We all think our jobs are safe from automation(technologyreview.com)
technologyreview.com
We all think our jobs are safe from automation
https://www.technologyreview.com/the-download/610067/we-all-think-our-jobs-are-safe-from-automation/?utm_source=twitter.com&utm_medium=social&utm_content=2018-01-25&utm_campaign=Technology+Review
62 comments
That's the thing. You don't have to replace every human in order for it to be scary. Current tech is perfectly capable of replacing 4 out of every 5 low-skilled workers, and that's enough to royally screw our current economic model.
Idris’ codegen comes to mind... just write the types and automation will write most of the code for you.
That doesn't fundamentally change anything, though. It just puts the code that matters in the types. Just like macros, or procedures, or assembly; it only changes how a program is specified. It does not automate away the ability to transform fuzzy business requirements into deterministic logic by describing a way to achieve them in a formal language, i.e. "programming".
It is a new layer of abstraction. A new “secret weapon”: http://www.paulgraham.com/avg.html
It is not without irony that AI is going to replace the radiologist before the cleaning lady. First field of expertise requires specialized knowledge (which AI is great at) latter requires general intelligence (which requires Artificial general intelligence and is much harder)
My mom is a radiologist nearing retirement and from the time we were old enough to seriously consider career paths she would tell us "Do not become a radiologist". As she tells it, the reason is outsourcing. Radiology is probably the single easiest medical field to outsource, and more and more of her office's work has been contracted out to doctors in India over time.
She doesn't consider AI as much of a threat... but I don't think she fully understands the leaps and bounds by which software is getting better at detecting patterns in images. As a software developer, I know that my field is still in its infancy. I wonder what it's like to look back on your profession, knowing your generation is the last of its kind.
She doesn't consider AI as much of a threat... but I don't think she fully understands the leaps and bounds by which software is getting better at detecting patterns in images. As a software developer, I know that my field is still in its infancy. I wonder what it's like to look back on your profession, knowing your generation is the last of its kind.
Yeah your moms intuition is correct but yeah perhaps she don't get just how far ML/DL is (most people aren't really)
Image/pattern recognition is already being used intensively and the only thing that keeps it from being used more is legislation and humans.
Having more than a thousand moles and having had two melanomas I am still hoping for the arrival of a system that can do complete scan of me and automatically figure out the areas of concern.
Currently I am at Sloan Memorial and using Molesafe as a more manual way to do the same. My dermatologist on of the best with regards to melanomas in the world using colonoscopy to see if there are any melanomic cells, there is even a 3model of me in blue spedos.
But I really would feel much better when they start taking the shots regularly and have the computer look for the odd ones out.
Image/pattern recognition is already being used intensively and the only thing that keeps it from being used more is legislation and humans.
Having more than a thousand moles and having had two melanomas I am still hoping for the arrival of a system that can do complete scan of me and automatically figure out the areas of concern.
Currently I am at Sloan Memorial and using Molesafe as a more manual way to do the same. My dermatologist on of the best with regards to melanomas in the world using colonoscopy to see if there are any melanomic cells, there is even a 3model of me in blue spedos.
But I really would feel much better when they start taking the shots regularly and have the computer look for the odd ones out.
I just really can't imagine, given the progression we see in AI image recognition on a daily basis these days, that medical fields like dermatology or radiology that revolve around scrutinizing things visually will continue to exist in anything but a dramatically shrunk supervisory form 20-30 years from now.
Neither can my dermatologists :)
I bet an AI would be much more likely than a janitor to know about Moravec's paradox.
Know about it? I’ll take that bet. Plenty of smart janitors out there.
Do you know many? I don't mean the janitors at Google who are probably qualified software devs who are also Google fanboys and are using it as a foot in the door. I mean genuine blue-collar never-worked-in-anything-but-manual-labour cleaners. I know a few. They're really good people, but the ones I know aren't book-smart in the same way that we are lucky enough to be, who get to sit around reading everything we can find on the internet.
Yes i know many, cleaning people too. You do know that being a bluecollar worker doesent make you dumb or ignorant and cant be booksmart right? Furthermore how many ais do you know that know about movarec let alone the paradox. How many normal academics do you know who know movarec?
Well, Roombas do exist...
They are a far cry from what you need to do actual cleaning in any professional sense.
They don't clean walls. Or toilets.
That's just a hardware problem.
With Roomba quality software a blower, an electric pressure washer, a few soap reservoirs, valves and maybe a squeegee can do the majority (by time) of a janitor's job in a bathroom more thoroughly. Being able to press "go" and have it find the floor drain, wash-down everything in the room below a set height and then herd all the water to the floor drain with the squeegee and blower and make a noise when it's done or needs to be refilled.
