Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
The Robot Revolution Will Be Delayed
Every so often on Ricochet, I read another thread about automation, the decreasing demand for factory workers and what this bodes for the future. Or about education and training our workforce for tomorrow. Someone on these threads always asks the titular question, although I’ve never seen it put so indelicately: “Your robot factory of the future will need scientists and engineers, but not guys turning wrenches on the assembly line. What about the people who just aren’t that smart? What will they do when their jobs get automated away?”
Well, I walk the concrete for a living and I’m writing this just after my night shift support tech job let out. I’ve got a couple points to make, which the pundits don’t usually cover.
Automation Is A Diminishing Returns Relationship
That bit with Charlie Chaplin? He’s the next guy to lose his job to a machine. It’s very easy to build a robot to tighten a bolt. It’s much harder to build a robot that can feed you lunch. Setting aside the implausibility of that particular brainstorm, any particular job in a factory will range on a scale from “Very easy for a robot to do” to “extremely difficult for a robot to do.” A clever engineer will be able to put a price tag on each robot on that scale.
People Are Robots
We see the converse in science fiction all the time. But think about it from a management standpoint? You can design a really kick-butt and expensive piece of vision software to inspect your parts for defects, or you can employ a legion of low-wage drones to do your inspections for you. It’s often economical to use each for different tasks. It’s always a question of cost.
Machines Are Fallible
You know that vision software just mentioned? It will lie to you. It will tell you good parts are bad and bad parts are good. Again, it’s a cost question; how much do you lose in scrapping good product versus how much does it cost to upgrade your inspection program? It’s cost effective to back up your vision software with more people to evaluate the results. (Sure, they’ll lie to you too, but you can fire them for that.)
Case in point, I work for a company that make hard drive parts in lots of about 250,000. Early in the process, we have a machine that inspects each one for defects. It spits out a yield figure, say 90% good parts. Then, because that’s not good enough, we have several people auditing the images the machine gave us to see what happened to our product and why. This isn’t charity for the low-IQ: we do it because we make more money that way. The machine simply can’t give us the answers we need on their own.
Humans Can Cheat
From the standpoint of the guy making the rules, that isn’t always a good thing. On the other hand, rules systems need some grease around the edges to keep things moving. When Google made its self driving cars they had a problem with stop signs: the robot would follow right-of-way laws, which isn’t how intersections work in the real world. So the car would just sit there.
Our machines at work have route enforcement: they check the incoming product for the process step it ought to be on and only allows the correct material through. Then a new product number rolls around, someone forgets to set the correct permission and I get called out to override the route enforcement. You can’t build a robot that knows when to break the rules.
People Are Easy To Program
It might not seem that way when you find the toilet seat up again, but it’s easy to tell a person to do a relatively complex task and have a reasonable expectation that they’ll be able to do it. You could build a lawn-mower bot (and usher in the robot rebellion, mind you), but it’d be very difficult to then train it to clean the bathroom. Some days, your factory owner just has to say “grab a mop and come with me,” and you darn well need a human for that.
So What Does All That Add Up To?
Same thing I’ve been hammering on: it’s a question of costs. At every point, your business owner has to balance the cost of hiring people to do the task at hand, versus the cost of building and maintaining robot. Every time the robot is the right choice, they will build that robot. But in many cases, it’s just cheaper to hire some schlub to do the grunt work. Despite all the clever people designing a better robot, we’ll still have factory jobs for low-skill, uneducated, low-IQ workers for decades to come.
There’s a simple way to make sure the jobs are still there: make it cheaper to hire people. Forget wages; the real problem holding back American labor is — you guessed it! — government. There are a thousand-and-one regulations protecting worker safety and the environment, making sure people can’t be fired and that you’re hiring enough people of the preferred sort, fining you thousands of dollars because you missed a jot or tittle, and all the other myriad headaches you have to endure to run a business. Some of those things are necessary, but darn well not as many as we have right now.
