At Davos, IBM Chief Predicts Artificial Intelligence Won’t Be a Job Killer
wsj.com
wsj.com
So if our social safety nets (retirement) and cultural expectations have kept pace with people being employed less, (be it in hours/day, hours/week, or just what years you'll be working), I'd expect unemployment numbers to not show the effects of technological unemployment.
Of course, it's too simplistic to just assert what I have and ignore forces on the other side (for example, dual income households are now the norm). That is, however, my point: Looking at the unemployment numbers is a poor measure of any effects upon people's ability to work as it has so much else lumped in there.
I'll all for technological advancement. As a society, we've accomplished marvelous things. But it's useful to remember whether individuals are actually improving their standard of living or not.
> now you're commanding power roughly equivalent to 100 horses when you hop into your car
The flipside of this is that (for most US) you must have a car to perform basic necessities such as purchasing food to eat. Food that would have otherwise been in your vegetable garden/chicken coop/pond.
Our society is immensely better for well-off city dwellers, but there are large portions of the population whose level of freedom is only slighly better than servitude.
That doesn't mean technology and AI are bad, just that we have the benefit of hindsight and we don't have to just blindly plunge ahead and hope it all works out eventually.
I'm sympathetic do the idea of saying "we've go to do something" but history unfortunately shows instances where "doing something" is often worse than "doing nothing" so you have to be very very careful.
Despite what Dickens may have written, the period was marked by faster gains for the lower classes than the previous century. And perhaps in part due to what he's written, we now have a social safety net -- I'll admit the quality of this system could be improved. What concerns me is making broad, major social changes in anticipation of a completely speculative (and very specific) view of the future.
I'd be curious about your answer to this: If conditions for the bottom third improve -- instead of falter -- as this trend develops, does that change your policy recommendations (about presumably, UBI)?
[1] http://www.econlib.org/library/Enc/IndustrialRevolutionandth...
2. "the lowest 65 percent" is a pretty big segment. How were the lowest 40% doing, or 20%? I don't know the answer, I just think it's a useful question.
3. Much of this article talks about how rising income is not the same as a rising standard of living, and how it is "plausible" that standards of living decreased for the first 70 or so years of the industrial revolution.
4. Even if wages and standards of living both went up, inequality on its own is a concern. Your quote shows that inequality was worse a century later.
5. I'm not calling for anything specific. My only point is that it's not acceptable to say "we don't need to worry at all, everything will work out". We do need to worry, but it's totally possible that after worrying we'll come to the conclusion that no major changes make sense at this time.
I've just seen a lot of people say "look at the Industrial Revolution, we'll be fine" and I think that's exactly the opposite lesson we should have learned. And I honestly didn't have UBI in mind when I wrote that comment, I was thinking about maybe some kind of subsidized education program for people who need to switch careers because their old career no longer exists. Or anything, I don't know. This isn't my field. I'm just saying this is worth a lot of thought and study, preferably by people who know what they're doing.
>"we don't need to worry at all, everything will work out"
Strawman of course. A nuanced understanding of the historical context of other technological booms is the point of my original post and it's specifically the 'omg AI gonna take our jerbs' kneejerk reaction that I wanted to preempt. I would add that statisticians seem to be less worried about AI threats than your average HNer [1]. And although I don't have a handy source, I know that economists are also likely to brush aside predictions of major impacts.
[1] https://ai100.stanford.edu/2016-report/executive-summary
Where are you getting that information???
"Skilled labor shortage will continue to plague construction companies" [1]
1. http://www.constructiondive.com/news/10-construction-industr...
Think about it this way - previously if someone came up with a more efficient way to do a given job, out of 10 people 9 lost employment, but you still needed that one person to do the job. With automation you get rid of 10 people, and add one additional "worker" to the pool of workers. So now 10 people have to compete not only with themselves but also with one robot that can do job of 10 people.
There are ways in which H1B is imperfect, doesn't work, and isn't fair. There will be ways in which making automated cars for hire illegal will be imperfect, doesn't work, and isn't fair. But those attributes haven't driven the former into the dustbin yet. If anything it's about to get stronger under the current political climate, so there's no reason to think the later won't be successful, at the least as a delay tactic so that automation isn't as disruptive as hypothesized.
Also, as a country establishes a more isolationist attitude, it matters less what the rest of the world does, and the rhetoric right now is distinctly a more isolationist attitude.
Wage stagnation is a real problem. It's a kind of vectorized unemployment where everyone in the middle class on down is partially unemployed through a combination of more part time work rather than full time, and stagnant wages. The broadness of the problem makes it seem like we have good employment numbers, but wage stagnation is a big problem, especially in contrast to the rise in cost of living, in particular housing.
Unemployment is a synthetic measure. A mom working 39 hour shifts at Walmart so she doesn't get benefits isn't the same as a cush office job that's rolling in benefits. When we look at post-industrial societies, we look at late stage job creation being problematic. There's a lot of Mcjobs, temp jobs, and contract jobs. Technically those people are employed, but its not sustainable for them.
Yes, the US has done well but its an exception for the most part due to the massive economy at work here and the relative lack of competition, until recently, from other regions. French youth unemployment hit 26% a few months ago. Spain and Greece are just as bad, if not worse. Jobs in many European economies are lifer jobs because if you leave, your chances of getting a new one is slim.
The real question is, when is the camel's back going to break. Another good career job replaced by a McJob thanks to automation won't change the unemployment stats, but it will make people unhappy. It will drive down the economy with less spending and other issues. In other words, wages matter, not just employment numbers.
I would not assume we are sitting pretty right now. This dam is going to break. Arguably, the surprise win of anti-automation, anti-liberal, anti-globalization, and anti-intellectual candidates in the UK and the USA are the first signs of a major fault.
I like the comparison to horses. Steam engines and trains threatened to replace horses in the mid 1800s. A lot of things that were previously done by horses were replaced with trains and steam engines.
Yet the horse population grew, and cities remained full of horses. There were countless transportation innovations in the 1800s, from canals to streetcars and omnibuses to bicycles. But the horse population kept growing and showed no signs of being threatened by this "automation". Whenever something took a horses job, there will always be other jobs that technology can't do yet, right?
