I spent a lot of time and effort deploying automation in customer service. The idea was to do it in ways that were really valuable, as opposed to spreadsheet positive. Our management had been toasted too many times by folks selling snake oil, but my god savings were needed.
They just had to be real...
So first issue: developing measures that measure real overall service productivity... this is hard, the gaming is intense.
Second issue: automation generates work. Yup. When your service workers get time they use it to address difficult cases and to do the work that removes regulatory and safety risks. The rest of the time this stuff is ignored, building hidden risk for the enterprise. Relieve the pressure, and it reemerges.
Third issue: peak demands. Ideally automation would help you deal with peak demands enabling human staffing at sensible levels to handle most of the traffic most of the time. Sadly it doesn't. Peak demand often seemed to be for the work that could least be automated.
Forth issue: the tech is a castle of lies. Ok, that's not quite true... but there is a lot of lying. In the academic work the lying is of the form of what is left out of the experiments and evaluations - for example that the algorithm cost $100k to run or something. In the commercial world there is flat out lying - HAL I look at you, you bastard. How many RFI's were derailed by some regional President at a supplier ringing my CIO or CEO and explaining that I was "a problem" and had "an attitude"... well, all of them. The issue is that there is no sanction. HAL is still pushing Holmes and winning contracts, and not delivering what they promise, because it's all a lie. All that happens is that they move to the next sucker and wait for management churn at the old sucker to erase the corporate memory, and this does not take long. In reality they should all be drummed out of the business and no one should ever speak to them again.
But, five years later I am still flogging my guts out and they are all on their boats and golf courses. One of the bastards had his own vineyard.
How many RFI's were derailed by some regional President at a supplier ringing my CIO or CEO and explaining that I was "a problem" and had "an attitude"
I had issues with this too. The only solution is to prep your CIO whenever you play hardball, and return the favor to the vendor.
For customer service I find that the real automation "gains" tend either to be in debugging the rest of the organization or in driving a customer to give up on finding a human to talk to.
This is true of automated IT processes in general; for human-operated processes there's always a certain amount of "slack" between how the process is written and what needs to be done to make it actually work. The knowledge for the latter is concentrated in lower-ranking staff - the "NCOs" - and often unknown and unknowable to those running the system. Converting the system to automation only converts the known process, so it tends to fail hard until the slack can be taken up elsewhere or learned and incorporated.
I've worked indirectly with customer support, in that we built FAQs and flows that made users do some troubleshooting themselves before contacting support - a simple couple-of-steps plan of checking your fuse boxes or checking the website for expected maintenance or known outages before calling support.
The next layer of automation would involve people going through one of those call menus; ideally once the customer ends up at an agent, they will have all their data and files right in front of them already.
Oblique references to IBM and their product Watson. I assume they're afraid IBM would otherwise find their comment and sue them for libel (implausible).
For those that need further explanation, HAL is the killer computer in the movie 2001, the name was picked for the movie because the letters are shifted by one from "IBM." IBM's product Watson is named after Sherlock Holmes's sidekick, Dr. Watson.
Edit: the product was named after a prominent, early CEO of IBM, not the character; Sherlock Holmes & Dr. Watson is what the commenter was referencing (I've thought the Watson name was a Holmes reference since it won Jeopardy).
the $100k thing is referring to the costs of deep learning, in my case these costs torpedoed an otherwise good looking business case. Other things that academics do is to report "human level" results when they mean "10 people I recruited on AWS mechanical turk, who really didn't give a monkeys about what they were doing". For academic papers indicating that technology might be valuable in the real world it's clear that they should be read as "this technology might be valuable after 5 years of development". The problem is that this tech then gets hyped by thirdparties which then obscure the 5 years of work required - sometimes meaning that the 5 years doesn't get done...
HAL is just IBM encypted using Caesar cipher / ROT25 :)
Holmes must be Watson (but "Holmes" is also the marketing name given by another tech vendor in this space to their own AI product inspired by Watson - very creative)
Automation is capital intensive to set up, so it may take years for it to pay for itself. This delays adoption considerably for smaller businesses. Hair cutting is one of those jobs that resists automation. Same for burger flipping. It's not as if salons and restaurants can afford expensive robots to automate those tasks. Customer support automation may mean lost business due to angry customers.
