Support NeoGAF

[Trumplestiltskin]

I see a major flaw in their argument: the incessant need for higher and higher levels of education in order to just achieve a middle class lifestyle. This will inevitably lead to disaster, as the vast majority of the population will either not have the motivation or capacity to obtain that level of technical knowledge and know-how.

As the labor pool increases in size due to globalization and includes highly educated, highly skilled workers from poorer countries where education is highly subsidized or free, native American workers, on average, will fall further and further behind, being underemployed or exiting the workforce altogether. In this paradigm of a race to the bottom in terms of wages and standard of living across the population as a whole, Bernie Sanders is the only person stating what needs to happen: Americans need access to highly subsidized or free higher education in order to compete and not start life with crippling debt.

Each generation knowing more than the past one has been the case for centuries. It's not a different expectation that people twenty years from now should be better educated and trained than ones today.

 

[tokkun]

I was trying to imply that as long as the problem is parallelizable then scaling applies today, the gains in that graph are products of improved transistor density and efficiency.

If we're talking about AIs / robots replacing your average joe worker, then the key to that is going to be things like cost, size, and environmental impact. Even if you have a perfectly parallelizable problem - which incidentally is not the case with neural nets, as they require a large amount of inter-chip communication - it does not help unless you can also reduce the energy per operation at an exponential rate. It is one thing to have a warehouse-sized supercomputer beat the best human at task X, but a company is not going to pay the cost of a warehouse-sized supercomputer to do job X, if it is cheaper to hire a human.

We are getting big boosts in AI processing power right now (in cost and energy normalized terms) because companies are switching to more tailored processors, be it the latest generations of GPUs or FPGAs & ASICs. I'm not optimistic about that rate of improvement being sustainable beyond a few years, though. People take the "Law" part of Moore's Law a little too literally I think. You can't expect that exponential improvement can continue indefinitely.

 

[night814]

Robot will try to weasel out of it. Something along these lines.

This is fucking great

 

[TakeItSlowDude]

If we're talking about AIs / robots replacing your average joe worker, then the key to that is going to be things like cost, size, and environmental impact. Even if you have a perfectly parallelizable problem - which incidentally is not the case with neural nets, as they require a large amount of inter-chip communication - it does not help unless you can also reduce the energy per operation at an exponential rate. It is one thing to have a warehouse-sized supercomputer beat the best human at task X, but a company is not going to pay the cost of a warehouse-sized supercomputer to do job X, if it is cheaper to hire a human.

We are getting big boosts in AI processing power right now (in cost and energy normalized terms) because companies are switching to more tailored processors, be it the latest generations of GPUs or FPGAs & ASICs. I'm not optimistic about that rate of improvement being sustainable beyond a few years, though. People take the "Law" part of Moore's Law a little too literally I think. You can't expect that exponential improvement can continue indefinitely.

I don't think the strategy for replacing human workers will be to provide a high level of intelligence to each robot. I suspect that each robot will be given enough power solve most expected problems and edge cases could be dispatched to a supercomputer, that supercomputer will also be responsible for optimizing those robot workers over time. I can't help but to think there are a lot jobs that require human level intelligence but only for very small slices of time. This level of intelligence is the threshold to automate, though once that threshold is surpassed by a machine it can apply that level of intelligence constantly or to 1,000s of robot worker in time slices.

Yeah, from what I read traditional scaling and new transistor topologies will enable scaling until 2025. If it peters out by then I would still expect to see a lot of jobs automated or situations where machines augment jobs previously done by many humans and leaving only hardest tasks to a few humans. There is also a rapid lowering of the price floor in remote sensing/logging which is having a pretty big impact in the market and provides a place to apply significant amounts of machine intelligence to optimize processes, if the gains from these optimizations outstrip demand then I would expect more jobs to be killed.

 

[tokkun]

I don't think the strategy for replacing human workers will be to provide a high level of intelligence to each robot. I suspect that each robot will be given enough power solve most expected problems and edge cases could be dispatched to a supercomputer, that supercomputer will also be responsible for optimizing those robot workers over time. I can't help but to think there are a lot jobs that require human level intelligence but only for very small slices of time. This level of intelligence is the threshold to automate, though once that threshold is surpassed by a machine it can apply that level of intelligence constantly or to 1,000s of robot worker in time slices.

I think it is easy to underestimate the amount of processing power it takes to do things that humans consider trivial. They only seem trivial because we are very good at them. Even doing one of these "McJobs" you are constantly engaged in tons of visual, audio, and motor processing.

 

[plagiarize]

Self driving trucks are still going to need a human onboard, even if they're less a driver and more a security guard. No way I'm sending tens of thousands of dollars worth of parts or product cross country without a person to deter theft and protect those goods.

Look at the insane supreme court case where they told a driver to stay with a broken down truck in dangerously cold weather. That guy should never have been fired, but it should show you the thought process these companies have. The product will need to be protected.

Cab drivers are fucked. Truck drivers not so much.

 

[Muppet of a Man]

Each generation knowing more than the past one has been the case for centuries. It's not a different expectation that people twenty years from now should be better educated and trained than ones today.