Then the car was invented, and within 2 decades the horse population crashed. Suddenly the price of feeding and maintaining a horse was much higher than the alternative. There is no law of economics that says the supply and demand of a good can't fall below the cost of maintaining it. Or that wages can't fall below minimum, in our case.
Robotics has made incredible advancements over the past 50 years or more. They have taken over entire factories, doing countless routine tasks previously done by humans. But they are still very limited. They have 0 intelligence - they can't see, they can't learn. They can only perform a rote series of movements. So there is still tons of work available for humans.
But with recent advancements in machine learning, this is about to change. A robot will soon be able to be trained to flip burgers, or drive a car, or take a customer's order, etc, etc. I can not imagine any jobs that an average, unskilled human can do, that a machine won't soon be able to do. Maybe skilled professions will be protected - I can't imagine robots being able to program computers for awhile. But the vast majority of humans can not be trained to be computer programmers.
Another way of looking at this is that we weren't ever under threat of millions of displaced horses rising up and overthrowing the system. Since that is a possible outcome of too many unhappy people, changes will be made to ensure people have some base level of happiness on average to prevent that. It's one of the things that makes our current government as stable as it is - the need to respond to the people.
Putting aside my dislike for considering horses as tools, you're missing an important point. Economies only have to function for those at the top in order to be promoted. You could say that the majority of humans are just seen as 'tools' to the people in power that are driven by profit. To them, as long as they get to maintain their quality of life, what do they care about the masses?
If you think this is far fetched, look at what's happening in places like Flint, Michigan. People were drinking poisoned water for decades, and those in charge knew about it. There's even a story that GM had to get a factory in Flint connected to the Detroit water supply because the Flint water was damaging their parts. Plus Flint is not an isolated case. Some people put profit ahead of people, and sadly a high number of those people are in positions of power.
Well, I labeled humans as tools as well, so I'm not treating horses any differently in that respect than I am people, so I think you inferred a slightly different connotation for tool than I meant. Interestingly, the ability to see horses as non-tools is possible largely because their usefulness as a tool was surpassed by other technology. The same with cats and dogs.
> You could say that the majority of humans are just seen as 'tools' to the people in power that are driven by profit.
Yes, that is one aspect of the current system. The main difference is that it's not the only aspect, as it was for significant periods of history for significant portions of the world.
> If you think this is far fetched, look at what's happening in places like Flint, Michigan. People were drinking poisoned water for decades, and those in charge knew about it.
This is only a counter-example if the problem comes out, everyone wants a change, and nothing happens, and then only somewhat, depending on how much people actually care. I'm not arguing that the world has magically eradicated corruption, negligence and shortsightedness, but I do believe that once exposed the public then gets to steer the response if they care enough. Special interests will still have a disproportionate amount of influence, but when the populace cares enough, that disproportionate influence is a drop in the bucket compared to the influence of an incited populace.
For you maybe, I'd say it's largely because they're sentient beings we share this planet with.
>"when the populace cares enough, that disproportionate influence is a drop in the bucket compared to the influence of an incited populace."
It's a nice idea, but compassion fatigue is a real phenomenon.
http://www.compassionfatigue.org/
If injustices were few and far between, I'd have more hope your approach could work, but considering the scale of the issues we face I don't think we can rely on mobilising the population into action over every injustice. Furthermore, the mass media still plays a large role in spreading misinformation, and smaller news outlets often lack the resources to carry out in-depth investigative journalism. I'd suggest any widespread political movement that doesn't tackle these issues head on is fighting an uphill battle from the start.
And my point is that a view such as this is a luxury people can now have because they are no longer required for survival. There have of course been people in the past that did not view these animals as tools, but for the vast majority of our history the vast majority saw them at least partially as a tool. The vast majority of people were seen as tools as well. Again, what I mean when I say tool and what you interpret when you read tool are very likely subtly different, otherwise I don't think you would have much objection. As I stated, I'm not treating them any different than people in this specific respect.
> It's a nice idea, but compassion fatigue is a real phenomenon.
Which doesn't really matter all that much when the whole point is that it may affect massive amounts of people. If it's affecting that many people, it's not compassion that's being tapped, but the changing circumstances of yourself and your immediate loved ones. Compassion isn't called for when it's happening to you; you don't become inured to pain like that when it's your own pain.
> I don't think we can rely on mobilising the population into action over every injustice.
I'm not proposing we do or can, for every injustice. I'm just noting that the people still have a massive amount of power, and when things affect them enough, they create change. The recent election is proof of this. People felt a change was needed, and felt they weren't being represented (whether true or not), and now we're in for something different. I don't think that necessarily means better, but I doubt it will be the status quo.
Your point about millions of displaced people rising up and being unhappy is 100% valid. But you'll be hard pressed to find any company willing to sacrifice a slice of their earnings to employe people just because it's the right thing to do. Especially in the US where the entire culture is based around corporate profits only ever increasing at a rate greater than inflation. The US is in one big race to the bottom to see who can shave $0.01 off the sale price of their product. That naturally leads to two things. Monopolies and hyper efficiency. Because when your profit margin is pennies per unit you have to move one hell of a lot units to make hundreds of millions a year in revenue.
Sure, but that's what legislation is for. We have a minimum wage and laws protecting workers for a reason. Companies do have a disproportionate influence on politics, but only to a point. When it comes to a truly unhappy populace, politicians know what's more important to continuing to perpetuate the system.
Unfortunately, there's a long history of people thinking of and using other people as tools.
I'm unsatisfied with the ubiquitous horse analogy as well. The economy exists to solve human problems, not horse problems. We are constantly generating new problems and new social performances. It would not surprise me if the economy of the future was based around elaborate, manpower-intensive tea-ceremony-like rituals or something equally preposterous. Human obsolescence is not inevitable, but might be engineered.
If, in the far future, there are greater returns from employing robots than humans, then those humans who need to work will starve.
Please explain this basic fact to the people at Davos.
Heidegger's question concerning technology is essential reading, but his basic argument is the technological process subordinates all others to be ends to itself. Notice how people don't so much adapt technology to their needs as adjust their needs to the available technology.
