For the kind of work you go into with a PhD, the work done during the PhD itself counts as work experience. During my PhD I was doing the same kind of work I would have been doing in industry, but at a University.
I use ROS for two things only: the interprocess communication system (with the nice data logging capabilities) and rviz, the data visualization tool. They can keep the rest. Also, it irks me that they call ROS an operating system, when it runs on top of Linux.
True to a point. One thin I've found with ROS is it encourages a lot of "black box" development. "connect these components and you'll have a working robot!" And you do.... until it doesn't work. And then you don't know why. One of the big promises of ROS is code reuse, but in my experience, the stuff that works reliably isn't particular sophisticated to begin with, and the rest you really should be writing yourself to ensure a robust robot you can actually debug. We've tried the "connect black boxes" approach, and by the time you've got a system that works the way you want, you've rewritten most of the code anyway.
I typically don't use adblock. In general, I don't mind most ads. If a site has really bad ads that are intrusive and annoying, I just avoid it. There's enough content on the web. But it was Slashdot's idea to say "Hey, if you stick around and make comments, and the community thinks you're adding to the discussion, then we won't show you any ads" and that's been the agreement. So as the ads Slashdot displays get larger and more numerous and more intrusive, when they don't hold up to that agreement it's very frustrating, because I frequent this site a lot and would rather not move on because of the ads.
The kind of processing done for the Kinect is well suited a GPU: facial recognition, skeleton tracking, hand tracking, voice recognition. These feature use a lot of matrix math, geometric algorithms, and signal processing... stuff you want to do on a highly parallel processing architecture. So yeah, doing them on a serial architecture like an x86 CPU would be much more taxing.
I've been seeing ads for a while now with Disable Advertising checked. At first it was just the ad under the story and the annoying pop up ad at the bottom of the screen. Unchecking and checking again would make them go away. Now all the ads are enabled and re-checking the box does nothing.
Seriously, this shit is incredibly annoying. The ads are obnoxious, gigantic, and intrusive. I've got a little checkbox that says "Thanks again for helping make Slashdot great!" under it, and it does nothing. So where's the thanks, exactly? Definitely not feeling it.
you're still employing only one man for every hundred people worth of productivity
You're still making two mistakes. One assuming that every robot is used to kill already existing jobs, and two assuming that for every job lost two more cannot be created. I've already argued to second point in several other posts, but I'd like to try and convince you on the first point where I believe you are mistaken. In this post I argue several points where robots can create a net gain in jobs, some of which cause the loss of no jobs at all. The problem you and others are stuck on is arguing about current robots while supposing a future robot. To this point, all robots have ever done is displace factory workers through automation. This is because they are uniquely suited for this type and work and really nothing else.
But robots of the future will be suited for much more, to the extent that they 1) augment human abilities and 2) extend to endeavors otherwise inaccessible to humans. In these scenarios, you aren't replacing anybody, or maybe a few people, while enabling work for dozens, hundreds, or even thousands by creating jobs were none previously were able to exist. If we're talking about robots of the future, you have to give them more capabilities than just those that replace menial labor.
And it's worse than that, because unlike humans robots lend themselves to modularity - there's no sense in maintaining a vision-systems repair specialist if you can simply unplug the faulty system and replace it with a new one more cheaply.
Again, you are assuming a robot of today for a robot of the future. Robots *today* lend themselves to modularity. Robots of the future? Already we see emergent behavior we can't fully comprehend when a whole system is working together as a whole. For instance (I don't like to go down the sci-fi very distant future path, but since everyone here likes to argue about machines with extreme intelligence replacing humans I'll do so for the moment), consider a robot brain with billions of interacting neurons. Perhaps it's not possible to download and replicate such a thing. Consider robot with some proportion of biological parts. Perhaps such a thing is not easy to repair. I don't know that's just a little sci-fi there.
Back to reality and with respect to vision in particular, we employ many computer vision experts. Robot vision gone wrong isn't just about the camera. When robots are seeing the wrong thing, sure you can replace their vision systems, swap in a new one, but then when it's still seeing the wrong thing, what do you do? It takes a specialist to figure it out. This is the difference between a robot repair man and a TV repair man; one is dealing with a single-purpose solid state device, while the other is dealing with a highly nonlinear system which is interacting with its environment and making decisions on its own. One of the two is exponentially more complex to service.
You only need a human involved for non-obvious problems, which probably means someone who is reasonably well versed in *all* the component systems of at least a specific class of robot
And you expect such a person to exist? Already in robotics we have dozens of specializations.
Not something Joe Sixpack can reasonably be expected to learn to do.
That's why Joe Sixpack doesn't become a surgeon when he's displace from his job, he trains to be a nurse or an orderly, or a sales rep.
And as for the technician support staff - how many do you really need?
