was simply a big steam powered planer for stone slabs. This required far fewer people. But it didn't increase demand for stone lintels, because they were a minor building component, not significant enough in cost to increase demand for brick buildings.
> So anyways, my next post will explain why this is all a lie and you’re actually maybe doomed to lose your job to automation. If you’ve found the problem with this analysis already, feel free to leave it in the comments.
Any computer person who has automated their jobs to the point fires aren't flaring up every day and their bosses and coworkers wonder what they do all day has felt the reality of automation killing jobs.
I hope they do not start using remote controlled cars on highways.
A car going at 60mph (~100kmh) travels at 26m/second. 4G has a ping of ~70ms* on top of the time from the cell tower to the operator's computer.
However, I found quite a variance in pings when browsing on the phone, anywhere from 200-300ms at times (maybe due to changing the tower).
So round trip might be at 600ms. Then we account for variance in bandwidth, human reaction time etc.
All in all we might be looking at a delay of at least 1s. Guesstimating, of course. That's still 26m of a 2 ton vehicle going at 100kmh without oversight.
If you need to have enough people that someone can intervene "instantly" then there's not likely to be much of a saving anyway.
Instead imagine a system that is "good enough" that at it's most risk averse configuration it will drive itself in normal conditions, but occasionally it will seem to slow down for now good reason, or even stop, while it waits for a human to intervene because its not sure if its correctly reading the environment ahead.
It would need to be good enough to be able to work well enough of the time to be able to reliably come to a safe stop, but it would mean you'd be able to have one person monitor many vehicles.
This is the supermarket self-service checkout model. Nobody likes them, because most of the time it doesn't work and then you have to stand around waiting for someone to intervene.
Interesting. Here in Norway I have nothing but good experience with self-checkout. It just works. Only thing is if I buy alcohol I have to wait for someone to click that I am over 18. It even has less queue, and queues are the worst!
In the Netherlands they've closed all but one regular check out line, and instead of paying cashiers they pay people to tell you to go to the automatic one. The reliability of the self check out (how often it craps itself and waits for assistance) really depends on the supermarket chain you're at though. At some it never breaks, at some it breaks almost every time (so hardly anyone uses it there).
I've found it highly depends on the store and how recently the machines were installed. My Walmart that just got renovated with self-checkout lines has snappy machines and a very intelligent weight system (often allows you to take items off the scale without supervisor intervention), while a Kroger that has had the machines for at least 5 years can be super slow and may take ~3 seconds for the scale to detect a new item.
I now tend to prefer to go to the self service, because it's usually faster than the normal checkout.
The exception is when I know I will trigger an age check.
But to the car comparison: If that happens enough that people won't use them, then they're not good enough unless it drives the cost down enough that people decides it's worth the wait.
Latency could be significantly reduced with 5G, but would still be a problem if the operators are in other countries.
But, I think the bigger problem is coverage and reliability. Relying on mobile networks for this type of life-or-death functionality strikes me as extremely dangerous, and that would remain true even if we increased reliability by a factor of three.
quick internet search says 5g is mainly a range of 1K, usa is 9.834 square km, yes this doesn't cover areas that have no roads but lets call it 1/4, that's still around 2.2M towers you'd have to add to make this feasible. Plus all the support infra. But hey at least we'll have power everywhere so we can put super chargers at every one of these towers!
I don't think realtime control will be much of a thing; keep in mind that companies don't actually want to pay people, that's why they're investing so heavily in automation and robots. At best it's going to be humans that will direct, tell a vehicle how to handle a situation it doesn't know yet. But all of the basics - driving, steering, avoiding obstacles - should and will be automated.
This is why connected cars are hoping for a quick spread of 5G. A fully-fledged 5G network would have transmitters placed on every other lamp post with a stable connection to the backbone. With such a system, the rtt should come down to sub-50ms for high priority traffic if properly set up. The obvious downside is the amount of transmitters and fiber you need to set up for this, but the smart vehicle people are seemingly quite confident in their ability to convince operators to install such a network.