I haven't seen any data, but my guess is that the average person's commute is roughly the same as it was 150 years ago. And if at some point we have self-driving cars or flying cars, we'll still spend roughly the same time commuting as we do today.
I think you'd probably be wrong (though it may depend strongly on which average you take); even with the industrial revolution, working at home, or essentially so (whether yeoman farming or living upstairs from the shop) was a lot more common 150 years ago than it is now, though it may be more common now than it was just before telecommuting.
Whenever I hear statements like this, I cannot help but think that the person making the statement has never done a week of work as "unskilled labour" themselves. (No offense intended.) Be it flipping burgers or building houses, there is improvisation both when it comes to high-level intelligence and in the ranges of motion the human limbs go through that make robots a complete non-starter.
Adding to that, a cheap robot arm that can lift as much as a four-year-old is $20k purchase price, plus another $20k for the controller/AI system, and probably $2k-$3k a year in maintenance. Figure four year depreciation when run 8 hours each day, you're looking at a yearly cost of $12k. That's compared to ~$15k for someone working full time on minimum wage. Take into account the huge process of switching your business over to robots, and it just doesn't look good.
Also, we don't need AI to take a customers order, you just need a touchscreen, or even just an app. The tech has been here for more than a decade. Why don't restaurants do this?
What jobs have been created by this automation? This certainly didn't create 8 (times the # of stores) jobs for software engineers - otherwise, this wouldn't have been profitable for QFC.
Does anyone seriously think that those 8 people went on to do work that paid better, or was more valuable then putting groceries into plastic bags? If so, why haven't they done that before they were put out of work by a machine? Do engineers not see a connection between this, and the fact that they get to step over human excrement on their way to work?
still, there is rarely a line.
You have some might fine AI machines ;)
HackerNews is read by a very weird set of people. All highly motivated and smart. Now, I've worked in a lot of small companies with people who are fairly smart but have no realy desire to win at work (or get promoted) they just want to turn up, do the work and go home. Home is time for family, hobbies, pub, etc.
I've noticed that simple web apps replace people. Excel or an access database, some scripts and someones job is suddenly much easier. People move on, or get demoted to part time, or some small pay cut (or years without a pay rise) and eventually leave.
These are usually the 18 - 25 year olds, employed to help out accounts or a manager or as a secretary while the PA is busy. These are the people that I don't see around so much these days. The gossips, the ones who remember your kids names, who have time for a chat. Who organise a night down the pub where the whole team bonds. They are the glue and the social oil that makes the machine move smoothly.
But they don't seem to get hired these days. Who needs them to print out a thousand letters and put them in envelopes? gmail will do that. Who needs them to find that document filed away in the back office? Or to go to the other office on the other side of town to find that blueprint?
Automation does kill jobs. For good and for bad. Life goes on.
But once competition automates, all bets are off everyone must automate or perish.
Its very much like the prostitution industry or pr0n.
None of those are going to automate any time soon.
https://www.youtube.com/watch?v=34T3IkAtOrc&ab_channel=McDon...
For restaurants where one of the main draws is price I think automation is inevitable as a cost saving measure.
I think we'll adapt as a race, but it'll be mighty uncomfortable for some (say those reliant on low skilled jobs, truck drivers, fast food workers, etc.). It's not to say that the way things will get done will even look the same as the transition occurs. I suspect many things will look different. For instance, rather than speaking your order to a person at McDonald's, you make your selections from a menu and a machine creates it, and then you pick it up and take it back to your table. Redbox already does this for movie rental. Think about how they've changed what we expect of renting movies... Same for home construction, particularly of the cookie cutter variety. Rather than a crew of people going through the motions to excavate a hole, lay a foundation, frame the house, put up drywall, paint, etc., Maybe it'll just be 3D printed with a huge machine (this tech actually exists, if in its infancy). Or maybe we'll switch to brick houses because they make brick laying robots (not humanoid, more like boxes which can move around). The next decade or two will be interesting!
But high end pizza places still don't use them and instead rely on stone / brick / coal pizza ovens instead.
Let's assume your robot arm example is about a frycook at a burger joint. I agree a robot arm isn't an effective way to replace the frycook, but a specialized system like Momentum Machines is developing[1] will be.
The automated fast food joint isn't a regular one with robot arms instead of people, it's a box the size of a small office that takes raw ingredients from a stockpile inside of it and creates fully assembled burgers, fries, chicken strips, and onion rings without any human interaction.
Automation hardware will strike a middle ground between specialization and reuse. Eg the deep fryer will be able to cook several different meals, but it will still only have 1-2 degrees of freedom for putting stuff in and taking it out. Similar story for the grilling surface, veggie chopping system, and the plating / packaging mechanism.
This is much cheaper, faster, and more reliable than truly general purpose robot arms. More importantly though, it takes dramatically less space than a conventional restaurant, so you can plop your company's food-making-box down in Times Square and actually make a profit, while everyone else is stuck four blocks away to break even on the real estate.
The stories will be different in different sectors, but in general I think the coming automation wave is a bunch of quantitative changes coming together to form qualitative ones like this. You could probably build a burger-maker a decade ago too, but the hardware and design know-how necessary have gotten orders of magnitude cheaper and more abundant, and so now it's actually happening.
I think this also has some interesting implications for real estate long term - if every mcdonalds needs 50% of the space they currently occupy over time that real estate will be put back in the market, multiply that by every organization doing that and it has some follow in implications that I can't forsee but find very interesting.
OTOH, the nice dining experiences will be done by people who can improvise and improve day-to-day based on their own tastes. eventually, the stuff people do will be the "good" stuff, and the stuff robots do will be the "bad" stuff. Hopefully. A more negative viewpoint says that robots will do the "high value" stuff and the people will only do the "low value" bad stuff.
In fact, if you consider the partial oxygen pressure (which is what matters for your lungs), this air is no worse than the air you find inside an airplane, or at a 2500m (8200 ft) mountain peak; Wikipedia says 140 million people live permanently at above 2500 m.