Knowledge growth is exponentially accelerating at this point. Fewer and fewer people are able to keep up. I don't think what was true for the past will be indicative of the future here, as we will hit hard limits of intellectual capacity or at the very least have to narrow the focus of our knowledge to compensate.

 

[Kill3r7]

These aren't estimates on technological unemployment, it's estimates on when AI will be able to perform certain tasks. The point isn't that automation won't automate jobs, but that it won't get rid of work and people will shift to being employed in different tasks where they have comparative advantage over AI.

My biggest issue with studies like this is that they assume that folks will be able to transition to new jobs, doing new tasks, in the automated world. That might be true at a macro level but I have a hard time believing that low skill high paying jobs are going to pop up from the ether. Many of these individuals lack the necessary skills to qualify for these jobs. I highly doubt there will continue to be $100k+ jobs for secretaries, low level analysts, paralegals and many, many other such jobs. I mean look at certain industries post 2008 and you can see how many jobs never returned.

 

[mike6467]

Each generation knowing more than the past one has been the case for centuries. It's not a different expectation that people twenty years from now should be better educated and trained than ones today.

The number of people in that category would shrink though right? Tuition costs, and expected return on those costs are insane as it is. Eventually people will stop going entirely if wages aren't increasing enough to justify those costs. Costs are not showing any sign of showing down (teaching seems to be the best example of this). I'm lucky enough to have a company paying for my EE, but I see this happening in a lot of STEM fields right now, which are supposedly the "do this to have a steady well paying job."

How is that handled? Less people means colleges are forced to shrink until demand goes up? I'm legitimately asking, I'm not informed enough to make an argument either way.

 

[platakul]

Self driving trucks are still going to need a human onboard, even if they're less a driver and more a security guard. No way I'm sending tens of thousands of dollars worth of parts or product cross country without a person to deter theft and protect those goods.

Look at the insane supreme court case where they told a driver to stay with a broken down truck in dangerously cold weather. That guy should never have been fired, but it should show you the thought process these companies have. The product will need to be protected.

Cab drivers are fucked. Truck drivers not so much.

A self driving truck doesn't have to stop except to refuel or weather conditions. Neither is prime for theft, and the cost of putting in security guards and that insurance is going to cost more than the product on one truck that gets robbed if that ever did happen

 

[Laiza]

Self driving trucks are still going to need a human onboard, even if they're less a driver and more a security guard. No way I'm sending tens of thousands of dollars worth of parts or product cross country without a person to deter theft and protect those goods.

Look at the insane supreme court case where they told a driver to stay with a broken down truck in dangerously cold weather. That guy should never have been fired, but it should show you the thought process these companies have. The product will need to be protected.

Cab drivers are fucked. Truck drivers not so much.

This argument has never made any sense to me.

What's stopping people from stealing those trucks now? Is one armed individual really that much of a deterrent? Do you really believe that bandits will give two shits if they're really determined to steal their cargo?

And what kind of bizarre assumption is that, anyway? Are we really such a lawless and disorderly country that highway banditry is something we're actually realistically concerned about? Why haven't I heard anything of this, if this is the case? I just don't understand.

 

[Amibguous Cad]

If he wants to argue that automation isn't cutting into employment figures, he'll have to address why labor force participation cratered during the great recession and hasn't improved much during the recovery.

 

[tokkun]

A self driving truck doesn't have to stop except to refuel or weather conditions. Neither is prime for theft, and the cost of putting in security guards and that insurance is going to cost more than the product on one truck that gets robbed if that ever did happen

That isn't true. Trucks are forced to stop periodically for various regulatory reasons. That's why those "weighing stations" you see signs for exist.

 

[KingV]

If he wants to argue that automation isn't cutting into employment figures, he'll have to address why labor force participation cratered during the great recession and hasn't improved much during the recovery.

Not really. The productivity numbers tell the tale. GDP/hour worked is only growing marginally. Less, in fact, then the historical average.

So if all of those people that are no longer in the labor forced have been replaced by robots, how is that number not growing? If you lay off 100 workers and go from 101 workers to 1 robo-supervisor at a factory, replace them with 100 robots, but the output is 100 widgets per day in both cases, the productivity for that factory exploded by 100x. If this phenomena is widespread it would show up in the manufacturing productivity numbers. But it doesn't? Why not?

Output is up, so it's not that output is falling faster than people are laid off.

Manufacturing pay is going up faster than the national average... why is that happening if people are being replaced by robots? Shouldn't the opposite be true as people clamber for the last few available jobs?

The only possible way this can work is if it's hitting certain industries only, but in a way that is almost exactly offset by productivity losses in other industries. If you look at the numbers some are up and some or down, but without having any context I don't know why coal mining productivity is up while dog food manufacturing productivity is down.

 

[SummitAve]

That isn't true. Trucks are forced to stop periodically for various regulatory reasons. That's why those "weighing stations" you see signs for exist.

Yeah I've been watching this trucking show on netflix for some reason, and they are stopping all the time for all sorts of reasons. Checking the load, tightening straps, checking and changing tires, mechanical issues, ect...