How many does a hospital need? There are many more nurses than there are doctors and surgeons.
just look at what's happened with customer service in the last decades, it used to be you talked to actual humans on the phone, now you've got to go through several layers of poorly-automated systems design
Self repairing machines maybe science fiction now, but so were cell phones with internet browsers in 1995
Smartphones were an exercise in making things we already had in 1995 smaller and faster. i.e. largely an engineering task. When you talk about robots that can repair themselves i.e. something which requires creativity and intelligent thought, you're talking about pushing the boundaries of several scientific disciplines far beyond where they are today. That is to say, in 1995 you could definitively answer that yes, it's theoretically possible to create an internet enabled hand held device. Today, there is no theoretical basis for even believing machine learning can *ever* match the reasoning or creative capabilities of a human brain.
If that ever gets significant progress it wouldn't be too far fetched for machines to self diagnose and self repair.
Good luck to them. Hopefully they make some headway by the time I'm 120 and can upload my consciousness to the internet.
I heard a VR software developer say "People overestimate technological change for a year, but under estimate change when you talk about a decade."
And yet self driving cars and true artificial intelligence are always "ten years away".
So its worth to give a bit of thinking on what happens when machine learning is good enough to eliminate current jobs and all possible jobs after that.
The same amount of thinking I'm willing to give what I should do after the sun explodes or the heat death of the universe. Thinking about such things is a flight of fancy, fueled by a science fiction and a popular media depiction of machine learning and AI.
And they'll be designed to make it easy to repair them. Hot swappable modules for each major component. Easily automatized repair. Most broken modules won't be repaired. The goal will be minimal downtime (we had contracts for under 4 hours unscheduled downtime per year). So that means the entire unit, or an entire module is swapped out and the unit is functional again.
You're arguing about the future while thinking about the complexity and capability of robots today, and again (as I argued in another post to you) considering only robots that replace factory labor... which robots of the future will do of course but go well beyond. You think it's as easy as just swapping out components? Maybe my future robot butler keeps knocking over my lamp with its arm. So I swap out the arm and it keeps doing it. I swap out the eye and it keeps doing it. I swap out the brain and it keeps doing it.... now what? This is a common problem that I face every day... debugging and repairing robots is not as easy as "hot swapping" modules because we're talking about machines which interact with their (nonlinear dynamic) world and make choices on their own based on input. Robots of the future that everyone is so worried about will not be your grandfather's robots.
Specialists cost money and are likely to only be used in the initial design, creation, and debugging of the robots
You think the debugging stops when it's shipped? With robots it's an ongoing process.
Our mainframes today already self analyze and even send emails saying they need a specific part replace. Heck- our automobiles tell repairmen what part is broken.
And yet my car has been to the mechanic four times and they still can't seem to get the check engine light to turn off... cars, mainframes... these machines are weaving looms compared to the kind of complexity robots of the future will have. How much of the outside world does your mainframe interact with? How many autonomous decisions does it make? For a machine that sits in one spot and does one job, it's not amazing that it can tell you what's wrong with it.
Dude, if robots didn't result in a net loss of employment, there would be no reason to buy them.
Wrong. First, the decision hinges on a net reduction in *costs* not employment. A net reduction of costs can cause a company to grow, which can result in a net *gain* of employment. Consider a factory which takes 10 workers on an assembly line and 10 workers to manage operations. Robots replace the 10 workers on the assembly line, allowing the company to save money, and open a second robotically-controlled factory which again takes 10 workers to manage.
Second, when you use the words "net loss" you're talking with respect to the company that purchased the robot. What about the company that sold the robot? They need workers too, and they now have money to grow and expand, creating jobs. What about the company that sells to the company that makes the robot? Now they have more money to grow and expand as well. Perhaps the workers fired from the assembly line can now be re-trained and hired elsewhere.
Third, you're neglecting the other D's. Robots are best suited for dull, dirty, and dangerous jobs. Humans are mainly replicable in the dull category, which your post and most other posts here are focusing on. But some jobs are too dirty or too dangerous for humans to manage. Consider nuclear waste disposal. Or bomb/landmine disposal. These are not suitable jobs for humans, and a net gain of jobs can be created for humans who support robots that do these jobs. Or consider jobs that are impossible for a human, like exploring the deep reaches of space or the ocean. Robots in these fields replace zero human jobs and creating jobs for those who support these robots.
Fourth, your neglecting robots that help or augment human ability. Consider an autonomous wheelchair robot that helps an elderly man navigate his town. You've enabled a man who maybe stayed in his house all day to become an active member of society. Again, you create jobs by supporting autonomous wheelchair development and support more jobs by getting this man active and involved in the community/economy and extending his life. Jobs eliminated? Maybe a single support person who used to push him around. Or maybe that person is still around but just doesn't push him anymore.