A 200ms round-trip would still be comparable to some drivers on the highways right now, but it's certainly suboptimal.
There's no such thing as a stable connection to the backbone. That's a myth. A single construction crew digging in the wrong spot can take out a whole bundle of fibers. Now your 5G network has a huge dead zone.
I know that's a bit tongue in cheek there but what a fascinating idea. Programmatically transferring liability to different locations or entities that are better able to diffuse the consequences.
Imagine a HFT bidding market like adwords. Guilt as a service. Wonderfully dystopian. Paging Mr. Doctorow....
That almost sounds like what happens when you look at a text message while driving. Your eyes are off the road for only a second. We already know the dangers of texting while driving.
But, I there are scenarios where remote operation is safe. Maneuvering in a tight alley. Any scenarios that don't require speed and where a failure scenario is bricking an already stationary vehicle. It could also be identifying obstacles (car thinks steam venting from the sewer is a solid object) or approving rule-breaking (allowed to place a wheel on the sidewalk.
When Shuri was basically live-driving cars in South Korea from her home base in Wakanda in Black Panther, I thought, "wow, those Wakandans must really be advanced! They got the latency on a transcontinental wireless link down to nothing!"
Makes you wonder how Wakanda flew under the radar while simultaneously setting up advanced tech in other parts of the world. I assume they would have launched satellites and laid pipe and what not. But then it's similar to Pym particles or Tony figuring out time travel in a night.
It's a lot easier to dehumanize a person when you are sitting behind the screen of Twitter / Instagram / UAV (unmanned air vehicle), pushing buttons...
When you rob someone of their humanity, then you too lose your humanity, making us all robots that can be disposed of anytime...
When have we not been seen as disposable. The common man has been reduced to just a cog in the machine, with the exceptions being our concerted fights for what little protections we have.
The modern person in a developed country seems to me more human than ever.
Think about the pin makers at the beginning of John Adams' "The Wealth of Nations." Atomation of meanieal tasks like these means the pin makers are replaced by educated generalists (at least to some extent) who operate and maintain machines.
...remote operation could allow companies to outsource driving, construction and service jobs to call centers in cheaper labor markets.
This terrifies me. The speed of light RTT from the west coast to India and back again is around 100msec. This is on the order of human reaction time. Real world latency we are talking way more than that. I get wacky routes to India traversing most of the world, so I get pings >400msec from the US to India. 400msec is more than enough latency to kill you. Teleop is a silly idea for big fast death machines on wheels (cars), especially if there’s significant latency in the teleop.
Not that I think it'd necessarily be a great idea, but you could probably get a decent bounded latency by using dedicated / leased lines, which is e.g. good enough for remote robot-assisted surgery in some places.
The Earth is 134ms around at the speed of light. That's how long it takes to communicate in real time with the other side of the world (your antipode). No technology or routing policies can change that.
This sort of real-time control would probably have to be done on the same continent at most, because of purely physical limitations.
I think it's worth introducing to the discussion that human reaction time to an unanticipated stimulus is somewhere around 150-250ms.
The ~140-150ms practically achievable latency over a hemispheric arc in real world IP networks is already fairly close to the limit imposed by the human element.
However, latency in IP networks is measured as RTT while human reaction time to a stimulus is given as the one way path latency.
The more accurate comparison would be between the ~70ms one way path latency of the optical terrestrial network with the same 150-250ms of one way path latency for the human to react to a directly applied stimulus.
At the extreme of the range of human factors and accounting for the differing scale, human factors are responsible for up to 80% of the reaction latency.
But if you're talking about having human operators in the loop then any communications latency would be in addition to the reaction time of the operator.
Note that "cheaper labor markets" doesn't have to mean overseas. Mexico is only a few hundred miles from many US population centers. And US federal minimum wage is still only $7.25. "Cheaper labor market" for California city jobs could just be Nevada or Utah.