Photo Credit: “Toyota Plant Ohira Sendai” by Bertel Schmitt – Own work. Licensed under Creative Commons Attribution-Share Alike 3.0 via Wikimedia Commons – http://commons.wikimedia.org/wiki/File:Toyota_Plant_Ohira_Sendai.jpg#mediaviewer/File:Toyota_Plant_Ohira_Sendai.jpg
Published in General
As I said above, humans are motivated by far more then their intellect. We are pain and pleasure motivated for example. Fear motivated, instinct motivated. The way we tend to envision a future with intelligent machines involves a great deal of projecting human traits onto them that they won’t posses.
Ones and zeros, my friend. Pleasure is the one and pain is the zero. Whether or not we can imbue robots with intelligences comparable to our own (or even comparable to a squirrel’s), we will certainly be able to simulate intelligence to a degree that the distinction between simulation and reality is blurred under common circumstances.
For example, two major factors that distinguish animal intelligence from programs are individual variation and degree of certainty.
Individual variation: Two people ask the same question and come to different conclusions. This will eventually occur in software when two adaptive programs base their calculations, animal-like, on different “experiences” (stimuli) and different information access. A program with a physical presence (a robot) may acquire unique input without direct human control simply by operating in a different environment than a fellow robot. The variation won’t be equal to the differences between human personalities, but it will be a small step on the road toward individualism in software.
As for certainty, today’s software is always certain. If a program doesn’t know an answer, it will either give up or crash while trying to solve. When animals choose the best among imperfect solutions available, that uncertainty is exhibited in emotion. This will eventually be seen in programs via hesitation, at least, when an action is stalled or stopped as the program continues to calculate and arrive at conflicting potentialities. In other words, for practical reasons, we will eventually give some machines the ability to initiate behaviors before they have finished calculating all input and potential outcomes (as is common in animal behavior). This will introduce the possibility of a machine catching itself in an “error” and changing its behavior mid-action.
Right, but when we get to that point Skynet will become self-aware and we’ll all be screwed anyway.
There is no reason for a machine to be built to feel pain or pleasure. Sensors that can identify when there is a problem with a component are useful, but will not be understood by a machine in anyway that resembles pain. Similarly, machines are not going to experience any emotions of any kind.
Fears of the singularity are based on this idea that we have no concept of how software will react when it reaches a certain level of complexity…but we do.
All assumptions about problems on this front stem from an idea that the AI will start bending and breaking rules it is given…for no reason, as it will have no will or desires of its own.
Does government regulation artificially increase the cost of labor? Of course it does.
Would eliminating those artificial costs increase employment and reduce the level of automation? For a while, yes, but not for the long term. We might not be there yet, but machines are getting much better relative to human labor and it’s not the government’s doing.
Look, I’m a lawyer. Thirty years ago I would have had someone whose full time job was to take dictation. Today, I type for myself. This is not the result of government policy. It’s the result of technological advances.
Automation isn’t just about manufacturing. Realtors, travel agents, insurance salesmen, grocery tellers, secretaries, cab dispatchers, and telephone operators have all lost ground as a result of technological advances. There used to be a lot of people employed as blacksmiths and chimney sweeps. They didn’t lose their jobs because of government.
Aside from lust for global domination.
Personally, I can’t wait until Skynet becomes self-aware. At least something around here should be self-aware. The lady in front of me at the Starbucks sure isn’t. Representative Sheila Jackson Lee sure isn’t.
If it has to be a machine, so be it.
Well, to the extent that those artificial costs led to sub-optimal outcomes then removing those costs will result, long term, in a bigger economy with more total employment.
I’d go farther and say that heartlessly running down squirrels is a feature. They’re rats with better PR.
You make a good point in that humans aren’t perfect. When evaluating driverless cars we expect them to be though. The first time one of them locks up and kills someone pandemonium breaks out, even if statistically it were safer than driving yourself.
This is an extremely good point. I’d also add that tastes change; no on can survive building the same product for fifty years, and you need people to reconfigure your robots to build the new stuff.