It occurs to me that buying a robot arm to literally flip burgers is highly anthropomorphizing the solution space. You want a machine that gives you a burger at the end. This is a pretty easy thing to automate if you had a huge, constant demand for burgers at one single location using standard factory automation strategies. Might not be feasible at all for the typical franchises with spikey demand and relatively low volume.
I say we increase the minimum wage to about $20/hour, adjusted for productivity growth (per capita GDP) since the 1960s and see what happens.
Assembling different kinds of sandwich is the kind of problem where a robot arm may be among the best solutions. But yeah, it's bad for flipping burgers alone.
Burger assembling may resist, but I know of at least two burger chains that use conveyor charbroilers where you put the patties in on one end and they come out nicely cooked on the other. I don't think those have a large capital cost either.
McDonald's is always inovating on labor reduction. They've got robots pouring drinks now, but they also used to have the size buttons to dispense a specific amount of soda which reduced labor. When you add up enough savings, you can run the restaurant with one less person; which is handy.
Still the best solution is to combine robotics with humans. Humans are still infinitely flexible and much easier to dispose off than an expensive robot /sarcasm.
There's a 1959 Asimov short story where a futuristic government overrelying on automation realizes they can replace expensive computers driving ships and missiles with humans to save on expenditures.
You laugh, but that was literally the strategy behind XCOR and Virgin Galactic picking non-automated, non-fly-by-wire guidance systems for their space vehicles. The regulatory overhead and avionics development costs of using uncrewed systems is far higher than for crewed systems with test pilots. Or at least, that was the argument in the late 90s and early 2000s.
But only with regards to health issues that affect their workplace performance. I don't think they are covered for problems that may affect them outside of normal working hours. Speaking of which, I don't think their work/life balance is that great.
Not until the [redacted] laws, when they decided humans should work at least 10 hours a day and robots 1 hour a day, given that they produce 10x more work in the same amount of time.
Ironically, that’s sort of drone for drones in aviation. Any person can just fly an ultralight. Without any kind of pilot’s license. And without nearly the kind of regulatory overhead of even smaller drones.
There is definatley something here in terms of using human intelligence to control physical actions, perhaps as a prelude to full automation for a given task.
It also opens up interesting questions as to the future of the labour markets around the world where somebody in one country can control the unit of production in another, I mean we are comfortable with software outsourcing so perhaps this is not such a reach.
In my experience, the best bits of automation are those set up by the people doing the tasks, once given autonomy (heh) to do it themselves.
Then, of course, they get promoted, the tech becomes a "product" and they try to push it on other departments, but the original grassroots efforts by software engineers doing things other than software engineering (e.g., operating spacecraft) is solid gold.
>I have not seen good examples of companies successfully deploying robotic systems in low-margin, public-facing settings.
— Matt Beane, University of California, Santa Barbara
Me neither, which has kept me at the opposite end of the spectrum for decades.
To this day, good apps linger where it's still cheaper to use a human operator.
Going to build me a robot anyway, as soon as it's complete it'll be doing an invoiceable job.
That's a key.
So it starts making money right away, admittedly not as much as having a human performing its tasks.
Human's going to have to fill in when the robot is down anyway, plus serve the robot instead at intervals when it is running, just to keep it fully supplied and maintained.
So it's going to require a higher-skilled human than before.
That won't be cheap but I can then sell the higher skilled output for a better premium to clients most interested in the robot.
Ok for the cases where they found the robots weren't flexible enough. But the "people are cheaper anyway" that they add on nearly every anecdote isn't reassuring at all.
What do you expect to be better in 30 years? We have NOW insane computing power, cool stuff like insane size FPGAs and super fast GPUs. Mechanical parts work with fraction of micron accuracy. And yet we have millions of workers picking simple shaped objects manually every day around the world.
> We have NOW insane computing power, cool stuff like insane size FPGAs and super fast GPUs.
They are still slow, compared to human brains. To compete with biological neurons, computers may have to be 100x-1000x faster.
Your phone is massively more powerful than a DEC Alpha based workstation from 1992 and it can fit in your pocket. Moore's law may be slowing, but I still expect compute to be 100x-1000x performance/$ and performance/W in 30 years.