Isn't that what some companies claiming to be AI/ML champions are doing when they are actually "employing" real humans through MTurk and the likes to do the work.
Yep. One of the earliest examples of this I can recall is the VoxSci voicemail to text message fiasco that unfolded over ten years ago.
They fraudulently claimed that they had some kind of machine learning system that enabled them to quickly, accurately and privately convert voice mail to text. The reality was that they had a call centre in India manually transcribing.
There is some version of this that creates a dystopian future where AI software companies shed liability by outsourcing ethical decisions to armies of contractors.
With AI currently set up as the "next big thing" this article is just another version of the "Natural Language Processing" and human augmented AI, that is being used by almost everyone to "fake it until you don't need humans any more" (see Apple, Amazon, Google listening to our audio as a backup for Siri, Alexa, and Now). A necessary stop gap.
I think there is a big opportunity to look past the stop gap and talk about technologies that are going to 10x the things humans are already good at (creativity, empathy, intuitive problem solving, ...).
I don't think "the contractor did it" will get very far, either in court or as PR?
Though, there is legal indemnification which could be used to get a contractor (the company) to pay when there is a screwup. But that's a matter of contract negotiations between companies.
My 2c, instead of building autonomous cars/robots which work in our current cities, we should just redesign our cities to cater for these new ideas. We could have an autonomous delivery system with current technology if we just put aside some dedicated space for them to operate. No fancy AI, just good old fashioned sensors and control systems. Put roughly pallet sized tunnels under the roads, and it doesnt even have to encroach on public space. Way less problems in the long run IMHO.
Yeah, it's not a far-fetched idea either. It's already been proven that dedicated lanes for buses improve road performance overall (especially improving bus performance) so at that point it should be easy to use autonomous operation as a way to justify more dedicated space.
Making tunnels that are large enough and safe enough to transport people is a much bigger and more expensive endeavor than making ~1m diameter tubes for transporting goods.
A tunnel would need to be large enough to be maintained by humans when all the pallets crashed into each other and someone had to unjam it. We'd have a new kind of plumber
is as good a place as any to start. It also has plenty of citations of its own if you want to dig further into any aspect.
Basically. the biggest cost of building a tunnel is excavation cost and the cross sectional area of a tunnel is the biggest factor affecting excavation cost for any given ground condition.
Secondly once you add people you have to add emergency escape tunnels and related systems which is just more tunnel that needs digging.
Much of the cost of the subway is the access caverns, life safety systems, etc. For example, NYC built a 2-mile water tunnel for $250M [1]. A subway tunnel costs in the ballpark of $500M-$3B per mile[2].
You might look at Elon Musk's The Boring Company. It's based on the ideas that self-driving electric vehicles permit smaller and simpler (and thus cheaper) single-lane highway tunnels, with simpler cheaper surface access, constructed rapidly at large scale (and thus cheaper), using mainstream vehicle tech, in a dedicated environment (permitting full automation, greater-than-highway speed, and vehicle coordination), woven in spacious 3D underground, with blended surface operations to whatever extent driverless driving becomes available. And thus traffic volumes/access/costs competitive with highways, buses, and subways. Perhaps picture something vaguely like Tesla passenger vans doing 100+ mph under bus routes, with van-is-the-cab elevators or ramps in parking lots.
Why is it ridiculous? If you look at Transport for London as an example, the organization employs 25,000 people, of which 3,000 are train operators, with 600+ trains. The salary + indirect wage costs for those people is on the order of $250 million per year. Which I think is pretty cheap, compared to what you'd guesstimate for CAPEX and OPEX for an automated system.
Now they are actually in the process of automating the trains, which has been criticised by many as a vanity project. Certainly it has not been shown that it makes financial sense. And it has been decided that they will retain a "captain" aboard the train, for safety reasons, as someone has to lead passenger evacuation in case of an emergency. So the total salary savings will be essentially zero.
