People Who Train Robots to Do Their Own Jobs

https://www.nytimes.com/2017/04/28/technology/meet-the-people-who-train-the-robots-to-do-their-own-jobs.html

by daschaefer • 9 years ago

109 84 9 years ago

sharkweek3• 9 years ago

I imagine this feels slightly worse than training your replacement in a different country, where the labor is cheaper, which is something I have done before. At least in that case, another human is making a living.

jstanley• 9 years ago

Why not just leave without training your replacement? You're going to be leaving anyway.

tryitnow• 9 years ago

I'm guessing that's not a realistic option for people in jobs where automation is taking place. The companies that would otherwise hire them are probably automating or planning to automate.

If they have the resources, then retraining themselves for another job might be worthwhile, but just leaving their current job probably won't help them much.

sharkweek3• 9 years ago

Severance payouts

fuzzybeard• 9 years ago

Your severance package is dependent on your training your replacement. We're talking months worth of pay which is nothing to sneeze at since you have nothing to gain from walking away from that severance.

gnaritas• 9 years ago

Besides severance, there's unemployment which you can't claim if you quit.

wu-ikkyu3• 9 years ago

I don't mean to bash you because it is a valid point considering the current state of affairs, but this comment really highlights how we as a society don't understand that the purpose of automation is humanitarian.

nemo• 9 years ago

In theory the purpose of automation is humanitarian. In practice results vary.

sharkweek3• 9 years ago

If I truly believed that automation, especially in the short term, would include UBI or some other social safety nets, I'd be far more optimistic about this.

I'm nervous that in the short term, fewer people will continue to get very rich, and more people will fall out of the middle class.

laughingman2• 9 years ago

Is it because we are not ready to do away with capitalism as it becomes obsolete?

random28345• 9 years ago

we as a society don't understand that the purpose of automation is humanitarian.

The purpose of automation is humanitarian in the same way that politicians serve the needs of their constituents, and police protect their communities.

While I reap the positive benefits of all of the above, there are plenty of marginalized people who are harmed by exactly the same forces.

In the gap between theory and reality, there's a lot of misery.

ssalazar• 9 years ago

we as a society don't understand that the purpose of automation is humanitarian

If "we as a society" don't understand the supposed purpose of something that we as a society are enacting, then that supposed purpose is academic and ultimately meaningless. There is no universal agency or purpose behind automation other than that of those who are building it, who to date have largely been capitalists and not utopians.

jtraffic• 9 years ago

Interesting that the possibility of training a computer to replace parts of your job has existed for a while. There is an excellent, thought-provoking paper by Robin Dawes that shows that a linear regression trained on human decisions will soon beat the decisions of the human, because it will pick up the trend and follow it with less variance in the future. He refers to this idea as "bootstrapping."[0] (obviously not the same bootstrapping statisticians are used to).

The takeaway was that experts should make decisions about which variables should go in the regression and the signs they should have. The model should produce the forecasts.

Real world application: doctors should figure out which tests to run, models should say whether the patient has the disease conditional on the test results.

[0] http://www.niaoren.info/pdf/Beauty/9.pdf

sitkack• 9 years ago

If the apps you are using for your job, store or communicate data in the cloud, you are unwittingly training your replacement. These massive troves of data that map problems to solutions are being used to train ML models that will replace or augment those tasks. On its own, I don't have a problem with the auto-macroisation of tasks. But those teachers will be invariably thrown to the wolves. I wouldn't be surprised to see startup pitch decks outlining domain specific tooling explicitly designed to capture expert-level worker knowledge.

olewhalehunter2• 9 years ago

btw thinking this threat excludes engineers and programmers is delusional

wyldfire• 9 years ago

I guess. But it strikes me as engineers and coders will just gain more power to work in higher-level abstractions.

vog• 9 years ago

I think this happens all the time. Since good programmers can automate practically everything, they tend to automate as much as makes economically sense for them.