Software is already on its way into a paradigm shift, where it will increasingly be written by AI (look at the state of github copilot). Hardware (both eletronic and mechanical) design will continue to see similar developments.
At some point, spot-like robots will reach a price where it starts to become affordable by some households. That moment will be similar to the appearance of PC's in the 80s or phones in the 2000x. As the market explodes, prices will drop really quickly.
And even before that, militaries will drive development. Drones (ie robots) are already becoming a key factor in the war in Ukraine.
Battery tech is the main limitation at the moment. Though I do expect significantly better energy density over 30 years, and reduction in power consumption should make up the rest.
So basially, I expect robots will have a development over the next 30 years similar to how PC's developed from 1980-2000 or phones developed from 2000-2020, if not more.
We'll have increasingly more incentive if birthrates (and thus the size of the workforce) continues to shrink. Moreover, robots of all sorts will probably have come down in price which is one of the stumbling blocks mentioned in the article.
Robots are too expensive; that is to say; humans are (a lot) cheaper. If you want to buy a Spot robot to pick up garbage, you are out 70k for the base model, 100k+ with some plugins, they need to charge every hour so you need multiple and there are maintenance costs (it’s not like you buy one of these and have no other costs outside electric for 10 years). In many countries humans are a lot cheaper than that for those manual jobs; depending on the work which determines the maintenance costs, maybe even yearly.
We either need to acknowledge humans are more valuable (which is utopia and won’t happen easily) or have high quality robots for 5k or less with 8-10 hour battery life (or hot-swappable; you buy an extra robot that just runs around bringing batteries to and from robots and chargers). And, as with all mass production, that will take time.
> If you want to buy a Spot robot to pick up garbage
All of the things you say are true. There is one additional thing you don’t mention:
You can drum up people on minimal wage, you give them a roll of bin bags and point at a street and tell them (verbally with words) “pick up the litter on that street, once the bags are full put it in the truck. At the end of the day drive to <address> for disposal” And it will with high probability happen!
Yeah they might goof off and you might need a foreman with a slightly higher wage to keep them busy, but that is basically all the “programing” you need to do.
How would the same work with a robot? Someone needs to program it. It is not unreasonable to think that one day we might have a “general trash collection” AI you can licence, but it is a lot of work to develop that system.
Whoever programs that system will need to solve issues such as:
- cover the whole street, but don’t wander into private estabilishments, peoples gardens or block the street
- pick up cigarete butts from the planters, but not the flowers, but do pick up the dried up fallen leaves/petals
- pick up the half shoe a reveller left on the corner, but don’t pick up the street performer’s hat with his coins in it
- hastle / don’t hastle the homeless based on opaque rules subject to change all the time
- shovel up the dirt from the pavement, but don’t try to dig a hole to china in the spots where grass should be but it dried out and it is basically just a patch of dirt
- empty the street bins, even the one which looks a bit odd (it was a one-off pilot project and we decided to go with a different suplier eventually) or the one which is hard to open (give it a good thunk, it jams since someone drove into it 5 months ago)
- leave the traffic cones be, but bag that one which was run over by a truck, dragged out of place, and got crumpled
And probably 15 other edge cases I can’t even think right now. And when your cleaner robot gets into a bad situation it probably won’t learn to avoid the same situation the next time.
And what happens if something changes? Let’s say you now want to collect aluminium into a separate bag, or there is a street festival with different temporary rules, or something?
To reprogram your robots you need highly trained, specialist crew, costing big bucks.
If your human crew needs “reprograming” you can email a middle manager paid 120% minimal wage who will make them change their way.
And all of this is “just software”. You have all this hassle even after all the battery and hardware issues you mentioned.
We might get street cleaning robots one day, but it is a very though task and i wouldn’t hold my breath.
Yes exactly. Jobs that are labelled "low wage" are often incredibly subtle and complex. And people doing them actually can get better at them with time, and more efficient, too. I remember the story of a lady whose job was to dispose cakes in boxes, and make assortments (boxes of 4, 6, 8 cakes, with 10 or 12 different type of cakes to choose from: cherry tarts, lemon pies, chocolate éclairs etc). No robot is able to do this, and it's already probably quite less complex than cleaning up a street taken at random.