Fully level-4 automated may makes sense for new builds. But for retrofits, you can't satisfy the overall requirements without a human physically present.
>Now they are actually in the process of automating the trains, which has been criticised by many as a vanity project.
The DLR has been driverless, since it was introduced in 1987 ─ it is a working example of the paradox of automation, which states: the more efficient the automated system, the more crucial the human contribution of the operators. Humans are less involved, but their involvement becomes more critical. This will be the unwritten hard limit for the time-being, when it comes to implementing the levels of autonomy, irrespective of the condition of the rolling stock or wage costs, which are moot.
No, it’s not ridiculous. It’s complex, time consuming, and expensive to retrofit fully automated systems on to existing lines. Like any other project, the costs (disruption, financial) must be balanced with the gains (capacity, safety). Automated systems will continue to be rolled out worldwide at a reasonable pace.
I think Swizec's point is that the current pace, even accounting for those factors, is not reasonable. Let's not pretend those decisions have some sort of mathematically correct answer; how much to invest in public transit, for example, is a political choice that one can legitimately disagree with - and even find unreasonable.
I don't think anyone is recommending remote controlled cars. It's about having a fallback to handle edge-case scenarios (e.g. toll ways, construction zones), likely at slower speeds, as AI is already quite good at reacting rapidly to sudden movements. The collusion avoidance will still be enabled while operated remotely, so a failed connection, can just result in the car pulling over and waiting.
It's interesting the effect that uber, amazon a gig working have had. Where "traditional jobs^" are created, the immediate expectation is that this will be uberified... More like selling icecream during summer tourist season than working in trucking.
^Traditional in the "standardized job that 5k people do for us."
Who will be held accountable when one of these robots inevitably harms or injures someone? Since these sorts of robots seen to be semi-automated, does this effectively abstract away any real responsibility? Can robots only be responsible for nothing more than an "unfortunate accident"?
This reminds me of "god of the gaps" idea where the receding domain of things doesn't show a clear line of stopping. It seems more like a limitation of the current implementation as the opportunity but watch out for the next version.
Concerns about ping assume all tasks will require real-time control by the human. Realistically, a supervisory system will have the human identifying hazards and gaps in the autonomous system's sensory capacity.
Quick-reaction collision avoidance will be controlled by radar (77 GHz etc.) or lidar.
It would be an interesting proposition to have the option to drive yourself or to seamlessly switch it over to autonomous driving monitored by a remote driver for $0.0X per minute.
It's an in-between step. Already they're reducing costs by making work simpler, reducing the workforce, etc. One person can operate a dozen delivery robots, vs a dozen delivery people doing the same work. It's the common trend in automation and industrialization.
That depends on where you want the money going: everybody involved in developing and maintaining automated systems at home and abroad (high-skilled labor) or everybody involved in controlling drones from afar (low-skilled labor). Of course, the amount going to either group will be drastically different depending on a multitude of factors.
If you look at the first chart in that article, bank teller employment peaked at a bit over 600,000. If you follow the link to the BLS projections given near the end of the article[1], you'll see that bank teller jobs in 2018 are down to 472,000 and expected to decline by another 58,000 over the next decade. Bank teller jobs per capita is declining more rapidly, since population is still growing.
Yes and that's a tall order considering even we humans do not even have general intelligence. Our minds are just a federation of special intelligences.
we have identified what could be automated but we haven’t taken any real steps to implement it. there is a lot of resistance when it comes to automation in the workplace.
There are actual startups that are "pretending to be robots", so the title is off: This article is about legitimate remote-operations work, and it's a really great thing, not something to be ridiculed as the title makes it. Some of them like postmates or remote driving seem like typical VC silliness but others could be lifesaving, like , imagine a remote-controlled da-vinci surgical machine.
[1] https://www.nytimes.com/2013/05/04/us/where-mail-with-illegi...