Now, this already exists and is called a hypoxic air fire prevention system, and is used e.g. in some data centers and in archives of valuable documents. It's not used in fast food restaurants though, because they have better ways of preventing and extinguishing fires at lower cost.
The majority of space in a McDonalds is the dining area and ordering counter area These areas are occupied by customers. The kitchens in a McDonalds is already pretty compact and dense.
I mean, real-time voiced-over translation on Skype? Eventually even mimicing the other person's voice? Yes please!
All thanks to this shallow form of AI that Nick Bostrom coined as "artificial narrow intelligence" (as opposed to "general" intelligence, i.e. human-like; and "superior" beyond ours). It's becoming possible for businesses of all kinds and all trades, thanks to the drastic cost reduction of deep learning since we moved these tasks to GPUs (and these bad boys keep improving at +25% performance per generation roughly every 18 month). And what AI does, sometimes is order of magnitude better than what any human individual/team could do (think Spotify's recommendations, anything with "big data" and "subjectivity" in it).
Which leads to the same conclusion as your last sentence.
It's orders of magnitude more scalable but you're music nerd friend will beat Spotify's recommendations any day.
Second of all, following their links we find that in 2012 they had a full scale machine built that had a 95% success rate, and they projected 11 months until they could launch the concept in a restaurant. What has happened since? No word, AFAICT. Happy to be corrected if I missed it somewhere and they have actually launched.
But otherwise, it doesn't exactly convince me this is "the next industrial revolution" if they can't polish out that last 5% of a single-purpose burger-making machine in almost half a decade.
A more concrete if less evocative example is Amazon's warehouse automation. Relatively simple robots with 3 DOF (2D planar movement + a container jack) are wiping out enormous numbers of jobs, both in obvious immediate effects (less fulfillment center jobs) and by decimating conventional retail.
edit: Momentum still has 19 employees on LinkedIn, so they're not completely dead it seems.
Even if all of those assumptions hold today, what makes you think the the robot arm of a decade from now won't cost $20,000, but, with minor maintenance, operate for a decade?
And no one complained (child work) now we are able to do exactly that.
I disagree with the general sentiment. What makes humans different is that we can become skilled.
Imagine the big slave population that was used to build the pyramids in ancient Egypt. Suddenly, modern construction machinery appears over there, making them "unemployed" (slaves had a cost too). I would bet only a very very small part of those slaves couldn't be trained to do something more skilled than pushing rocks around.
Unskilled labor, as in repetitive, boring work where little creativity, reasoning, or dexterity is involved, is an infliction humans have to endure, for the moment. It's a waste of our real ability.
http://harvardmagazine.com/2003/07/who-built-the-pyramids-ht...
I think there are plenty of manual jobs that I don't think will be done by robots any time soon. Plumbers, electricians - just don't see it happening.
Digital or easily digitised jobs on the other hand - lawyers, accountants, possibly certain type of programming jobs - will be automated
Like many other "optimistics" about AI I think reality will be much more of a hybrid one, i.e. plumbers or electricians will be aided, not replaced, by AI and robots. It's like the 100 horses in your car, it didn't kill the transportation industry but took it to a whole other level. Imagine as a plumber having an army of tiny drones to fix things you can't even see, let alone reach, as a human; from diagnosis to actual repair. You would wield such power that eventually one man could fix/maintain a whole skyscraper alone and still be home for dinner with the kids.
It's the difference between what an office worker could do with pen and paper and the post office, and what they could do once came the telephone: suddenly you could talk to hundreds of people every day without moving. Then came computers, internet, now AI and robots... A century ago, thousands of men couldn't do some of what you alone accomplish in your office today.
The most menial tasks will be automated for sure, however whether or not a given job is entirely made of such tasks is another matter entirely.
I, for one, don't think it a bad thing that we eliminate brainless or skill-less jobs, I'd much rather pay for UBI for these people until they learn new skills more likely to be rewarding on a self-growth level. I also think it a good thing that we supplement most professions with ever better tools -- think what a doctor or lawyer can do with a computer and a few ad hoc databases that they couldn't do without. Even as an assistant, if you are freed from most menial tasks (e.g. "computer, send a thank you email to these customers with our classic style") then you can focus on creating more value for the person you assist. You can reclaim your worth as a highly-functioning human being, add much more value to the table, while robots and AI takes care of things your kid could do (but never will, and that is a good thing).
It's basically what allowed civilizations to work twice less (in time, work hours) and actually keep growing that GDP. I don't know where AI takes us in this regard, but I wouldn't be surprised if on average we could eventually maintain the same level of growth working orders of magnitude less. But since humans love doing stuff, I'm sure most won't work less, they'll just do so much more --and we might yet reconnect with double-digit growth in western countries.
There may not be robot plumbers, but that's not the only possible outcome.
I've seen tradespeople complaining about others who don't really know what they're doing and just look things up on YouTube in the van outside.
If we had an AI system that could look at images/lidar of plumbing and interact with you to diagnose issues and fast self-driving part delivery, that could result in a system where routine plumbing work gets done by people who would otherwise call a plumber.
I'm not saying this will likely happen soon, but there are other ways this could have an impact besides full-on robotic automation.
From what I've read, we're not going to have completely driverless cars anytime soon (too much difficulty with all the potential edge cases). If we look at productivity growth, it's actually somewhat slow at the moment. People are claiming that soon it will be rapidly increasing, but they've been claiming that for several years now and it's yet to materialize. The evidence of a "robot revolution" is pretty weak at the moment.
And that's not even getting into the question of what effect this would have on the job market. The horse population might have crashed, but that's because you can't retool horse to weld skyscrapers or make advertisements. You can use humans for a variety of tasks, and until we have a post-scarcity society there's always going to be a desire for more production. And even in a post-scarcity society there will probably be a demand for more researchers or artists.
Sure, you might not be able to imagine what people of the future will do, but that's not surprising. If you told someone 150 years ago that less than 2% of the population would be involved with agriculture, they probably would imagine a mass employment catastrophe. They wouldn't imagine jobs like social media manager or video game developer, nor would they probably imagine that there would be enough offices to provide employment to so many administrative assistants. If you go back far enough, people would probably have a hard time believing that so many people could be employed in work that required reading and writing - surely the vast majority of humans can not be trained to read and write!