Or what about robots that fit on to a soldier, allowing him to carry more weight. The soldier is still there, so you didn't eliminate his job, you've just made him better at it. Jobs eliminated? Zero. Jobs gained? Maybe dozens to construct, build, market, and support that robot exoskeleton.
And others (including models that see better than humans and can throw and catch objects and have manual dexterity equal to humans) very close to production.
This is vastly overstated. We have robots that can see and throw and catch well in carefully controlled laboratory experiments, and published results and videos carefully present the 10 successes out of hundreds of failures. I saw a talk in the fall by Marc Raibert, the former CEO of Boston Dynamics. I'm sure you've seen the impressive Cheetah and Big Dog videos. He referenced those and then told us "and these are the videos you don't see online" and proceeded to show us about a dozen clips of big dog falling off cliffs and the Cheetah running in to a parked car at full speed. When you say these things are "very close" to production, I'd still given them 15-20 years minimum.
In 15 years, almost any non-creative job a human can do you will be able to automate at a cost lower than starvation / poverty level wages.
You'd be surprised how many jobs require even the tiniest modicum of creativity and reasoning, and how impossible even the tiniest modicum of creativity and reasoning is hard for a machine. Let's come back here in 15 years and see who is right, eh?
but that seems to be your only counter argument to the concerns others are expressing about the impact on the real job market in future.
Because we're talking about different degrees of "future", one of which is much closer than the other, and is therefore practical to consider while the other is at this point a fairy tale. When you talk about robots repairing themselves, you're talking about first diagnosing the problem, which requires logic, inference, rationalizing cause/effect, problem solving, creativity, etc. Robots of the future will be very complex, nonlinear, dynamic, interacting systems, and most likely will not be able to self-diagnose, the same way even a human cannot self-diagnose most problems. The robots I work with do some very strange things sometimes, and it takes a long time to come to the exact reason *why* it behaved as it did, and fix it, even with an intimate understanding of all the implemented systems. I can't even imagine how impossible a robot of the future will be to diagnose.
Then when you reach a diagnosis you're talking about the actual repair job, which again is a hard job that often requires some creativity and problem solving, something machines are not well suited for. We're not talking about replacing a panel and a headlight on a banged up car. We're talking about complex machines that make decisions and interact with a dynamic world in a nonlinear way. Fixing such a machine will not automatable any more than fixing a human is.
So I've used a couple of words above (creativity, problem solving, rationalization, inference) that hint at some of the deepest most profound questions of human understanding and knowledge. Talking about machines capable of these tasks is some serious science fiction. When we start talking about robots possessing these qualities, let's also start worrying about a robot apocalypse while we're at it.
That doesn't work. Ten people can maintain the machines that do the work once done by 100.
A couple of points here. First, it takes 10 people to maintain today's machines that do the work of 100 people. These machines, as I've posted elsewhere, are highly simplistic as far as robots go. Limited sensing and perception, limited cognition, very limited degrees of freedom, no mobility, specialized actuators, etc. Fixing simple machines is simple. A robot of tomorrow will be much more complex, requiring more people with more specialized knowledge to service them. Much like you have mechanics who specialize in transmissions, or even more aptly doctors who specialize in hearts, you will have robot "technicians" who specialize in perception, locomotion, "brains", electronics, drive systems, etc. Think about how many doctors a human needs, due to their sheer complexity. This is more along the lines of how a robot repair industry would develop.
Now if you could train everyone to be robot maintenance technicians that would be fine (ecological implications notwithstanding), but that's not possible
But not everyone needs to be a robot repair technician, just as not everyone in the healthcare industry is a doctor or surgeon. You've forgotten that a robot technician also works for a company. A robot technician would also be supported by non automatable non-technical jobs (management, sales, marketing, HR, legal... anything with a human-facing or creative component). I could even imagine different tiers of knowledge, where some technicians perform routine maintenance (like a nurse), some technicians simply diagnose (like a doctor), and some technicians repair (like a surgeon). Maybe someone replaced on the assembly line could re-train to a human-facing job that doesn't have to be highly technical. Will there be enough such jobs? I don't know, I can only guess. But I can see the job creation/destroyed ratio is much better than 10/100.
tl;dr - repairing robots is/will be a new *industry* not a new *job*.
K. S. Kyosuke's reply mirrors some of what I have to reply to your post, so I incorporate what he posted here, but would also like to add the following:
The most deployed robots today are industrial arms like Kukas, and I suspect that's what your brother repairs. The reason for this is because they are very simple as far as robots go. The future I'm talking about will have several orders of magnitude more robots than there are today, they will be as ubiquitous as cars, and will be so complex that they make current industrial robots look like tinker toys. Given the complexity of robots of the future, I would expect the support industry around them to resemble something like the healthcare industry, filled with specialists who focus on diagnosing or repairing the brain, electronics, perception, mechanics, etc.; general practitioners who would be like your local corner auto mechanic, who could fix the common problems and give referrals to specialists; and a highly specialized components industry which would resemble medical equipment manufacturers and distributors. An industry like this can support hundreds of thousands, if not millions of jobs.