You're not replacing the janitor just narrowing the scope of their job. Putting up the toilet seats and letting a robot foam and pressure wash them down with some chemical you can't sell in California (so you know it must be good) is a big improvement over cleaning toilets manually, especially for anyone with back problems.
With Roomba quality software a blower, an electric pressure washer, a few soap reservoirs, valves and maybe a squeegee can do the majority (by time) of a janitor's job in a bathroom more thoroughly. Being able to press "go" and have it find the floor drain, wash-down everything in the room below a set height and then herd all the water to the floor drain with the squeegee and blower and make a noise when it's done or needs to be refilled.
You're not replacing the janitor just narrowing the scope of their job. Putting up the toilet seats and letting a robot foam and pressure wash them down with some chemical you can't sell in California (so you know it must be good) is a big improvement over cleaning toilets manually, especially for anyone with back problems.
Or corners, or windows, or sinks, or counters, or stoves, or bathtubs. Nor do they take out the trash, or even dump their own waste into the trash. Nor can they do the laundry or the dishes or dust, or fold towels, or go up and down stairs...
As a data scientist, I have the lingering fears of job automation. At least for applications-oriented work (as opposed to research). There's no reason Auto ML software couldn't take away most of the difficulty of model building in simpler scenarios. I mean, the software is already here. Just not widely used. And as computing power increases, no reason they can't tackle more of the complex problems where you might need to call in an expert. Feature generation based on the context of the industry, etc. Translation of findings into business-understandable language. One new algorithm or software package can automate much of what a data scientist might do. I mean, the unofficial slogan for some of the data science positions I've found were to "automate myself out of a job."
That's why, although I LOVE what I do, part of me thinks I need to make sure I'm in a position to move into management at some point - because I see a lot less ability for automation in areas that require "people" skills over the next few decades. On the other hand, I LOVE what I do. What's that French phrase? C'est la vie? I think it translates into "you're screwed, one way or another."
That's why, although I LOVE what I do, part of me thinks I need to make sure I'm in a position to move into management at some point - because I see a lot less ability for automation in areas that require "people" skills over the next few decades. On the other hand, I LOVE what I do. What's that French phrase? C'est la vie? I think it translates into "you're screwed, one way or another."
Machine Learning today is where web-development was twenty years ago. As more tools become available, everybody will be able to do it.
I feel general ML is already extremely accesible. There are tons of frameworks that trivialize impleneting various types of classifiers. I am actually surprized it's not more popular. I think the only thing left is to make GPU processing more available or find a cheap alternative to expensive Titan cards (maybe Google's NPUs?) and not really a requirement but abstracting away some of the configuration parameters could help the general population that doesn't know or care about the math behind.
I can already see a future, a dark future, where pre-trained models on pictures of people are sold on darkweb and you can just plug the model and hook it up to some live cam in your area and stalk anyone. This is almost if not already possible today and I can't imagine Facebook not already having something caple of this.
I can already see a future, a dark future, where pre-trained models on pictures of people are sold on darkweb and you can just plug the model and hook it up to some live cam in your area and stalk anyone. This is almost if not already possible today and I can't imagine Facebook not already having something caple of this.
I'd really love to see some GPU-like capacity integrated into the main CPU ISA so one could be certain to find those instructions and cores as you can be sure you'll find AVX.
Such a technology is the McGuffin in the last two Fast & Furious movies where it’s called God’s Eye (or The Devil’s Bumhole, by one character)
and yet - most people still can't do it (webdev), or do it poorly, the expectations are higher, often the stakes are a bit higher, and many people try to skimp out on the lowest priced vendor (or DIY). Will we see those same trends in ML over the next 10-20 years?
Another reason to strive for financial independence ASAP.
I have to wonder though: if >50% of people currently employed lose their jobs to automation... who will be there to fuel the consumerist economy? Who will buy ads on Facebook and Google that peddle widgets and services for which there no more consumers with disposable income to market them to?
I have to wonder though: if >50% of people currently employed lose their jobs to automation... who will be there to fuel the consumerist economy? Who will buy ads on Facebook and Google that peddle widgets and services for which there no more consumers with disposable income to market them to?
Another reason to strive for financial independence ASAP.
How would this work? You can’t invest in shares, if companies have no consumers. You can’t invest in property if no one can afford rent. You would need not financial independence but independence from finance
How would this work? You can’t invest in shares, if companies have no consumers. You can’t invest in property if no one can afford rent. You would need not financial independence but independence from finance
If every single job was automated... What do you think would happen?