So programming, almost by definition, always involved all tasks that are not automated yet. Programming gets easier and harder that way at the same time: Previous tasks (e.g. taking care for memory management) become easier or automated way, while harder tasks remain (larger, more complex applications, higher abstraction levels).

So I'd rephrase it that way: Engineers and programmers are effected every day by this "threat", but so far worked quite well for to their benefit, they just operate mostly on higher and higher levels, while still having to keep in mind (and having to fix) the lower levels.

Kind of full stack, with a growing stack, thinner and thinner on lower levels, but overall higher and higher.

shaggerty2• 9 years ago

I disagree. Programming is the act of interacting with AI (the compiler) in language (code) detailed enough to specify what we want it to build for us. As tasks become abstracted to a single click or line of code, the programmer is free to work at higher levels of abstraction.

jameslk• 9 years ago

Most (e.g. [1]) don't think of their job as being automated because most don't understand how it will happen. The issue is we only have the present way to think about how the future will work. The future will likely eliminate a lot of mundane engineering jobs (e.g. CRUD apps) with tools that provide automation (learned or programmed) and more intuitive interfaces for the layman. Not all work can be automated, but those are the edge cases. The focus is to eliminate most work, by building smarter and more niche services.

For example, Squarespace, Wix, Shopify, et al have reduced the number of jobs for CMS and ecommerce developers. Zapier and IFTTT have reduced the number of integration jobs. The Grid was an experiment in creating whole websites automatically (most regard it as a failure, but the tech will likely catch up). There's research into using ML to identify and correct bugs in code. These are the low hanging fruit that will erode away at the jobs easiest to automate, leaving behind more niche work.

It's hard to imagine less work for engineers in the future because there's so much work to go around right now. That's likely the case presently because there's so many industries where manual labor can be eliminated, creating lots of new labor for engineers. Whole industries may be disrupted until corporations with automation emerge and take over. We'll likely see engineering work keep rising until there isn't as much left to automate and engineering supply catches up, at which point we'll hit "peak software engineering."

There will be engineers in the future just like there will be medical professionals in the future, but they'll probably be doing a different job and there will be a lot less work to go around. I like to think of that line of work as "escalation jobs," where when the machine fails, it goes to the human. Other work will likely get more and more scientific, theoretical and research-oriented.

1. http://www.cnbc.com/2017/02/24/americans-think-robots-will-t...

rm_-rf_slash• 9 years ago

When I first read "The unreasonable effectiveness of neural networks" I realized that the writing is on the wall for the software development profession as we know it.

I'm glad my employer is sponsoring my part-time M.Eng in AI. Otherwise I fear my job will eventually go the way of the telephone switchboard operator.

xenihn• 9 years ago

Which program are you doing?

bjornlouser• 9 years ago

yes, they will automate away your programming job, but they will still need you to run xgboost ;)

sebringj2• 9 years ago

As AI attains emotion...the boss is like, "Why is this AI telling me to fuck off?", "Why does this AI hate me?", "Why does this AI always have a bad attitude toward work?". It was trained to at a very specific time. Hmmm. This could be termed "Human Cruft" at some point.

blacksmith_tb• 9 years ago

Well, we are all trained (as children) to hate people who mistreat us, perceived unfairness, etc. The question would be if the AI was capable of learning to like a good boss if it was trained to hate a bad one first. This is a common problem for humans and their relationships, for the same reasons...

sebringj2• 9 years ago

Right, but not started on a good footing for sure, kind of like if you are born into abuse and your epigenetics are all peaked on stress expression, harder to integrate harmoniously into positive environments. Speculation vacation... ok I'm back.

joeybaker2• 9 years ago

As an engineer, I consider it my job to automate myself out of a job. It's a sisyphusian goal, but a healthy way to think about your duty to the company and leads to a good kind of laziness.

rz2k• 9 years ago

Sisyphus’ fate could be a punishment because of the effort, or because everything he does is undone each day. Bad jobs when people are young or have poor bargaining power often seem to involve a manager who is disconnected from profitability of the business telling their inferiors to continue doing something in an inefficient way.