The vast majority of work available for people throughout the world is manual labor, including trades
And most of that work isn't going away in the near future. With the current state of robots, you're talking about taking away the most dull, dangerous, and dirty jobs out there. Some robots will even have jobs that humans aren't capable of doing because they are so dangerous or dirty. Any jobs for these robots will be a net gain in employment, creating jobs surrounding and supporting the robot that were not possible before.
Again, as for those replaced by robots, well, tough. Your job is now done by a machine. Find something else to do.
If you can't see that coming then I wonder if you've given much thought to this issue at all.
If you think that's coming any time in the near or even distant future, you have absolutely zero knowledge of what robots are actually capable of. As someone who designs robots for a living, you can rest assured that humans will be the ones designing and repairing robots for a long time to come.
In this case, replacing horse drawn carriages with cars was of the same type.
It was the same "type" insofar as both made you go forward faster than walking. That's about where the similarities end between the horse/buggy industry and the automobile industry. Horsewhip makers really have no transferrable skills in a world where horsewhips don't make cars go faster. And yet the world moved on. Shocking.
And what of the millions of auto workers and those in peripheral industries who gained their jobs due to automation. I mean.... do you realize an AUTOmobile is a form of automation in itself? We automated the horse. Sure, all the horsewhip and buggy manufacturers lost their jobs, but in their place sprouted an even larger industry. I mean, there is a gas station and auto repair shop on almost every corner in my town. When robots become as ubiquitous, there will be many industries surrounding their support. Expect to see robot repair shops, with robot mechanics and technicians some time in the future.
Horse whip makers were once made obsolete, but the automation that replaced them (automobiles. Auto is right in the name!) created an industry that is now many orders of magnitude larger than the horse whip industry ever was.
The reality is, automation has about the same effect as off-shoring on productivity... the jobs go away and don't get replaced.
Maybe your job goes away. As a roboticist, I get even more job opportunities. Sorry you chose the wrong field. For those who were made obsolete by robots, well that's progress. Maybe they can retrain as someone who repairs the robots that replaced them.
It was also criticized for not having the same way to get into safe mode as previous versions and for having watered down BSODs
Wow, Windows really is reaching new lows here.
Regardless, "only interface criticism" is a pretty big one
The word only here was not used to diminish the significance of the interface criticisms, but to refute the claim that Windows 8 is "Windows 8 is shit, from top to bottom." If Windows 8 is shift from bottom to top, yet the only thing people really have a problem with is the interface, then Windows 8 isn't really shit from bottom to top now is it?
considering you can't use win8 without some apps hijacking the fullscreen or the stupid start screen hiding your desktop.
Um, you can *very easily* use even stock Windows 8 without ever using the Metro interface or a metro app. Since update 1, Windows 8.1 is now even easier to use on a desktop setup. They have addressed pretty much every criticism of the Metro UI except for the start menu at this point. As far as I can tell, most of the criticisms on this site are outdated by about 6 months with regards to the Metro UI, because people here would rather talk out of their asses than from true experience.
If a deep neural network is biologically inspired we can ask the question, does the same result apply to biological networks?
No. Artificial neural networks are inspired by biology, but that's where the similarity ends. Any conclusion drawn from an ANN should not be cast onto their biological counterparts.
I agree with everything you said except fitting an IMU in the pen. I think it wouldn't be that hard or expensive to fit one of these in a pen:
http://www.invensense.com/mems...
GPS is no longer distorted since decades, at least 15 years.
Sorry, I thought you were using a broad definition of distorted. Nominal accuracy of GPS is 1 - 10 m, depending on conditions, in my experience. I'm not sure where you're getting this idea that it's accurate to 20 cm. That's the kind of accuracy you'd expect with something like DGPS. If this were the case, competitors in the DARPA Urban Challenge wouldn't have needed DGPS and expensive inertial navigation systems. Would you care to offer a citation for this claim?
When I used the term "distorted", I was referring to multipath errors you will get from GPS signals reflecting off buildings. This, combined with low satellite visibility makes GPS-only localization a poor choice in urban environments.
With visible land marks you don't need GPS, or in other words it wont be helpfull.
Unless you have a map with absolute coordinates of those landmarks you still can't position yourself on the globe, only relative to those landmarks. With GPS you can create a map as you explore (SLAM), and refine the absolute locations of those landmarks over time using sensor fusion and filtering (GPS, LIDAR, INS). With survey results (as you say, millimeter level precision) as ground truth, we can say this method achieves sub-decimeter accuracy in global pose.
Sorry, but not being able to pinpoint your position in a city down to less than a centimeter is fail (for a robot).