Another one: solid state batteries. It'd be quite surprising if they're not commercially ready in ~15 years and they should offer at least 2x energy density, better safety, faster charging, more charge/discharge cycles. All would be huge for mobile robotics. There are big expectations: QuantumScape has a market cap of over $3B, Solid Power over $1B, etc.
I agree with your sentiment but just a nitpick - tasks considered "easy" by humans aren't necessarily easy to automate and vice versa. Eg. beating a computer at recognizing trash vs beating a pocket calculator at 6734100522 * 714261898941.
> And yet we have millions of workers picking simple shaped objects manually every day around the world.
It's designed to be like that
> Automation, the most advanced sector of modern industry as well as the model which perfectly sums up its practice, drives the commodity world toward the following contradiction: the technical equipment which objectively eliminates labor must at the same time preserve labor as a commodity and as the only source of the commodity. If the social labor (time) engaged by the society is not to diminish because of automation (or any other less extreme form of increasing the productivity of labor), then new jobs have to be created. Services, the tertiary sector, swell the ranks of the army of distribution and are a eulogy to the current commodities; the additional forces which are mobilized just happen to be suitable for the organization of redundant labor required by the artificial needs for such commodities.
The problem is there's a lot of cheap labour that is more competitive than these robotics systems.
This is bad as it removes the incentive for innovation and keeps productivity low (profit per hour worked) so that ensures wages will stay low - it's bad for the workers and bad for the economy as a whole.
How exactly will that happen? Most Western governments are happy to import immigrants after their own populace wakes up to their bullshit and stops tolerating it. Many immigrants don't mind working for 20% of what a local would work.
I don't see us running out of Indians, Chinese and Africans in the next 200 or so years at least.
Strangely enough a former co worker at the Angelika is now doing video archives. In a local interview they asked him the most interesting thing he’s found so far. His response:
“You’d be surprised to see we’re still debating and arguing over the same issues we had in the 60s, 70z, 80s…”
Basically same shit different generation - so far.
I think they are being held back by trying to make them human-like. I doubt that people will find plastic attractive. instead they should be focusing on sensation
> If getting jiggy with an industrial machine that could tear you in half is what floats your boat...
People already get jiggy with other people who could easily bite off important bits.
The issue is trust, and I think that even with the simplest of failsafes[1], the machine juggling your important bits can be made more trustworthy than the human currently doing it.
[1] For example, failing in an off position, or using equipment that, even if overdriven at full force, will not have enough force to do damage. Having the clamping or piston mechanism driven off a spring can ensure that the maximum force applied depends on the stiffness of the spring, which gets softer with age, not harder.
There's multiple ways currently being used to make machines that interact with humans safe for humans no matter what happens.
Even if they make perfectly human-like, realistically acting sexbots with HW failsafes, there's the question of trusting the SW. It would be inadvisable to use such a black-box proprietary machine which is probably recording and sending detailed (and potentially embarrassing) telemetry. I'd expect a long wait for community-designed, open-source equivalents.
They just had to be real...
So first issue: developing measures that measure real overall service productivity... this is hard, the gaming is intense.
Second issue: automation generates work. Yup. When your service workers get time they use it to address difficult cases and to do the work that removes regulatory and safety risks. The rest of the time this stuff is ignored, building hidden risk for the enterprise. Relieve the pressure, and it reemerges.
Third issue: peak demands. Ideally automation would help you deal with peak demands enabling human staffing at sensible levels to handle most of the traffic most of the time. Sadly it doesn't. Peak demand often seemed to be for the work that could least be automated.
Forth issue: the tech is a castle of lies. Ok, that's not quite true... but there is a lot of lying. In the academic work the lying is of the form of what is left out of the experiments and evaluations - for example that the algorithm cost $100k to run or something. In the commercial world there is flat out lying - HAL I look at you, you bastard. How many RFI's were derailed by some regional President at a supplier ringing my CIO or CEO and explaining that I was "a problem" and had "an attitude"... well, all of them. The issue is that there is no sanction. HAL is still pushing Holmes and winning contracts, and not delivering what they promise, because it's all a lie. All that happens is that they move to the next sucker and wait for management churn at the old sucker to erase the corporate memory, and this does not take long. In reality they should all be drummed out of the business and no one should ever speak to them again.
But, five years later I am still flogging my guts out and they are all on their boats and golf courses. One of the bastards had his own vineyard.
So, sigh