Ideally nobody has to work and everything you'd ever want is free.
Ideally nobody has to work and everything you'd ever want is free.
> Ideally nobody has to work and everything you'd ever want is free.
but in reality, people who don't have investments in the automation will just starve and die, while people who _do_ have such investments reap many rewards.
but in reality, people who don't have investments in the automation will just starve and die, while people who _do_ have such investments reap many rewards.
I think that, given the historical evidence (specifically the french revolution being driven by the price of bread), it's more likely that there will be mass unrest and those who own the robots may get executed to cheers of those whose jobs have been automated.
And, again judging by the French revolution (and the Russian, and numerous others), an elite subset of the revolutionaries will seize control of the robots, and the situation will return pretty much to status quo ante, except for the identity of the owner of the boot stomping the average guy in the face, and their philosophical justification for their exalted position.
Yes but, even those who invest in automation need a mass of middle class consumers to buy their products.
Ford famously gave employees relatively high wages and weekends off specifically so they had money and reason (free time) to buy the very products they manufactured.
Ford famously gave employees relatively high wages and weekends off specifically so they had money and reason (free time) to buy the very products they manufactured.
I'm curious to how relative wealth (in regards to income inequality) impacts ones Standard of Living. After the bare necessities are acquired, is all 'standard of being' relative to how you perceive your place in reality?
UBI? Great question.
Is anyone out there working to automate middle management?
If you use the Dilbert comic for training data, I suppose you could get pretty far.
In fact, turns out Scott Adams wrote a post on the topic: [1]
> ... My prediction that robots will dominate management before they dominate blue collar jobs is based on The Dilbert Principle which observes that the least skilled employees are promoted to management. You need your most skilled people doing interface design, engineering, and the hard stuff. Management is mostly about optimizing resource allocation, and that is something a robot can learn relatively easily, at least compared to most skilled jobs.
> You might wonder if a robot can have enough leadership qualities to be a manager. I would point out that most humans in management have zero leadership skills, so the bar isn’t set high ...
[1] http://blog.dilbert.com/2012/12/19/the-future-of-middle-mana...
In fact, turns out Scott Adams wrote a post on the topic: [1]
> ... My prediction that robots will dominate management before they dominate blue collar jobs is based on The Dilbert Principle which observes that the least skilled employees are promoted to management. You need your most skilled people doing interface design, engineering, and the hard stuff. Management is mostly about optimizing resource allocation, and that is something a robot can learn relatively easily, at least compared to most skilled jobs.
> You might wonder if a robot can have enough leadership qualities to be a manager. I would point out that most humans in management have zero leadership skills, so the bar isn’t set high ...
[1] http://blog.dilbert.com/2012/12/19/the-future-of-middle-mana...
Yes, but they are tackling middle management skills individually. Here's some progress on one front: http://projects.wsj.com/buzzwords2014/
Chaos Monkey?
They would need to outsource middle managers first, strangely that never seems to happen
Who are the middle managers going to be managing?
There is a particular slide in this talk I gave which was mindblowing - https://www.youtube.com/watch?v=HEnFBHqHJT8
Slide 58 to be exact - https://www.slideshare.net/mintek/future-of-engineering-and-...
Between 2015 and 2022 most of the industries on the list are expected to go from a 5-10% automation rate to a 60-90% automation rate.
Slide 58 to be exact - https://www.slideshare.net/mintek/future-of-engineering-and-...
Between 2015 and 2022 most of the industries on the list are expected to go from a 5-10% automation rate to a 60-90% automation rate.
I work in digital marketing and really think my job will go in the next 10 years due to automation (if not in the next 5 years).
I'm trying to stay a step ahead and work "above the algorithm" - learning skills that will always be needed and can't be automated (creativity, team management, translating analytics into actionable insight).
It's difficult to tell how quickly the need to adapt is and the best way to adapt to automation.
I'm trying to stay a step ahead and work "above the algorithm" - learning skills that will always be needed and can't be automated (creativity, team management, translating analytics into actionable insight).
It's difficult to tell how quickly the need to adapt is and the best way to adapt to automation.
I've been thinking of getting into the field, so this is very interesting. Can you please explain a bit about how do you see digital marketing becoming automated ?
I'm guessing when they have your inputs and outputs matched in a hypercubed skinnerbox, "marketing" would seem as quaint as "alchemy" of yore.
(If that is not clear enough - a scientific way to extract all your available economic margin by ways of manipulating you. Of course in a race with competitors to the bottom, knifing each other with the same algorithms. I should write a dystopic scifi book.)