Depends on what your goal is. If your only goal is to get a position, then there are better techniques. If your goal is a mobile robot that can move down a sidewalk without $60,000 worth in sensors, then sub decimeter is enough. With a wheel base of 0.6 m and an average sidewalk width of 1.5 m, a robot traveling down the center can easily be off 10 cm either way and still succeed in its operation. Typical error I've seen using the described system is around 6 cm with a minimum of 0.12 cm. This is enough to navigated sidewalks, crowds, and even through doors.
Hm, or you don't have access to accurate maps. Here in germany you buy them or even get them for free at the land-registry (and those maps are accurate below a milimeter level)
But GPS is *usually* distorted, especially as you point out next to buildings, where I want my robots to work. If I want a robot to follow a track on the ground using commodity GPS solutions, this is not possible. With GPS, and visibility to 3 landmarks, I can achieve sub decimeter accuracy.
Slashdot Mirror
... stuff that matters. On a scale of stuff that matters, this ranks pretty high.
For the kind of work you go into with a PhD, the work done during the PhD itself counts as work experience. During my PhD I was doing the same kind of work I would have been doing in industry, but at a University.
Who pays if you get into an accident using cruise control on current cars?
I use ROS for two things only: the interprocess communication system (with the nice data logging capabilities) and rviz, the data visualization tool. They can keep the rest. Also, it irks me that they call ROS an operating system, when it runs on top of Linux.
True to a point. One thin I've found with ROS is it encourages a lot of "black box" development. "connect these components and you'll have a working robot!" And you do.... until it doesn't work. And then you don't know why. One of the big promises of ROS is code reuse, but in my experience, the stuff that works reliably isn't particular sophisticated to begin with, and the rest you really should be writing yourself to ensure a robust robot you can actually debug. We've tried the "connect black boxes" approach, and by the time you've got a system that works the way you want, you've rewritten most of the code anyway.
I typically don't use adblock. In general, I don't mind most ads. If a site has really bad ads that are intrusive and annoying, I just avoid it. There's enough content on the web. But it was Slashdot's idea to say "Hey, if you stick around and make comments, and the community thinks you're adding to the discussion, then we won't show you any ads" and that's been the agreement. So as the ads Slashdot displays get larger and more numerous and more intrusive, when they don't hold up to that agreement it's very frustrating, because I frequent this site a lot and would rather not move on because of the ads.
The kind of processing done for the Kinect is well suited a GPU: facial recognition, skeleton tracking, hand tracking, voice recognition. These feature use a lot of matrix math, geometric algorithms, and signal processing... stuff you want to do on a highly parallel processing architecture. So yeah, doing them on a serial architecture like an x86 CPU would be much more taxing.
I've been seeing ads for a while now with Disable Advertising checked. At first it was just the ad under the story and the annoying pop up ad at the bottom of the screen. Unchecking and checking again would make them go away. Now all the ads are enabled and re-checking the box does nothing.
Seriously, this shit is incredibly annoying. The ads are obnoxious, gigantic, and intrusive. I've got a little checkbox that says "Thanks again for helping make Slashdot great!" under it, and it does nothing. So where's the thanks, exactly? Definitely not feeling it.
you're still employing only one man for every hundred people worth of productivity
You're still making two mistakes. One assuming that every robot is used to kill already existing jobs, and two assuming that for every job lost two more cannot be created. I've already argued to second point in several other posts, but I'd like to try and convince you on the first point where I believe you are mistaken. In this post I argue several points where robots can create a net gain in jobs, some of which cause the loss of no jobs at all. The problem you and others are stuck on is arguing about current robots while supposing a future robot. To this point, all robots have ever done is displace factory workers through automation. This is because they are uniquely suited for this type and work and really nothing else.
But robots of the future will be suited for much more, to the extent that they 1) augment human abilities and 2) extend to endeavors otherwise inaccessible to humans. In these scenarios, you aren't replacing anybody, or maybe a few people, while enabling work for dozens, hundreds, or even thousands by creating jobs were none previously were able to exist. If we're talking about robots of the future, you have to give them more capabilities than just those that replace menial labor.
And it's worse than that, because unlike humans robots lend themselves to modularity - there's no sense in maintaining a vision-systems repair specialist if you can simply unplug the faulty system and replace it with a new one more cheaply.
Again, you are assuming a robot of today for a robot of the future. Robots *today* lend themselves to modularity. Robots of the future? Already we see emergent behavior we can't fully comprehend when a whole system is working together as a whole. For instance (I don't like to go down the sci-fi very distant future path, but since everyone here likes to argue about machines with extreme intelligence replacing humans I'll do so for the moment), consider a robot brain with billions of interacting neurons. Perhaps it's not possible to download and replicate such a thing. Consider robot with some proportion of biological parts. Perhaps such a thing is not easy to repair. I don't know that's just a little sci-fi there.