That’s certainly true, but it’s a slightly different question.
Designing a self-cleaning bathroom isn’t the problem. Designing a machine that will mop up after itself when it accidentally spills gallons and gallons of etchant onto the floor is more difficult. Designing a factory that will recognize roof leaks and respond to them is more difficult.
Designing a factory that will recognize every maintenance exigency it’s expected to see over a 20 year lifetime and can compensate for them all, well, you’d better wait for the AI.
I once represented a company in a wrongful termination suit. The terminated employee was a quality control technician at a glue factory. His only job was to make sure the caps were properly secured to the bottles and that the labels were on straight. He was so fallible that he managed to let 3 million units of glue ship with the labels on upside down. Sure, he got fired for it, but the employer ended up spending about a hundred grand defending against his lawsuit.
I watched us go from breadboards with wires attached to computer-aided-design-and-manufacturing (CADAM) system components. There were the early, often self-taught engineers who could determine what was wrong by listening to a device, similar to those early, often self-taught mechanics who could determine what was wrong with a car or truck by listening to it.
I believe mechanics have been largely replaced by technicians who plug into places in the engine and determine if everything is okay. Yesterday’s mechanic is not today’s employee.
We don’t seem big on vacuum tubes anymore either, although I do remember what they looked like. Had some in a black and white television.
As was noted above, becoming a telephone operator at a long distance carrier is not a womb-to-tomb job guarantee anymore. One might look upward and gravitate toward some thing greater, such as being a stewardess for TWA or PanAm.
If a lot of the smarts are external to the device, making upgrading easier, I suspect that my upgradeability time is pretty much shot. My device is not what it used to be.
I was reading a history of the UP railroad during the transition from Steam to Diesel. Paraphrasing, repair times for locomotives were similar, but the skill sets were very different. With steam, diagnosis of the problem took 30 minutes or so, but then it would take hours for the mechanic to fabricate the part to repair it. With Diesel, diagnosis might take hours, but the repair would only take a few minutes to swap out the part.
No need to wait for the AI to get that good – just hire one guy at minimum wage to sit in the factory and push the button that says “leaky roof”, or “spilled liquid”.
Breaking one of my own rules here; I don’t log in at work because then I’d spend all shift on Ricochet. But hey, main feed. Scratch that one off my bucket list.
Rather than subsidizing the engineering profession, I’d like them to kick the scholarships towards a two year electromechanical degree. It’s interesting and fun work for hands-on kind of people, and you don’t have to be a genius to do it. ‘Course, you need to have factories in the USA for that to make sense. Discussing outsourcing probably deserves it’s own thread too.
I don’t like to speculate past a certain point because I can feel my predictions getting worse and worse. I can’t speak much about AI because I haven’t been following the field. So, for you and Mr. Soto and others, a couple questions:
If we make a piece of hardware as complex as a human mind, do we expect it to exhibit behavior as complex as a human mind? If we could lay down the neurons in silicon, would the hardware be enough or would we have to program it first?
How do you teach a program to learn from experience? I’ve read about robots which learn to move that way, is it possible in the general case? What rules do you have to hardwire in? Are we being restrained by sheer hardware capability or by not solving the initial software problem?
AI certainly might change everything. Or it might not, or everything might already be changed by the time it shows up.
Granted. But all my points still apply. Replace the stenography pool with word processors and spellcheck, but you still have secretaries in the office, right? You need less of them, and their job is more complicated and variable. It’d be extremely expensive to get rid of all of them.
Whenever people lose their jobs to technology, not all of them go. Whatever the job is, the interesting parts are harder to automate. The people who are left have better jobs because of it.
The others? Need to get a new job. It’s not fun, but it happens to all of us eventually.
One of the hidden costs of employment is how hard it is to fire anybody, specifically because of this kind of situation. Maybe I should have left that parenthetical off. When people lie to you, it’s harder to detect and to react to. Worth discussing, but I didn’t want to divert from my main point that you can’t always trust what a machine tells you.