I understand wanting to continue taking home a paycheck to support your family, but I don’t understand wanting to continue doing a job if there is a vastly less labor-intensive way to get the same output.

What you describe sounds like Sisyphus always finding out at the end of the day that he only reached a ridge partway to the top of the mountain. A typical menial job, where you do the same work every day without improving the process is more like the classic tale of the boulder rolling back down to the base of the mountain at the end of each day.

joeybaker2• 9 years ago

Fair point. Perhaps "perpetual" is a better word.

Kenji• 9 years ago

As an engineer, you are in the comfortable situation that if your job is automated, pretty much the entire humanity can stop working because your job is among the last to get automated.

Theodores• 9 years ago

For a year I did not automate my colleagues in Customer Service because I liked them and knew they needed income. When they outsourced everyone but the team lead I had a chat with the team lead and we decided that automation was the way to deliver timely customer service and set to work.

The hardest part was the spec. Simple and obvious changes to forms, message strings and processes were arrived at by my colleague teaching me her job. We worked collaboratively and now have happy customers.

Had I started this task earlier with the old team present then I doubt I would have done such a good job. The collaboration would not have been there, it could even have been hostile. Due to the outsourcing we got a break to work on the same side.

For our team lead we have better job satisfaction as everything is excellent with no backlog.

Now we do not need people to fill out forms and do other brainless stuff we find that we need to get more skilled and differently motivated staff for the job.

The old team are not suited to the new work that is there so I guess that is a problem for them but the new hires will be paid a lot better. I think 4 proper jobs is better than a dozen pointless jobs. We can also scale the business now without fear of recall or late shipping nightmares sinking the company.

If we scale the business 3x we will be having the larger team again, all paid better.

I don't see a magic AI system doing the real change, putting people like me out of a job even though I am the first to admit that most of my work is simple stuff.

zw123456• 9 years ago

Some years back I worked for a large corporation on a team that had a large number of "engineers" doing what seemed like pretty rudimentary work so I wrote a program to automate about 90% of the job functions. I thought I would be a hero and showed it to my boss. He asked me if it could be our "little secret" because he was afraid the whole team would be replaced including him.

leggomylibro• 9 years ago

See, the company should really be incentivizing this sort of thing. If you automate your team out of a job, great - you all get a pension at 33% wages as long as the company is in business and 6 months to find a new job.

Company saves a lot of money, workers are incentivized to this sort of behavior, and you could probably handle the upkeep by offering to keep one or two employees on at full pay if they'd prefer, or stipulating a day or two a month to perform any necessary maintenance on your automation.

But that's probably a pipe dream; it would essentially mean funneling capital to those who produce value, which is apparently a ridiculous concept in this modern economy.

ohazi• 9 years ago

you all get a pension at 33% wages as long as the company is in business and 6 months to find a new job

Boss: "Or, and take a minute to appreciate my cleverness before you respond, how about zero, and don't let the door hit you on the way out!"

These improvements eventually materialize regardless of any incentives for this behavior. Perhaps it would come marginally sooner with an incentive.

But since non-owner workers will no longer have a relationship where they provide continued value to their employer, the company is (correctly?) motivated to screw them over (think single vs. repeated prisoner's dilemma). The company could even pretend to offer the incentive to get these improvements sooner, and then fire them anyway.

The game theoretic optimum for capital owners is to screw over their (soon-to-be former) employees. This is why some suggest a tax on automation -- there's a market externality here that isn't going to magically go away.

whack• 9 years ago

In most successful companies, the guy who came up with the idea and his manager will be heavily rewarded, and taken care of in a meaningful way. Everyone else is given some time to either find a different team to work for, or asked to leave and find another job where they are actually contributing something of value.

Although this might seem tragic in the immediate aftermath, it's precisely activities like this which lead to GDP growth and economic prosperity.