This actually puzzles me (especially since writing code requires no physical robot, just "an intelligence").
Just recently, we had some AI beat the world's Go champion?
Poker, chess and countless other games are long done.
Programming, in many languages for many tasks, isn't that much of a stretch. In all these games you have a number of primitive operations, sets/rules of valid and invalid state, a goal to work toward.. "same as much of programming, in principle" ;)
I guess at this point the only things holding AI back are (A) the impossibility of parsing human-formulated requirements and specs accurately (we suck at this to, but we're also for now "the only alternative" and can be talked to by same-species inconcise/inconsistent spec-writers painstakingly), and (B) how to deal successfully with all the glitches bugs inconsistencies incompatibilities and special-cases in the various unavoidable-for-interop-required-by-specs human-made APIs, libraries, tools, stale repos, protocols, frameworks..
A more apt comparison would be having an AI create board games that have logical rules, are enjoyable, and are extremely deep.
In this case, specifying a fitness function becomes non-trivial.
Like with designing a board game, how do you know if the program you are writing is getting closer to the goal? Let's say you use the number of unit tests passed as a metric. Ok, but who wrote those unit tests? And more importantly, how do you know you're getting closer to passing some unit test?
In summary, we are still far away from an AI that can fully replace software developers, unless we get a general AI out of nowhere, of course.
In the popular "Humans Need Not Apply" video about this subject. It was suggested that there are signs that "creative" jobs are already being replaced by automation and that professionals are the ones creating tools that replace themselves.
Survivorship Bias; Not everyone is thriving under the current circumstances, not everyone was thriving during the Industrial Revolution or the Digital Revolution.
Also, lots societies needs don't even require Alpha Go level of AI; and computer will make these tasks automatic. And these class of worker won't beat machines logic until a deep learning (sic sic) phase.
It already does. Have you tried to build a house without power tools?
Strong-form, self perpetuating, Terminator style AI has not even been invented yet and you're jumping to the closing chapters.
https://www.youtube.com/watch?v=P73REgj-3UE is slower than a modern house, but not that slow. Further, with just hand tools the Amish where known for raising a barn in one day.
All you need is a bit of advance in robotics, coupled with stronger AI.
Perhaps a proper AGI is not even required for that. Just good enough multipurpose ANI combos.
Now it seems that AI might be able to handle any mind-based job just fine, in a not too distant future. If that is indeed the case, what will human workers do?
Right now there are 121.5 million people working full time in the US. The population is 320 million people and 250 million adults and 200 million adults under 65. So actual employment is around 50% depending on how you slice it.
Your 50% number is more commonly called the 'labor force participation rate' and it's near 63% right now [1], but has trended down for the past few decades. I would argue (as Keynes did) that if work force participation falls and standards of living continues to rise (or stay flat) then society is having good results.
[1] http://www.cnsnews.com/news/article/susan-jones/94333000-not...
Further just looking at the published unemployment rate in the US there has been several changes to the definition over time. On top of that other countries use different metrics. So no you can point to U3 and call it a day.
No of course it isn't. But because it's called "artificial intelligence" instead of "conditional probability" or "regression" or something people think otherwise.
The reality is that this kind of technology will make us more productive, bring down the cost of products and services and make them available to a mass market, in exactly the same way literally every tech advance has done.
Of course some jobs and disciplines will be lost, but they will be replaced with different, higher skilled jobs. Just look at the range of disciplines and jobs that the web has created over the last 20 years. I mean just look at Google. The world's largest "AI" platform has spawned entire industries. No evidence this trend won't continue.
- [1] Elon Musk https://twitter.com/elonmusk/status/702534707464896512?
- [2] Kurzweil https://en.wikipedia.org/wiki/The_Singularity_Is_Near#Expone...
- [3] Gates apparently endorsed the book: http://www.growth-dynamics.com/articles/kurzweil.htm
People who recognise the threat from cheap AI and robot.
And people who think there is some hand-wavy magical force that will ensure that people will always be employed.
You can see that doesn't work now, you can see low income jobs are down, and while they are rising in China and India, those people are already being replaced by robots.
* Delocalization of work. The need for physical proximity is still a pervasive bottleneck. As that further relaxes (VR, remote work), what changes? Think "N weeks to get there" Roman empire vs current "call/email/tweet the world". Maybe a world without commutes? Without two-body problems?
* More richly interwoven human/machine-learning hybrid work processes. Say CEO compensation plummets as you get jit virtual "CEO"s. What happens when most businesses can afford competent leadership? Management? Even if the number of businesses increases by an order of magnitude? Or three?
* What does an economy look like if most people are scrambling, looking for opportunities, not working n-hour "jobs"? Where you can again pick up a "help wanted" "paper", and have a work a few hours later?
Yes, it's easy to imagine dystopias. Especially with resurgent aristocracy, regulatory capture, a patent system I expect to remain broken my entire professional career, and so on. But it's also easy to forget just how much potential change is inbound. Modulo the usual dangers of stagnation and wreckage, "you've not seen anything yet". Maybe.
> Advances in artificial intelligence will lead to job losses, but new forms of employment will take their place...History, though, has demonstrated that technological breakthroughs lead to new employment opportunities
-- IBM CEO Ginni Rometty
But, I think we are approaching (if we haven't already crossed this point) the time where new jobs created by technology advancements will be fewer than old jobs displaced.
My example here is Instagram, a company of a dozen staff or so when acquired, is the digital replacement of photo labs in every city. New technologies are not the job creator that they once were and I haven't read anything to contradict that statement or posit that new tech is going to drive jobs growth.
If you look at it this way, it's possible that the new ecosystem is just as big in terms of number of jobs.
I think a bigger problem may be that the pace of change is accelerating and people used to sticking to a job for a decade or longer may find themselves having to do more drastic career changes every few years.