Back to reality and with respect to vision in particular, we employ many computer vision experts. Robot vision gone wrong isn't just about the camera. When robots are seeing the wrong thing, sure you can replace their vision systems, swap in a new one, but then when it's still seeing the wrong thing, what do you do? It takes a specialist to figure it out. This is the difference between a robot repair man and a TV repair man; one is dealing with a single-purpose solid state device, while the other is dealing with a highly nonlinear system which is interacting with its environment and making decisions on its own. One of the two is exponentially more complex to service.
You only need a human involved for non-obvious problems, which probably means someone who is reasonably well versed in *all* the component systems of at least a specific class of robot
And you expect such a person to exist? Already in robotics we have dozens of specializations.
Not something Joe Sixpack can reasonably be expected to learn to do.
That's why Joe Sixpack doesn't become a surgeon when he's displace from his job, he trains to be a nurse or an orderly, or a sales rep. And as for the technician support staff - how many do you really need?
How many does a hospital need? There are many more nurses than there are doctors and surgeons.
just look at what's happened with customer service in the last decades, it used to be you talked to actual humans on the phone, now you've got to go through several layers of poorly-automated systems design
Self repairing machines maybe science fiction now, but so were cell phones with internet browsers in 1995
Smartphones were an exercise in making things we already had in 1995 smaller and faster. i.e. largely an engineering task. When you talk about robots that can repair themselves i.e. something which requires creativity and intelligent thought, you're talking about pushing the boundaries of several scientific disciplines far beyond where they are today. That is to say, in 1995 you could definitively answer that yes, it's theoretically possible to create an internet enabled hand held device. Today, there is no theoretical basis for even believing machine learning can *ever* match the reasoning or creative capabilities of a human brain.
If that ever gets significant progress it wouldn't be too far fetched for machines to self diagnose and self repair.
Good luck to them. Hopefully they make some headway by the time I'm 120 and can upload my consciousness to the internet.
I heard a VR software developer say "People overestimate technological change for a year, but under estimate change when you talk about a decade."
And yet self driving cars and true artificial intelligence are always "ten years away".
So its worth to give a bit of thinking on what happens when machine learning is good enough to eliminate current jobs and all possible jobs after that.
The same amount of thinking I'm willing to give what I should do after the sun explodes or the heat death of the universe. Thinking about such things is a flight of fancy, fueled by a science fiction and a popular media depiction of machine learning and AI.
And they'll be designed to make it easy to repair them. Hot swappable modules for each major component. Easily automatized repair. Most broken modules won't be repaired. The goal will be minimal downtime (we had contracts for under 4 hours unscheduled downtime per year). So that means the entire unit, or an entire module is swapped out and the unit is functional again.
You're arguing about the future while thinking about the complexity and capability of robots today, and again (as I argued in another post to you) considering only robots that replace factory labor... which robots of the future will do of course but go well beyond. You think it's as easy as just swapping out components? Maybe my future robot butler keeps knocking over my lamp with its arm. So I swap out the arm and it keeps doing it. I swap out the eye and it keeps doing it. I swap out the brain and it keeps doing it.... now what? This is a common problem that I face every day... debugging and repairing robots is not as easy as "hot swapping" modules because we're talking about machines which interact with their (nonlinear dynamic) world and make choices on their own based on input. Robots of the future that everyone is so worried about will not be your grandfather's robots.
Specialists cost money and are likely to only be used in the initial design, creation, and debugging of the robots
You think the debugging stops when it's shipped? With robots it's an ongoing process.
Our mainframes today already self analyze and even send emails saying they need a specific part replace. Heck- our automobiles tell repairmen what part is broken.
And yet my car has been to the mechanic four times and they still can't seem to get the check engine light to turn off... cars, mainframes... these machines are weaving looms compared to the kind of complexity robots of the future will have. How much of the outside world does your mainframe interact with? How many autonomous decisions does it make? For a machine that sits in one spot and does one job, it's not amazing that it can tell you what's wrong with it.
Dude, if robots didn't result in a net loss of employment, there would be no reason to buy them.
Wrong. First, the decision hinges on a net reduction in *costs* not employment. A net reduction of costs can cause a company to grow, which can result in a net *gain* of employment. Consider a factory which takes 10 workers on an assembly line and 10 workers to manage operations. Robots replace the 10 workers on the assembly line, allowing the company to save money, and open a second robotically-controlled factory which again takes 10 workers to manage.
Second, when you use the words "net loss" you're talking with respect to the company that purchased the robot. What about the company that sold the robot? They need workers too, and they now have money to grow and expand, creating jobs. What about the company that sells to the company that makes the robot? Now they have more money to grow and expand as well. Perhaps the workers fired from the assembly line can now be re-trained and hired elsewhere.