The biggest gains of course still go to the business owners and executives. Not the working class. Hence why we need higher taxes on the wealthy and a stronger social safety net.

iamacynic• 9 years ago

at a major legal industry firm, i had an ML project torpedoed even though it showed reasonable promise. i was fairly young at the time but now thinking back, this may have been part of it.

i was an hourly consultant and they did pay me handsomely to do that little experiment though. can't complain. the world is a complicated place.

maxxxxx4• 9 years ago

That's the stupidity of companies. Instead of people being afraid to automate their work they should be encouraged and rewarded.

clarkmoody3• 9 years ago

Companies don't make decisions. People do.

Markoff2• 9 years ago

was there significant incentive for you? sure, go for it, let those people lose jobs

though it's sad some people do this even without any incentive not thinking about consequences of their actions, they just wanna brag how they improved things, they will not benefit in any way from this improvement, while it will make lives of fired people miserable and only one profiting from it will be company with higher profits for those few owners on top, do you really wanna support such thigs?

but yeah, I was once young and idealistic too...

LrnByTeach2• 9 years ago

based on your description of Manger and team, I assume the Team size is 5 to 6 people at least.

If that company run so inefficiently that it could not identify 90% of work of a 6 people team can be replaced by bunch of programs, so asume that company or the division may be closed by now ( or severely down sized as competitors took their business )

It would be great if you can find out what happened to that team

Chathamization2• 9 years ago

What's impressive is how much office work, from my experience, could easily be automated away. Not even with anything fancy - very simple scripts would be enough for a lot of stuff.

It's not uncommon to find people in offices doing things like manually generating the stock reports and graphs each year, collating dozens of Excel spreadsheets by hand, or even doing data cleaning by hand. For all the excitement in the news about machine learning and robots, a huge chunk of work could be automated away with relatively simple tools.

stinkytaco3• 9 years ago

I hear this often, but have yet to see it in practice. I have worked in government and a number of underfunded non-profits who would dearly love to have some of these things automated. But in my experience it just shifts work. It automates something, requiring extra work somewhere else. Or it automates, but then breaks one, mission critical function that needs to be taken care of. I truly don't think this is just about "save my job", because they feel the pinch of lack of personal and funding, but about the difficulty in truly enumerating what the workflow actually is. Once you've paid a consultant to take care of it, you realize you forgot 3 steps in the process or that they were not listening when you described them.

dahart• 9 years ago

Can human agents find new ways to be valuable as quickly as the A.I. improves at handling parts of their job?

The classifiers we call AI right now can take over a lot of tasks if not whole jobs -- which points at how remedial many tasks & jobs are, for better or worse.

What ML can't do yet is tell you that you asked the wrong question. Some jobs will go away, but not all of them, for now. And some jobs have gone away with every technical development so far.

The interesting and open question is whether it's turtles all the way down - is it possible that a large enough classifier that's seen enough examples of everything will be sentient? Is it possible that we are just walking fleshy classifiers, and there's no actual line between consciousness and a large enough neural network?

jorblumesea• 9 years ago

As a junior developer, is it even worth pursuing an "standard" career in technology if line engineers will be replaced by some kind of deep learning algo in 10-15 years? What career is even safe if we get to that level? When I see events and the pace of progress, it's hard to even justify any long term planning and I feel that (American) society is completely unprepared for the impact automation will bring.

Might as well go for my masters now, given that no one with a BS degree will be needed long term? Given a long enough timescale, even that won't matter I guess...

dorgo4• 9 years ago

As long as I get my food from humans in McDonalds I'm not too concerned about me being replaced by a robot. Society is lagging behind state of the art AI and robotics by decades.

drenvuk2• 9 years ago

That is literally two years or less away from today. Please look up momentum machines. They're beginning to hire people from the food industry and have stopped hiring people from the tech field vs a couple of years ago. It's getting close.

grandDesigns• 9 years ago

When they invented the COBOL compilers, the assembly language programmers were scared. "Now every business man can write their own code!".