It all seems so obvious to me that I feel like I must be the fool in the room, but that can't be. Technology and progress in science changes the kinds of work we do. The processes are accelerating, and we're reaching a point where work is better done by machines. I can see no viable future for humans other than Universal Basic Income, with options for additional creativity and work unbounded by time spent doing menial tasks to earn wages for shelter and food and clothing. Yes that all sounds utopian, but what other futures are there? The masses won't have jobs - the machines will. What will people do when there are no jobs? We'll have UBI and we'll have the strongest entrepreneurship and artistic discovery we've ever had. It doesn't matter what side of the politics you're on - UBI is going to be the future simply because it has to be.
"Job creator" is the most intense insult I can think of to call a politician but everyone else seems to love it. Aren't we supposed to be destroying jobs as fast as possible so that automation can make our lives easier and provide higher quality of life to all?
Creating jobs sounds like the worst possible path for people to take - and for CEOs like Rometty to encourage the idea that jobs are here to stay seems dangerous. Jobs are not here to stay, even as the work changes and some new jobs are created along the way, the numbers of jobs will decline to 0.
Another way to look at it is that some people like working, and find having a job fulfilling, regardless of whether it pays the bills or not (many senior people get simple jobs after retiring because they like it).
If we think of a "job" as a means for acquiring the resources for happiness (all too often money for basic needs, but also often a feeling of advancement and status among peers), then it's not something we necessarily want to do away with. It may be that it's less based on money and more on some other capital (social?) instead though.
I've never actually thought of it this way. That's potentially a revolutionary thought that we should be seeking to destroy jobs, not create them.
No one has explained to me how this is supposed to occur. I agree that if you were building a civilization from scratch and had this sort of automation available, you'd probably come up with a nice idealistic solution.
But as it is, starting from the world we live in today, these sorts of efficiency savings just seem to concentrate wealth in the hands of the robot owners.
I feel like companies would say "Oh, look at this, our production went up 10x at no increased cost. Now, should we use that increase to maximize our profits, or should we just raise wages and cut working hours in order to selflessly do our part towards creating a utopia?"
I can think of an alternative. Ahem, cough cough dystopia cough cough.
While not trying to dispute your main point, I hardly think Instagram replaces all of what photo labs do, such as large size prints, professional correction, and specialty products such as cards and books. There are other companies online that do compete in these spaces though. I think most labs are probably getting by with their services targeted towards professionals, who want a bit more accountability than they are likely to get using a large service. That may just mean a semi-large to large lab though, so small labs likely have a hard time of it.
Instagram 'models' come to mind.
On the other side, I agree with the manufacturing comment. There is a British documentary about building steam locomotives on Youtube. Watch what the skilled workers do, they just move things from side to side so a machine can do the hard work. Hell, I can learn to do that in a week.
So when your main/only skill is "driving a truck" what can you pick up as a new job when "driving a truck" has been automated?
Reading and writing where a skill at some point. So was driving. But at this point, they are necessities (maybe not driving, if you live in a civilized country). We shouldn't keep jobs around just because some people really can't do anything else. I'm not saying that we shouldn't find a way for these people to live, but keeping them employed in jobs that hold humanity back, just because, is beyond stupid.
What I was trying to address was: "Yeah, in the past technology advances always created new jobs, this will happen with AI too, do not worry". And I am arguing that this time we are not sure it will work because the skill gap between - for instance - truck driver and self-driving-sensor-suite repairman is much bigger than the one between farmer and factory worker.
But for the current/next transition...
Except history has never had to content with automation and AI. Sure, back in the old days you took farmers and showed them how to make steel. Then took the steel workers and showed them how to work in a cubicle in a service economy. Now we're not that far away from the machines doing cubicle work. What's past the service economy? What's past creative and technical work?
I think its a little naive to think we'll always find work for people to do. There has to be an endgame here and this might be it. Just because certain people were wrong in the past doesn't mean the end of labor as we know it won't ever happen.
Considering we live in a political environment paranoid about "job stealing" and a POTUS who reflects those views, well, of course IBM mouthpieces are parroting those views. IBM doesn't want to become the posterboy for job loss via automation. They want you to think its the other guys.
It's taken me awhile to wrap my head around this. I think I'm starting to understand the fundamental difference.
In the past technology revolutionized processes. E.g. Replace horse with trains and cars. But, the human skills were never touched--they all required a driver.
Whereas today, technology is reaching the point of replacing human skills. E.g. autonomous cars. So, this time as you say it is different because it is the human skills that are being revolutionized by technology.
> My example here is Instagram, a company of a dozen staff or so when acquired, is the digital replacement of photo labs in every city. New technologies are not the job creator that they once were and I haven't read anything to contradict that statement or posit that new tech is going to drive jobs growth.
Photo labs (places where physical photos were developed and edited) were replaced by digital post processing. Some of which is done by the photographers themselves, some is done in specialty shops, some is even done overseas. We can imagine that demand for the output (finished commercial photographs for products, offices etc) has likely continued to increase as the number of products, offices etc. have increased globally.
While many physical labs have shut down, many digital post processing shops have started. And where they haven't started, the photogs themselves are either doing the work or hiring people to do the work for them.
The net effect of digital photography is likely more jobs overall (post processing, web design, etc.) and product delivered for a lower price, but people who used to work for physical labs are completely screwed unless they completely learned a new skill.
Then comes along Instagram! Well, all of the sudden, there is a new place to display high & low quality images and some new editing tools. I am sure that some professional editing is now just a simple filter, but most finished commercial photography projects are still highly post processed by paid professionals. So what about all those images? Well, Instagram has actually resulted in multiple new business models, including "Instagram models" etc.
My guess (no data) is that the number of people producing commercial digital images has increased, not decreased and that instagram probably helps create demand for finished images. At the same time, instragram has likely helped launched new businesses without us even realizing it (access to random food blogs etc.)
While this is all great, and increases global net jobs, it doesn't help the poor guy who owns a photo lab in Missouri. That guy/girl is completely screwed and when people talk about technology jobs, he will have his photo on CNN.com, taken by an unpaid photographer who post processes his images himself. He was/is not newsworthy, but CNN now creates clickbait. In other words, there was never going to be a professional paid to shoot that guy in the past, but now there is a new market for previously worthless content. He will be photographed and CNN will make like $3 on the post in ad revenue.