Third, you're neglecting the other D's. Robots are best suited for dull, dirty, and dangerous jobs. Humans are mainly replicable in the dull category, which your post and most other posts here are focusing on. But some jobs are too dirty or too dangerous for humans to manage. Consider nuclear waste disposal. Or bomb/landmine disposal. These are not suitable jobs for humans, and a net gain of jobs can be created for humans who support robots that do these jobs. Or consider jobs that are impossible for a human, like exploring the deep reaches of space or the ocean. Robots in these fields replace zero human jobs and creating jobs for those who support these robots.
Fourth, your neglecting robots that help or augment human ability. Consider an autonomous wheelchair robot that helps an elderly man navigate his town. You've enabled a man who maybe stayed in his house all day to become an active member of society. Again, you create jobs by supporting autonomous wheelchair development and support more jobs by getting this man active and involved in the community/economy and extending his life. Jobs eliminated? Maybe a single support person who used to push him around. Or maybe that person is still around but just doesn't push him anymore.
Or what about robots that fit on to a soldier, allowing him to carry more weight. The soldier is still there, so you didn't eliminate his job, you've just made him better at it. Jobs eliminated? Zero. Jobs gained? Maybe dozens to construct, build, market, and support that robot exoskeleton.
And others (including models that see better than humans and can throw and catch objects and have manual dexterity equal to humans) very close to production.
This is vastly overstated. We have robots that can see and throw and catch well in carefully controlled laboratory experiments, and published results and videos carefully present the 10 successes out of hundreds of failures. I saw a talk in the fall by Marc Raibert, the former CEO of Boston Dynamics. I'm sure you've seen the impressive Cheetah and Big Dog videos. He referenced those and then told us "and these are the videos you don't see online" and proceeded to show us about a dozen clips of big dog falling off cliffs and the Cheetah running in to a parked car at full speed. When you say these things are "very close" to production, I'd still given them 15-20 years minimum.
In 15 years, almost any non-creative job a human can do you will be able to automate at a cost lower than starvation / poverty level wages.
You'd be surprised how many jobs require even the tiniest modicum of creativity and reasoning, and how impossible even the tiniest modicum of creativity and reasoning is hard for a machine. Let's come back here in 15 years and see who is right, eh?
but that seems to be your only counter argument to the concerns others are expressing about the impact on the real job market in future.
Because we're talking about different degrees of "future", one of which is much closer than the other, and is therefore practical to consider while the other is at this point a fairy tale. When you talk about robots repairing themselves, you're talking about first diagnosing the problem, which requires logic, inference, rationalizing cause/effect, problem solving, creativity, etc. Robots of the future will be very complex, nonlinear, dynamic, interacting systems, and most likely will not be able to self-diagnose, the same way even a human cannot self-diagnose most problems. The robots I work with do some very strange things sometimes, and it takes a long time to come to the exact reason *why* it behaved as it did, and fix it, even with an intimate understanding of all the implemented systems. I can't even imagine how impossible a robot of the future will be to diagnose.
Then when you reach a diagnosis you're talking about the actual repair job, which again is a hard job that often requires some creativity and problem solving, something machines are not well suited for. We're not talking about replacing a panel and a headlight on a banged up car. We're talking about complex machines that make decisions and interact with a dynamic world in a nonlinear way. Fixing such a machine will not automatable any more than fixing a human is.
So I've used a couple of words above (creativity, problem solving, rationalization, inference) that hint at some of the deepest most profound questions of human understanding and knowledge. Talking about machines capable of these tasks is some serious science fiction. When we start talking about robots possessing these qualities, let's also start worrying about a robot apocalypse while we're at it.
That doesn't work. Ten people can maintain the machines that do the work once done by 100.
A couple of points here. First, it takes 10 people to maintain today's machines that do the work of 100 people. These machines, as I've posted elsewhere, are highly simplistic as far as robots go. Limited sensing and perception, limited cognition, very limited degrees of freedom, no mobility, specialized actuators, etc. Fixing simple machines is simple. A robot of tomorrow will be much more complex, requiring more people with more specialized knowledge to service them. Much like you have mechanics who specialize in transmissions, or even more aptly doctors who specialize in hearts, you will have robot "technicians" who specialize in perception, locomotion, "brains", electronics, drive systems, etc. Think about how many doctors a human needs, due to their sheer complexity. This is more along the lines of how a robot repair industry would develop.