Same when they invented the spreadsheet.

Same when they invented garbage collection.

Same with every new library, framework, "its just like legos"-invention.

Yes, they make it easier. But, the world then demands more software.

Sofware begets software.

Without compilers, there would be many many fewer programmers in the world.

Someday we might have the computer from the Enterprise. Then we can ask it "Give us the answer to the ultimate question about Life, the Universe, and Everything".

dorgo4• 9 years ago

toexitthedonut• 9 years ago

That's the way I see it. Programmers have been automating themselves out of menial tasks since the first compiler, or possibly even before.

What is considered "menial" depends on the context of the problem needing to be solved and a mix of your personal experience.

The increase of hardware performance also helped allow software abstractions to the ridiculous levels you see. That's what also makes it possible for computers to solve a greater array of business problems.

Engineers making tools that are used for making other tools, that are tools for other tools for other tools that eventually helped make someone's wedding photography site on Wix.

wikwocket• 9 years ago

I wouldn't lose sleep over it. Deep Learning is great at making recommendations or classifications, given petabytes of data, but not even close to automatically solving general-purpose computing problems.

For example, I can type an error message into Google, and get sent to the solution, since thousands of others before me have provided the training data. But can I just describe my task to Google and have it produce the solution? Not even close.

Markoff2• 9 years ago

this is pretty much what microsoft has done with reviewers of their app store, they fired ("downsized" in newspeak) pretty much 80-90% of staff after automating their jobs with help of few young and eager people not thinking about consequences. at same time they lowered requirements for submitting apps due to falling amounts of developers willing to code for their platform. did it help microsoft? i don't think so, good riddance...

skydoctor• 9 years ago

Nitpicking but the article has 5 examples, only 2 of which are true to the title. The travel agent and the customer representative actually train the "robots" to do their own job. However the software engineer, for example, doesn't train A.I. to write software - she writes code which helps A.I. to get trained in driving better. At present, I think very little of a software engineer's task can be replaced by an A.I - understanding requirements, debugging, deployment, and the likes. Mind you programmatic automation has to be distinguished from A.I.

dorgo4• 9 years ago

for example, doesn't train A.I. to write software

Actually there is reasearch in that direction. Genetic programming comes to mind (https://en.wikipedia.org/wiki/Genetic_programming) It would be something which takes a specification and retuns a computer program. Somebody still needs to write the specification though.

hindsightbias• 9 years ago

What if we train them with bad traits?

csours• 9 years ago

The Toyota Way includes the idea that continuous improvement should increase quality and efficiency, and that the efficiency should be re-invested.

If the smartest thing your company can imagine doing with a person who is able to automate their job is fire them, your company is wasting valuable resources.

stinkytaco3• 9 years ago

This assumes that a company's ultimate goal is growth. For a large, publicly traded company like the article is talking about, this is probably true. But many companies I do business with on a day to day basis (local stores, contractors, etc.) have not interest in growth. Automation is a way of reducing workforce and costs.

amelius• 9 years ago

Waymo’s cars have driven two million miles in the real world and billions more in computer simulations. But it’s impossible to program for every event.

I wonder what will happen if two people cross the street with a big painting of ... a road :)

noonespecial• 9 years ago

I have it on excellent authority that if the paint happens of be of the ACME brand, its very hard to tell what might happen.

Paul-ish• 9 years ago

I know this is a joke, but wouldn't Lidar detect the obstruction?

id122015• 9 years ago

but dont you hate it when you call customer service and want to talk with a human but the robot asks you what you want ?

now I remember, I think it was PayPal that I called.

Last time I had to repeat twice that I want to talk to customer service. However nobody told me that repeating two times the same thing would do the trick. A novice would be lost at that.

downrightmike• 9 years ago

Its not the stupidity of the companies, its that you have a hammer and not everything is a nail.

People Who Train Robots to Do Their Own Jobs

Related Stories