Comment by pdkl95 | Hacker News Reader

Comment by pdkl95 | Hacker News Reader

Parent Full thread

Farming and other types of automation that put people out of work in the past used technology to amplify and/or replace physical labor. Now we are learning how to use the general purpose computer to replace mental labor. These are different types of automation that will have fundamentally different effects on society. While we have many examples of new technology replacing physical labor, automation of mental labor is new. I suggest watching CGP Grey's explanation[1] of why "this time is different."

> went into other endeavors

We are automating away entire classes of jobs, so this will require retraining. How, exactly, do you propose to send half[2] of the workforce to college for retraining in a highly technical field? Assuming this was even possible and assuming there will be jobs for all of these new programmers/etc, is this retraining cheaper than UBI or other plans? It is relatively easy for a farmer to learn another trade that requires primarily physical work. Retraining drivers for unautomatable technical work is a lot harder and more expensive.

[1] https://www.youtube.com/watch?v=7Pq-S557XQU

[2] Driving is the largest category of jobs. The next handful of job types (e.g. retail sales, various clerical work, customer service/support) are also being automated away. Together is very roughly half of all jobs.

TL;DR - just watch [1]

I watched the video, and I have to say it had pretty poor arguments. Too many to name, but I think the core is this:

If any conceivable software was free, required no human labor, and instantaneous: Would the world be a better or a worse place? (although I'd need to find something else to do)

If computers replace doctors, and I get better care for cheaper, aren't I better off?

...If every job was automated, and you can press a button to have robots make software or do medical tasks, that is the ultimate UBI for every person on earth. Everything is free, and then everyone can do whatever they want to fill their days.

P.S. - I want my own clothes-folding coffee-making robot.

P.P.S. - One tiny thing the video said, that is a piece of incorrect common wisdom: Auto insurance companies will not be helped by automated cars, they'll be hurt greatly. If accident rates decrease, premiums decrease, and so the insurance companies margins (and float) decrease, making less revenue and profits.

> If computers replace doctors, and I get better care for cheaper, aren't I better off?

Doctors have a state-guaranteed monopoly that is created via lobbying. That is why, in many countries, doctors are extremely expensive -- even though most of the miracle of healing is done by medicine.

In the future it will probably be a doctor owning and "supervising" the machine. Patients use the machine (paying $XXX). A doctor will take 30 seconds to one minute to look at the results, sign it and your prescription and then charge you exactly the same as before.

The only way the general public will benefit is from more accurate prescriptions. But I sincerely doubt that the cost of a doctor's visit will be reduced significantly. But don't worry, the laws will the there for your own good.

PS: MYCIN was already developed in the 70ies, and showed higher accuracy than doctors. https://en.wikipedia.org/wiki/Mycin

The common wisdom of autonomous transportation being lucrative for auto insurance companies may be correct, albeit for some not-so-common reasons.

As an analog to Wirth's law I'd suggest that in the transportation industry, technological advances in mitigating damage risk are offset by increased risk tolerance in seeking higher throughput / lower latency in the transport of goods and people.

So while the absolute accident rate will go down and the number of meatbag casualties will be reduced by automation removing human error, the _severity_ of catastrophic failure damaging _property_ will increase due to the incentives to push more autos through the pipes in less time: extreme tailgating, hyper-dense cargo, excessive velocity, and bountiful heterogeneity.

Really this is about a transition from (mainly) insuring against loss-of-life to (mainly) insuring against destruction-of-property. Of course, humans will still die even in 'negligible' quantities. But an actuary's loss-of-life liability assessment doesn't need to discriminate between fatal blunt trauma and fatal atomization.

I thought the reaction on hn to a population aging story a few days ago was interesting. No one mentioned AI or robots. ( https://news.ycombinator.com/item?id=12225518 )

Some aging countries will face a severe loss in standard of living if they fail to automate huge parts of the economy. Presumably the worlds population will not continue to grow infinitely and at some point labor shortages will be a big problem.

One of the most destructive outcomes could come from poor government policy during an economic transition. Socialist or capitalist is one thing, but policies which are in direct opposition to another is another ( e.g. Venezuela, Zimbabwe.)

There are also big AI existential risk issues, but I think that is separate from the labor impact.

There will always be scarcity of some sort or another, and there will always be want of something. There is also a predilection to be the person with the most stuff, and robots only accentuate that those with capital are the ones who are winning.

Maybe at some point people will just throw up their hands and start to embrace higher taxes on capital and UBI, but I'm not holding my breath.

>Assuming this was even possible and assuming there will be jobs for all of these new programmers/etc, is this retraining cheaper than UBI or other plans?

That's assuming that programming won't be one of the first categories of jobs to be automated away by AI. Think of the cost savings! But thankfully we don't have to worry about what all those poor programmers are going to do since it's progress for humanity.