(If that is not clear enough - a scientific way to extract all your available economic margin by ways of manipulating you. Of course in a race with competitors to the bottom, knifing each other with the same algorithms. I should write a dystopic scifi book.)
There are lots of tools out there that already have automation elements built in (e.g. social media scheduling, programmatic ads). This will only increase as AI gets integrated into the plethora of adtech and martech products.
What will likely happen is that individual marketers will be able to do more with less, so marketing teams may well slim down to individuals who simply coordinate the tech involved.
There are still parts through that need a human touch - such as creative work and translating data into insights/analysis for various stakeholders.
What will likely happen is that individual marketers will be able to do more with less, so marketing teams may well slim down to individuals who simply coordinate the tech involved.
There are still parts through that need a human touch - such as creative work and translating data into insights/analysis for various stakeholders.
Creative and design may remain - but configuring and optimising CPC campaigns, content scheduling, automated comms (although the systems will have to be set up and reviewed), conversion tracking and targeting, converting assets to different formats, account managing agencies/clients etc... will all be up for grabs.
I'd recommend looking into ways to create/review/optimise automations. That way you get to blend tech and marketing, and get to lead on integration of new technologies.
I'd recommend looking into ways to create/review/optimise automations. That way you get to blend tech and marketing, and get to lead on integration of new technologies.
I don't think my job is safe from automation, and I'm an automation engineer :)
Outsourcing is a greater threat for software engineers. Why pay wages that reflect the costs of living of a city with first-world standards when you can just pick anyone in the world?
Some developer communities are very active in securing their employment by convincing the managerial tier that their yak shaving is essential work.
The problem is not that your job will be automated, the problem is that it will happen too fast and people will not have time to adapt. The disrupt mantra for startups can generate lots of money with the side effect that it hurts people's lives. If the changes comes slowly, across one generation, then people do not suffer because they already retired.
"too fast and people will not have enough time to adapt"
Thing is, this has been the case for at least 200 years, dating from the Luddites. Quoting https://en.wikipedia.org/wiki/Luddite , "Luddites feared that the time spent learning the skills of their craft would go to waste as machines would replace their role in the industry. ... The Luddites were not afraid of technology and did not attempt to eliminate technology out of fear. Their goal was instead to gain a better bargaining position with their employers."
(As a reminder, "It is a misconception that the Luddites protested against the machinery itself in an attempt to halt the progress of technology.")
Or Phil Ochs "Automation Song" from 1964, https://www.youtube.com/watch?v=WbSTnuXv_3Y
A strong labor union, or a more socialist form of government interested in supporting labor and workers, could help reduce the hurt of disruption, either by effective trade negotiations or higher taxes on those profiting from automation.
Thing is, this has been the case for at least 200 years, dating from the Luddites. Quoting https://en.wikipedia.org/wiki/Luddite , "Luddites feared that the time spent learning the skills of their craft would go to waste as machines would replace their role in the industry. ... The Luddites were not afraid of technology and did not attempt to eliminate technology out of fear. Their goal was instead to gain a better bargaining position with their employers."
(As a reminder, "It is a misconception that the Luddites protested against the machinery itself in an attempt to halt the progress of technology.")
Or Phil Ochs "Automation Song" from 1964, https://www.youtube.com/watch?v=WbSTnuXv_3Y
Now you've got new machines for to take my place
And you tell me it's not mine to share.
A strong labor union, or a more socialist form of government interested in supporting labor and workers, could help reduce the hurt of disruption, either by effective trade negotiations or higher taxes on those profiting from automation.
If you look at productivity growth[1], the current rate of automation is actually slower than in most of the prior decades.
[1] https://www.bls.gov/lpc/prodybar.htm
[1] https://www.bls.gov/lpc/prodybar.htm
The less your job requires understanding advanced ML concepts, the more likely you won't comprehend what it's capable of or where the technology is going.
That's why your average dishwasher is blissfully unaware of what's coming for a job function that's relatively simple while computer programmers may be similarly optimistic - but for good reason.
That's why your average dishwasher is blissfully unaware of what's coming for a job function that's relatively simple while computer programmers may be similarly optimistic - but for good reason.
Over a long enough timeframe everything will potentially be automatable. The only timeframe I care about though is the one between me and retirement. For now I am quite safe no automaton can manage to deal with my daily nightmare without pulling its own plug.
So far these all still involve a person operating them. However, what if things get good enough that you don't need massive teams of people as before? There are so many things that can be automated. There is no way the thought of being safe from automation applies to 94% of people. Just shows how naive/deluded we can be as people. If this poll's sample represents the average population's view on the topic we are in for a rude awakening. If that is the case it would seem majority of us are just coasting along in la-la land instead of being aware.