As the need for human labor decreases, costs fall to the floor. This is specifically true for knowledge work that is not constrained by resource inputs. Imagine how much HN would cost to run if it was a newspaper or magazine? This may seem like it's still bad news for humans, because where do the new jobs come from? Yet we regularly see an increase in consumption to compensate for the reduction in production costs. This increase in consumption generally overruns the job reductions due to efficiency. The number of newspaper writers may drop, but the total number of employed writers goes up.
We also see this trend in technology. Early computers took a highly trained team to keep a single computer operating. Now technician to computer ratios can be 1000:1 or more. Yet we seen an increase in total employment rather than a reduction.
> the response to previous waves was a shift to knowledge work.
Computer assisted knowledge work may be precisely the answer to the unskilled and unemployed. Ubiquitous computing made it so that your banker only needs 1/10th the knowledge in order to be effective (when was the last time you calculated an amortization schedule by hand?).
Make it so the "knowledge work" is done by the expert system and the human simply operates the expert system.
Furthermore cost doesn't necessarily fall to the floor just because cost of production does. Many products are being kept up artificially and even more is being subsidized.
No but that's the entire problem. Why computer "assisted" when computers are getting to the point where they can do it on their own completely?
Robotics already outperform humans physically, and AI is (in many areas) approaching humans. So the risk here is complete replacement, not supplementing.
In the near term your solutions are solid, but that's a band aid. Humans will soon simply no longer be able to compete at all except for niches.
>the human simply operates the expert system.
Operate what? At the current pace the only operating required will be flicking the ON button.
>when was the last time you calculated an amortization schedule by hand
5.5 years ago. :p
The steam drill didn't beat John Henry, the system of technology, resources, and people behind the steam drill did. Until a computer decides that it wants to win a chess tournament and teaches itself how to play, humans will always be at the helm.
The problem is that it could be just one human. One person could own all of the machines. They would advise him, protect him, and carry out all of his orders.
Just imagine the President having his entire White House automated. The NSA? Automated. The military? Automated. Police? You get the picture. Robots and computers replacing everybody. One person in charge. 100% loyalty. Small pockets of resistance trying to hack into and take over robots, data centres, etc. Vast computing resources under the ruler's command employing all manner of encryption, fuzz testing, logging, surveillance, etc. in an attempt to eliminate the resistance.
It's like the plot of the Terminator movies without the general AI in charge. Instead, it's a legion of special purpose AIs serving the one guy.
The AI doesn't need to be fully conscious to kick a human's arse in the workspace.
> Operate what? At the current pace the only operating required will be flicking the ON button.
Maybe GP refers to a human operating the expert system like a manager operates an expert. :) And managers make good money on that, so no problem here.
Not if the new wealth is eaten as resource or monopoly rent.
some people arent capable of making the shift to this 'knowledge system' operator based work you mention, for various reasons.
and employeers arent going to be willing to take the time\cost to teach them.
And of course in the face of a combo of increased consumption and automation the jobs created are part time entry level retail.
How? And what would be bad about that?
This sounds exactly the same as automation replacing skilled labour. Lots of things today that we take for granted used to highly skilled markets. Things like clothing, pottery, furniture, etc all used to be highly skilled trades where people would learn the craft through apprenticeship over potentially decades. The cost of these items was massive compared to today, most of them being items a families would save up for years to a afford and keep for a lifetime. While now the production of these goods are available cheaply to everyone, and there's massive industries around them.
Imagine a world where all you needed for an accurate medical diagnosis was a smartphone? Maybe the only doctors that remain are some highly trained specialists, but the number of nurses needed would skyrocket. The barrier of entry for a general practitioner could be much lower. The families who debate wether or not to feed their family or go to a doctor to ask about that cough they've had for 2 months which won't go away will disappear.
Yes technology makes things that used to be expensive and required highly specialized people much cheaper and thus accessible.
But Technology goes further than that and replaces, in the beginning the need for as many and with time for anyone.
It's not about how the market react it's what is outside the reach of AI that it can't do better than humans which can be turned into a big enough industry to compensate for the jobs lost. Thats the challenge here.
Maybe that's possible, but this blossoming we have as AI today is nothing like that. Knowing how we do AI today and the fundamental difference between general purpose AI, which is very stupid, and extremely specialized AI, which is very smart but only useful for very specific things, gives me zero confidence that the recent AI developments are headed towards an AI that's cheaper/smarter at everything than your average joe.