This can go on and on and on and on.
Technology in the past created new more efficient processes, freeing people to create new processes. That describes your example above and technology advancements largely to date. What's different now is not that we are making more efficient processes, rather technology is displacing human skills. Concretely, when the horse was replaced with a car both still required a driver. But, the driver was so much more productive thanks to the advancement in the process. However, today, it is not the process that is advanced it is the human skill that is replaced--ie. there will be no driver. So, it is the human skills that technology is encroaching on. So, anything that I can physically do, we could soon engineer a robot to do as well. Price/performance may not make sense today, but we know tech gets cheaper/better every year, labour doesn't. So, it is not processes that are being replaced but human skill.
So, with this in mind, I don't think you can argue that we'll have full employment in 100 years time.
Not at all. The mass of work was not done for professionals, but for the mass market. And that one totally disappeared, almost nobody gets photos developed any more compared to hundreds of millions (billions ?) of people before. Of course it was already automated, but now there is isn't anyone left to feed the machine, to check the result, to send the photos or deliver them at the desk, since even the machines are gone...
Where will the 3M jobs that automated trucking will remove come from? I don't think it is all doom and gloom but it begs the question of what fuels the consumption. If employment is super low, then we need money from somewhere to continue to live, so is that universal basic income? Or will things be so automated that everything will effectively be free?
Very interesting time to be alive and wonder what might be yet to come.
One of the bigger differences with AI is that it now may take away many of the lower tier "white collar" jobs along with blue collar jobs that were previously displaced by technology shits.
I'm not terribly familiar with my economics history but is it possible that a huge tide of AI backed apps could be the first time that a significant portion of white collar jobs are lost along with the typical blue collar jobs that are more traditionally lost with tech shifts?
Then again, in 2000, alot of first world programmers were worried about offshoring of programming jobs, I haven't seen any real impact of the loss of these jobs on programmers in North America so perhaps IBM is correct that AI will produce as many new jobs as it replaces.
Machines have already replaced animals including humans for the application of pure physical force to solving problems.
Now machines are now replacing animals including humans for the application of intellect to solving problems.
It is not clear that there is any next frontier. Would it be "love" or "spiritual practice" or what?
Some people say that AI will not progress as far as industrial-age machines did in obsoleting humans. They sometimes give specific examples, claiming that they find AI-generated art to be dull or that AI-guided robotic surgeons only succeed in narrow tasks, etc. It's OK to argue that this will take a long time but I haven't heard anyone explain why the AIs won't win in the long run. Maybe the run is so long we shouldn't worry about it?
Some other people say "we found new jobs before, we'll do it again." For example, the IBM chief. And that seems like empty wishful thinking.
Keep in mind that labour share of income has gone down drastically in the last few decades and that as a result, real wages have stagnated -- which has produced significant unrest in developed countries and unseated the established political order. If that trend continues...
Follow your own logic to its conclusion. If AI does that then what are the new jobs? We are eliminating entire job categories. While the new jobs will be in entirely different categories (e.g. manual labor will be all but eliminated, and the new jobs will be in creative industries). There is very little cross over between those job categories so we are heading to a future with a significant portion of the population jobless.
If we solve for strong AI, we might as well kiss the species goodbye, never mind just jobs -- but the AI we have in between is not going to completely eliminate every category of work in the present, never mind future ones that might arise.
Where do you think CEOs get their information from?
The AI that is so far real is a particular kind of software that has helped large-scale decisions and processes in large enterprises, large websites and etc - places with a lot of data to agregate.
Automation is an ongoing process of replacing people with machines and it is what has eliminated jobs. Advances in software has helped some automation and some of those advances are connected to AI but a lot of the advances involve materials, processes, a willingness-or-not of consumers to use automated processes rather than interacting with a human (whether buying online or from automatic kiosk or whatever).
Indeed, the AI-in-particular and job losses in general narrative is just about a sales-pitch for IBM's Watson etc rather than a serious threat.
Tasks, whether they are manufacturing or service, have a difficulty level in automation. That difficulty is generally scored as a 'cost' with respect to having a person do the task or service and the cost to have them do that.
There are two costs associated with execution, training and marginal cost of execution. So if you spend 10 hours training a factory worker to use a machine, that is your training cost, if it takes them 1 second to operate the machine to do the task, that is the marginal cost. Consider the printing press.
The printing press replaces a fully manual process. A scribe takes 3 to 5 years to write out a copy of a book. They are trained for many years in calligraphy, and each book takes years to produce. The printing press on the other hand you train a 'scribe' to arrange the letters from a type tray into a plate, to ink it, to press it onto sheets. That training takes less time than calligraphy training. Then for each book the scribe puts together the plates, prints out multiple copies of the page, and then assembles the book from the various pages. One or more copies of a single book can happen in a week. Your marginal cost has gone down as well.
Then plate printers, removed the type trays. Then mechanical paper handling fed paper into these machines. Etc to the point where now you need nothing, the author writes, uploads it to their kindle publishing account and the readers down load and read.
But all through this process "employment" around the notion of books changed. And who was employed and how also changed. The product, written and illustrated communication from one person to many, was fundamentally the same but delivered differently.
What we think of as "AI" today allows automating with less people training tasks which defied automation before. And by doing so it changes the market dynamics around that good or service. The most popular example is livery and delivery services with automated drivers. It only became possible to automate a driver when you had a machine that could recognize and respond to a wide variety of conditions in its environment.
But if you look at projects like Deep Mind Lab, Open AI Universe, Good AIs Lab and 5 million usd AGI contest, combine those 3d environments with incremental learning transferred across domains, leveraging deep learning, hybrid and other advancements, as well as (hopefully) the existing body of research from the field of AGI (which has existed for a number of years, despite many pretending it didn't) -- we now have very powerful parallel learning tools and many public projects aimed at general agents. And very convincing demos in more narrow domains. Because of all of this I personally believe the core AGI technology will be demonstrated within the next two years. How long before that makes its way into embodied robots and gets trained up to the point it can be readily adapted to various industries may be another few years.