Thus the reason I know carpentry :3 but, having programmed for a while, it seems like the hardest problems are converting plain english requirements to source code. At the point that that becomes trivial (and doesn't fall into WYSIWYG editors or graphical languages, which already exist and haven't brought about the apocalypse), wouldn't you just ask the computer to make itself smarter over and over? Or ask it to automate any other profession with stupendous ease?

  I think specialisation in software will be the first to go not programing itself.
  I expect machine learning software to get very good at finding and fixing software exploits in the next few years. Google used deepmind and machine learning to optimise their energy use. How long before (maybe they are doing it already) put it to work on optimising their software code like chrome and android etc to find to speed it up or find exploits.

A few jobs I feel pretty safe will still be here in 50 years. They include robot repair, nurses, and computer programmer. Maybe not in 100 years but I won't care then.

Why repair robots when you can just disassemble them the same way you assembled them(automatically) and reuse the functioning parts and recycle the rest?

If robots can do surgery, surely they can do nursing. And repair themselves. And write programs. And even new and better programming languages that humans can't understand.

So surgery is a mechanical thing. Take a heart like X, make it look like Y. I can see robots being able to get there at some point.

Nursing, at least part of it, is about empathy and care. If someone is in the hospital getting robot heart surgery, someone has to help guide them through the practice. I do not see robot empathy in the next 50 years. Same reason I do not see robot counselors on the horizon ;)

You say that like it's a bad thing.

I can see how the brain vs. brawn distinction could be pernicious, because mechanical robots can do one thing, but a robot that's smart like a human can do anything. Something about that formulation feels like an oversimplification though.

We've had technology replace mental labor before too.[0]

Spreadsheets democratized accounting and while making the work easier, also put a tax prep firm on every corner in the US. It vastly increased the number of people performing some sort of accounting. Some are small business owners that do it themselves, and can now hire a few extra employees.

We've had AI that can write beautiful music for ages, now it can duplicate the style of any composer,[1] but we largely don't care and still have pop stars. And composers.

The work Taylor Swift does for her millions of dollars probably involves layers of cultivating a story about a real person, responding correctly to paparazzi in different scenarios, and judging when to capitalize on interactions on Twitter between other celebrities. This probably sounds like I'm saying she's calculating, but really it has to be an enormous amount of instinct, because it's all probably not reducible to any series of tasks that can be enumerated or even perhaps knowable. Several people have tried to analyze her success, I'm not sure anyone actually understands it.[2][3]

More important for ordinary employees, we still have people compose film scores for some reason. Maybe because even if you can emulate other composers, someone has to decide which composer to emulate, which means having some ability to guess at another artist's as yet unexpressed artistic vision. Once you have the decades of training and experience to do all that, might as well write the thing yourself. Maybe recommendation engines could get there, but I'm not sure those work with novel creations, and I'm not sure we're even interested in perfecting those anymore for areas where they're proven (more or less) to work.[4]

Maybe the weird thing about mental assistants is that they expand the workforce, rather than reducing it. Robots are set up to run automatically, but all the AI we develop is driven by humans to solve day to day problems.

It's like if instead of robotics we built programmable exoskeletons or mechs. The number of employees would explode. There'd still be people building everything, inside the machines. That's what we're doing with computers in a lot of fields. Software is most often like a suit our brain wears to think faster or more reliably.

On the other hand, there are also examples of muscle jobs that haven't yet been perfectly replaced.[5]

The lack of a sewing robot (or ubiquitous clothes folding robots as another commenter mentioned) illustrates that we tend to overgeneralize. Some brawn problems are easy, and some are hard. Same for brain problems. Some problems will stick around, either stubbornly unable to be mechanized, or just not worth the cost.

Also, feedback systems kick in to ensure that labor force participation bounces between reasonable extremes. A lot of the recent downturn in the LFPR can be attributed to retirees.[6] It seems to be leveling off or even heading up again, but regardless the US rate remains well within historical norms.[7]

Sorry this turned out super long and rambly.

[0] http://www.npr.org/sections/money/2015/02/25/389027988/episo...

[1] http://newatlas.com/creative-artificial-intelligence-compute...

[2] https://blog.longreads.com/2014/07/07/taylor-swift-is-a-musi...

[3] http://www.gq.com/story/taylor-swift-gq-cover-story?currentP...

[4] http://www.slashfilm.com/happened-netflixs-recommendation-en...

[5] http://www.npr.org/sections/money/2016/08/03/488611449/episo...

[6] http://www.bloomberg.com/news/articles/2015-01-19/why-labor-...

[7] http://data.bls.gov/generated_files/graphics/latest_numbers_...

I can't seem to export the BLS page I generated. Basically fully expand the range of this page to see the entire history: http://data.bls.gov/timeseries/LNS11300000

Reply on news.ycombinator.com