Take a look at all the voice assistants. They're very good at some specific tasks(they aren't general purpose, and they fail at super basic things. These failures can be addressed, but Siri isn't going to learn them on it's own, some one needs to program new features still. As an example, say my wife name is Jane, and there's another Jane in my address book. If I asked a human assistant to call Jane and let her know I had a meeting and I'd be home late, and they said "Do you mean Jane Doe or Jane Smith?", or forced me to say "tell my wife" otherwise she wouldn't do it, I'd fire her. The human assistant understands that "home" is where I live, and the only person who would care when I got home named Jane is my wife. Also when they called my wife and my wife asked when I would get home, they would think to check my calendar for the meeting I had mentioned in the message, if it was there tell her how long the meeting ran, and then even people able to figure out that there'd be some transit time. They might even think to ask wether or not I was going to grab something to eat before I got home since they might know that sometimes when I stay late I get a sandwich at the cafeteria and that my wife doesn't need to wait for me to have dinner.
Identifying that "text" means send an SMS, and that the next could words need to be searched in my phone book, and the remainder is the message, is not general purpose AI. It's some voice recognition with some fuzzy matching and is not that far off from SmarterChild. All the connections a human assistant can intuitively figure out in todays "AI" need to be programmed in. They need to know to look for those variables, and make relations between them. If we stop developing Siri, she will never learn on her own to make those connections or look at those variables.
Finally, the idea that if machines can do everything we do today, there will be nothing left to do is the definition of the lump of labour fallacy. Just like everything else with technology in the past, reducing the cost to produce X increase the availability and accessibility of X, reducing the number of people who had to work on X, and allowing people to instead start working on new things Y, or to do interesting awesome things with X that weren't possible when X was scarce, creating industry Z.
Unless once AI is invented there is literally nothing else new we can create, I think we're okay. And I don't think AI will be out-inventing or out-creating human beings in me or my great grand kids lifetime. There's just no precedent I can see that this is happening.
tl;dr: our AI suck, AI will make things(and subsequently labor) cheaper, AI will enable us to focus on new things, AI will create new industries of jobs.
The economic cost is only a matter of time and the larger transition is already happening. Every time a company looks at how to improve productivity they will look more and more at automation.
You are making the mistake of looking at now and saying see it's not that good and thus missing the point which is that technology is exponential and will continue to be so. Even if there was a limit to the CPU speed the connectivity of things will provide a better and better datasource for machines to learn more and more on top of.
Of course there is no precedent thats the whole point. This is the first time machines are actually competing with what we normally thought required highly trained people. We now know it doesn't and you will be hard pressed to find a job that isn't soon going to be possible to replace with some sort of AI or completely disregard it because it's been digitalized away to some other process.
Which in turn leads to the (cheaper) diagnosis of more illnesses, which leads to an influx of treatments, that will require more Doctors and Nurses...which leads to more jobs.
"Short of moving to basic income..."
There it is.
People undervalue how radically the word processor and spreadsheet changed how offices are run. But it didn't lead to a mass layoff of office workers.
I bet not even a single doctor has lost his job to AI yet.
https://newrepublic.com/article/121712/slow-death-secretary
A similar shift is occurring with paralegals. Combine AI with massive legal data, firms could drastically reduce the hours of legal staff:
https://www.nytimes.com/2017/03/19/technology/lawyers-artifi...
The same could occur for support staff in the medical field, e.g., radiology.
Doctors and lawyers themselves are unlikely to be as affected in the next several decades, except they may be under pressure to take up more clients or including more sophisticated analyses increasing workload. Job requirements will likely require higher and higher levels of education.
Instead, what really happens is that someone makes an incremental advance that allows 9 doctors plus a machine (plus, probably, some tiny increment of people servicing the machines) to do the work of 10. All it takes is to find the most routine aspect of some job and automate that away...
You provide the perfect example in your utterly incorrect summary of the effect of office software. This almost entirely wiped out a huge number of secretarial jobs.
Isn't there a shortage of doctors already? Doesn't seem like the most vulnerable field to incremental automation.
Devastating to an industry in its current form. Taking medicine as an example, I just don't see any reason to expect that, if you automated away 50%+ of the current job, you'd require less, highly trained, medical staff.
Sure they'd be doing quite different jobs, for the most part. Sure some of the current crop won't want those new jobs. But everything I've seen wrt automation encroaching on highly skilled employment would suggest that there's little to no reduction in employment. Quite the opposite.
I tend to look at it a different way. What is the likely limit on the need for medical expertise ad how far away are we from that limit? I would suggest that we're currently a very long way from capacity for medical expertise and you could automate away a significant portion of the current roles without substantially affecting numbers.