So my own belief is that we should expect ALL jobs to start to be 'threatened' by AGI systems by around 2022-2023.
I'm sorry, but the stuff coming out in the last couple of years just seems like it's in a totally different league to me...
Look at the Open AI Universe, Deep Mind AI Lab, Good AI Universe, etc. Look them up. They all say that the goal is artificial general intelligence. All of these companies/programs have multiple literal geniuses and tens or even hundreds of millions or more in funding. The AGI environments are public, so who knows how many hundreds, thousands or tens of thousands of talented programmers and scientists are working towards the same goal.
Do _you_ think that all of these large programs, companies, investors would be so publicly pursuing this goal if it were not feasible? If not within two years, then how many years do you think it will be before these projects yield results that have human-like generality or abilities?
If you suppose they will never achieve that, then what level of generality do you assume them to be limited to? Certainly those groups expect to achieve some type of general intelligence. So perhaps you assume in the near term it will not be human-level, then at what level do you suppose it will be limited? Dog-like intelligence? Once agents achieve dog-like intelligence, why would you assume there would not be a way to progress beyond that? We have already achieved super-human intelligence in numerous particular domains.
Fair enough. Your post is proof
> Do _you_ think that all of these large programs, companies, investors would be so publicly pursuing this goal if it were not feasible?
Sure, it happens all the time.
> If not within two years, then how many years do you think it will be before these projects yield results that have human-like generality or abilities?
I'm not sure those projects ever will but in general before we get AGI? It's totally unpredictable. My current best guess is that self-driving car systems won't be as good as promised and because they've pierced the public consciousness, we will get AI backlash setting in another AI winter. This winter will trim the fat of all the crap that gets published now and maybe* a couple new fundamental techniques will be born. Maybe those techniques lead to AGI when we emerge from the winter. We'll get AGI like we'll get cold fusion. When it comes it comes.
> We have already achieved super-human intelligence in numerous particular domains.
Sure but show me a system that can do just two unrelated tasks. Doesn't even have to do them above human intelligence, but something that can say put blocks in holes like a two year old and do some basic language processing. Or play chess and recognize pictures of dogs. What kind of leap is even required to get there?
Yes, when agriculture, industrialization, etc, emerged, old jobs were replaced with new jobs that people could not have previously predicted.
But these new jobs are terrible. They are degrading, dehumanizing, and absurd. We have 63% labor participation, and about half* the people working do nothing. They make Power Point slides, they send emails, they form working groups, they schedule conference calls to schedule other conference calls. Apple hires a bunch of lawyers to sue Samsung. Samsung hires a bunch of lawyers to defend itself. Nothing is accomplished. They all feel like shit, they're addicted to the internet and they are on prescription anti-depressants.
These jobs are rituals, they are not productive in any sense.
When ancient Egyptians developed agriculture, they did not go on to live lives of luxury. Workers who would previously have spent their days hunting and gathering instead spent their days pushing limestone blocks up to the top of the pyramid.
*bullshit on my part, but look at these statistics (https://www.bls.gov/emp/ep_table_201.htm) and come up with an industry-by-industry estimate for yourself, it is depressingly high
That would be extremely interesting to read.
I'm not sure what "full automation" really entails, but those who envision to inevitable automation future don't really explain it either. Everyone focuses on the software without considering if there may be limits to the hardware, and by that I don't just mean the computing machines but everything in meatspace.
The big problem with immigrant-fueled growth in America is that the first derivative of most Latin American demographic trends indicates larger reductions in growth rate than the U.S. or Canada. So North America, in general, could become dependent on higher-risk immigrant populations from Asia to maintain any significant growth.
2. Most boomers will be dead and turned to dust before 2060. And age distribution could become flat as fertility rates go flat. After 2045, the over 65 population rate will probably decrease to under 25% for North America. Some actuaries are thinking that people will not necessarily continue to have increased lifespans.
3. Quality and financial incentives for production and design automation have been driving industry at increasing rate for at least 25 years. My three current employers (one full-time regular, two contract) are all under 200 employees, have all reduced head count by at 15% to 40% over last 5 to 10 years, and have mandated to either not increase bodies or to further decrease bodies.
4. Too much emphasis on self-driving vehicles. Most industrial automation is after the obvious low-hanging fruit - logistics and production, where you do not necessarily see actual robotics, but much process automation and decision-making sustained by ML.
5. Mid-level jobs are being increasingly targeted: parsing of legal documents, accounting, warehouse management and stock control, and some of the more simple product design functions.
6. Anecdotal stuff.
In Early 2016, employer's main site in Southern California terminated OVER half of the engineering group - mostly support and mid-level management. Only one technician, no PCB designers, no mechanical designers, and no engineering managers remain. The remaining engineers do it all - feasible because design automation tools have reached a 'critical' mass.
Spent most of October through December automating the crap out of a factory warehouse and related processes. Last week, 14 people were terminated at the Mexico factory site. The factory warehouse had 14 full-time and 4 part-time employees in warehouse; there are now 3 full and 1 part-timer; and the part-timer will probably go away next quarter. And no robots in sight...
Maybe not now, but once we all get automated, what is the goal we're trying to reach?
Where are we going with this?
Take CV for example. AlexNet is wakeup call for deep learning renaissance, however no one has predicted that we would go beyond human-level performance in just 3 years(ResNet developed in 2015). Same goes for AlphaGo/Machine Translation/Speech Recognition, or the latest photo-realistic stacked GAN.
All in all it is hard to predict what we will end up with by looking at history because we are in such an exciting time of ML/AI development. No matter how prestigious that person might be, his or her words affect very little what the future will actually be, only technology itself could.
Does anyone really believe the latter?
https://www.fastcompany.com/3065339/mind-and-machine/can-ibm...
> A doctor reads about a half dozen medical research papers in a month, Meyerson says, whereas Watson can read a half million in about 15 seconds. From that, machine learning (one of the key types of artificial intelligence today) can suggest diagnoses and the most promising course of treatment.
Much like IBM's bullshit 25,000 jobs PR stunt this one has no substance besides the empty words of a suit who will do anything to increase revenue.