Fascinating question. Here’s a thought for you; with modern robotics it should be possible to fabricate steam engine parts relatively quickly and easily. Assuming a comeback of the steam engine (what would have to happen for that?), you could end up with the best of both worlds.
And Mr. Todd? I’m a self-taught programmer. I can watch a program execute and just from the way it slows down at what points have a decent idea of what’s screwing up. People are still people.
If you can design it that far you’re 99% of the way to functional AI. The problem isn’t the unknown knowns (we know there will liquid on the floor but not when) it’s the unknown unknowns. The Superdome lost its roof in Katrina partly because of bullet holes in the roof. Bullet holes. Do you think when they designed that stadium they considered people just hauling off and taking a shot at the roof, and what that would imply in a hurricane?
If you can design an automated factory or fast food joint so well that you can rely on one minimum wage dork to recognize emerging situations an press the right button, then you can program your automat to just go ahead and solve it without him.
/Joke
Sell enough hamburgers and you obsolete yesterday’s canoes. Now you need to fire the industry up again to make them double-wide canoes. See? Problem solved.
I manufacture electronics. I’ve got 3 (soon to be 4) production workers. They do in 1 week what my father’s company in the 90s did in a month with twice as many people. What we haven’t automated cannot be automated YET on our budget, but that’s not to say it will always be so.
Put another way, the manual processes we use today are manual only because we don’t want to invest a million bucks in automating the rest of the jobs, we just don’t have the volume to justify that scale.
Take our wiring harnesses. We could fully automate those for around $500k in machinery and another $100k in tooling. Or I can buy $50k in machinery and 2 or 3 people to spend all day running crimp presses and stuffing connectors. We don’t break even until we hit about 3x our current volumes on the higher end stuff, but the equipment does exist.
Here’s a company which specializes in automation. Starting price is $500k for systems, but they can automate, or semi-automate about anything.
http://xasinc.com
With the caveat that we really don’t know: I suspect that in order to get general purpose AI we’re going to have to program robots with motivation. That motivation may be similar to humans’ or may not be, but the AI needs a reason to continue synthesizing data without human intervention. The AI needs to have considerable latitude to reprogram itself based on input, similar to what humans do. We have some vague notion of how to do this, but the specifics continue to elude us.
Hank Rhody: #52 “I’m a self-taught programmer. I can watch a program execute and just from the way it slows down at what points have a decent idea of what’s screwing up. People are still people.”
I managed to write a few programs and then my professional programming friends told me that nobody over the age of 12 writes in Basic. I was in my 30s then. Talk about bringing me up short.
In any case, those old programs did exactly what I wanted them to do, but then I wasn’t using machine language to automate a factory floor or instructing a 3090 in an SNA environment to do the next thing it was supposed to do. I wasn’t building a complex program to build telecommunications systems and account for power, real-time, traffic, shelf space, cabinet space, and other considerations. My old friends were moving through Fortran versions, trying to keep up with faster CPUs and additional memory.
Me? I was a mere plodder.
And the kids are more able to use more technology than I am.
I am no longer technically employable.
Thank you, that is exactly the kind of point I was trying to make.
How’s the hunt for the solderer going? I worked with surface mount components a little in school, and they were frustrating to place.
I program in BASIC. Right now I spend downtime at work writing macros in Excel. Very useful, and the guy pushing the button doesn’t see the code.
Put it another way, are you not allowed to work on your car since you don’t work for a dealership with a million bucks invested in equipment? The world needs tinkerers as much as it needs professionals.
Hank Rhody: #60 “I program in BASIC. Right now I spend downtime at work writing macros in Excel. Very useful, and the guy pushing the button doesn’t see the code.
Put it another way, are you not allowed to work on your car since you don’t work for a dealership with a million bucks invested in equipment? The world needs tinkerers as much as it needs professionals.
Thank you Hank and I do appreciate your effort. However when I reach the limits of what I am capable of, I am reminded of the words of one of m favorite philosophers: A man should know his limitations.