I could, of course, be wrong. And, of course, one can extrapolate to a point where AI has perfect knowledge of all human anatomy, natural and enhanced, as well as having innovation and design skills well in advance of humanity. But we're an awful long way away from that.
Near zero I'd imagine. That's beside the point though. The fact that it's possible is what matters. The actual roll-out of it is just a matter of time.
Under Bush and Obama almost all the jobs created were temp jobs, they even had to expand the definition of what constitutes a job.
Can you cite a source for this? The Bureau of Labor Statistics etc. has been using the same U1-U6 unemployment metrics for decades.
I will see if I can find a more official source than this:
http://www.washingtontimes.com/news/2015/feb/10/donald-lambr...
You asserted that "they [the Obama and Bush administrations?] even had to expand the definition of what constitutes a job," which is not supported by that article.
I have had this debate before, and I have yet to find any evidence for the claim.
They changed the U3 focus under Bush I will see if I can find the article about it I read way back.
https://www.wired.com/2017/05/googles-alphago-levels-board-g...
But then, about a year ago, DeepMind redesigned the system. In essence, they built the new AlphaGo without help from human moves. They trained it entirely from games where the machine plays against itself—part of a continuing progression toward AI techniques that truly learn on their own. “AlphaGo has become its own teacher,” says David Silver, the project’s lead researcher.
Answer: The machines just train themselves.
And that's a good thing. Less work for humans is a great thing, but we have to have new economic models to keep up with it.
When introducing a new ML system, it's generally trained off of people, but in most (though not all) contexts you get much, much more data once the system starts getting used in practice and you get feedback on the decisions that the machine made.
The human knowledge is generally needed as a seed to make the model sufficiently not-bad so that we'd allow it to be widely used in practice and learn on its own.
Who knows? Just because you can't imagine the step after the one I describe doesn't make the one I describe less valid. Maybe it'll become self-improving? We already need AIs to help build CPUs so doesn't seem far-fetched.
While knowledge tasks are different than mechanical ones, you could argue that they're both "tasks" that are being "done" and require thinking in some way, but just more (or different) thinking is involved this time.
I don't know if machines could automate things like ingenuity, although at the same time I don't know if the demand for ingenuity is large enough for people to have a stable job.
Many would likely still reach for the far left. A human would have to make a call saying "Wow, this is not a compromise we can make, other humans are not going to use this".
That said, it's a moot point - I am in full agreement that the two revolutions are different.
You might find the first two comments interesting:
@russell usability of the node on the far right is a nightmare. Where do the humans attach the cables? No empathy in AI design.
What makes you think humans will have to attach the cables?
The argument actually speaks against what he was trying to say. He is basically saying that humans aren't clever enough to utilize optimal solutions which is a weakness not a strength.
Your phrasing is too harsh since "optimal" is not defined thoroughly.
It is indeed defined as the blurp also says. I am trying to find the original paper again.
My immediate response is to think that the machine was told to "do something that does X, Y, and Z while W". The computer kicked ass at this in your link, but if XYZ were more fuzzy, or required more lateral thinking, how would a machine fare? I guess a machine could try keep trying, but the number of options to take might explode over time.
Thoughts?
Strong AI is not coming anytime soon. People will still be necessary.
In many cases, the design effort moves from making the structure to writing out all the possible structure constraints the design needs to satisfy.
PS. This is not new at all if you look at the literature.
I.e. computers are already used for ingenious solutions and that's really all that matters here.
To say humans have no place in this is the same as saying in the future, my compiler will be so smart human programmers won't be necessary.
Frankly that design is cool but not mind blowing. If you knew how much angst computer optimization packages can put you through, you might be more cynical. Even before NNs were all the rage there were (ex: genetic techniques, hill-climbing, simulated annealing, Monte-Carlo, pick-your-optimization-technique) to do exactly the things which that Twitter post referenced.
1) AI is not there
2) Human ingenuity is necessary.
With regards to 1) of course it's not there yet. But the progress is astounding and quite a few things will be possible even before we get to strong AI.
With regards to 2) already today there are several areas where technology is better than humans to detect mistakes. In fact we use computers in many cases to test our work. This is an increasing trend and will result in end to end computer design and implemented solutions.
Previous automation waves focused on taking a mechanical task a human was doing and making it faster by mechanical means. Compare that to the modern AIs outperforming highly trained doctors in diagnosis. Fundamentally different - instead of optimising a physical task we're replacing the need for a thinking human.
It's also worth noting that the response to previous waves was a shift to knowledge work. Now that is being threatened too. Short of moving to basic income I don't see where humans are going to go next.