Now if you could train everyone to be robot maintenance technicians that would be fine (ecological implications notwithstanding), but that's not possible
But not everyone needs to be a robot repair technician, just as not everyone in the healthcare industry is a doctor or surgeon. You've forgotten that a robot technician also works for a company. A robot technician would also be supported by non automatable non-technical jobs (management, sales, marketing, HR, legal... anything with a human-facing or creative component). I could even imagine different tiers of knowledge, where some technicians perform routine maintenance (like a nurse), some technicians simply diagnose (like a doctor), and some technicians repair (like a surgeon). Maybe someone replaced on the assembly line could re-train to a human-facing job that doesn't have to be highly technical. Will there be enough such jobs? I don't know, I can only guess. But I can see the job creation/destroyed ratio is much better than 10/100.
tl;dr - repairing robots is/will be a new *industry* not a new *job*.
K. S. Kyosuke's reply mirrors some of what I have to reply to your post, so I incorporate what he posted here, but would also like to add the following:
The most deployed robots today are industrial arms like Kukas, and I suspect that's what your brother repairs. The reason for this is because they are very simple as far as robots go. The future I'm talking about will have several orders of magnitude more robots than there are today, they will be as ubiquitous as cars, and will be so complex that they make current industrial robots look like tinker toys. Given the complexity of robots of the future, I would expect the support industry around them to resemble something like the healthcare industry, filled with specialists who focus on diagnosing or repairing the brain, electronics, perception, mechanics, etc.; general practitioners who would be like your local corner auto mechanic, who could fix the common problems and give referrals to specialists; and a highly specialized components industry which would resemble medical equipment manufacturers and distributors. An industry like this can support hundreds of thousands, if not millions of jobs.
The vast majority of work available for people throughout the world is manual labor, including trades
And most of that work isn't going away in the near future. With the current state of robots, you're talking about taking away the most dull, dangerous, and dirty jobs out there. Some robots will even have jobs that humans aren't capable of doing because they are so dangerous or dirty. Any jobs for these robots will be a net gain in employment, creating jobs surrounding and supporting the robot that were not possible before.
Again, as for those replaced by robots, well, tough. Your job is now done by a machine. Find something else to do.
If you can't see that coming then I wonder if you've given much thought to this issue at all.
If you think that's coming any time in the near or even distant future, you have absolutely zero knowledge of what robots are actually capable of. As someone who designs robots for a living, you can rest assured that humans will be the ones designing and repairing robots for a long time to come.
In this case, replacing horse drawn carriages with cars was of the same type.
It was the same "type" insofar as both made you go forward faster than walking. That's about where the similarities end between the horse/buggy industry and the automobile industry. Horsewhip makers really have no transferrable skills in a world where horsewhips don't make cars go faster. And yet the world moved on. Shocking.
And what of the millions of auto workers and those in peripheral industries who gained their jobs due to automation. I mean.... do you realize an AUTOmobile is a form of automation in itself? We automated the horse. Sure, all the horsewhip and buggy manufacturers lost their jobs, but in their place sprouted an even larger industry. I mean, there is a gas station and auto repair shop on almost every corner in my town. When robots become as ubiquitous, there will be many industries surrounding their support. Expect to see robot repair shops, with robot mechanics and technicians some time in the future.
Prove it.
Horse whip makers were once made obsolete, but the automation that replaced them (automobiles. Auto is right in the name!) created an industry that is now many orders of magnitude larger than the horse whip industry ever was.
The reality is, automation has about the same effect as off-shoring on productivity ... the jobs go away and don't get replaced.
Maybe your job goes away. As a roboticist, I get even more job opportunities. Sorry you chose the wrong field. For those who were made obsolete by robots, well that's progress. Maybe they can retrain as someone who repairs the robots that replaced them.
It was also criticized for not having the same way to get into safe mode as previous versions and for having watered down BSODs
Wow, Windows really is reaching new lows here.
Regardless, "only interface criticism" is a pretty big one
The word only here was not used to diminish the significance of the interface criticisms, but to refute the claim that Windows 8 is "Windows 8 is shit, from top to bottom." If Windows 8 is shift from bottom to top, yet the only thing people really have a problem with is the interface, then Windows 8 isn't really shit from bottom to top now is it?
considering you can't use win8 without some apps hijacking the fullscreen or the stupid start screen hiding your desktop.
Um, you can *very easily* use even stock Windows 8 without ever using the Metro interface or a metro app. Since update 1, Windows 8.1 is now even easier to use on a desktop setup. They have addressed pretty much every criticism of the Metro UI except for the start menu at this point. As far as I can tell, most of the criticisms on this site are outdated by about 6 months with regards to the Metro UI, because people here would rather talk out of their asses than from true experience.
If a deep neural network is biologically inspired we can ask the question, does the same result apply to biological networks?
No. Artificial neural networks are inspired by biology, but that's where the similarity ends. Any conclusion drawn from an ANN should not be cast onto their biological counterparts.
For some people, careers aren't about ROI. I wouldn't do mobile app development for a million dollars.
I agree with everything you said except fitting an IMU in the pen. I think it wouldn't be that hard or expensive to fit one of these in a pen: http://www.invensense.com/mems...
GPS is no longer distorted since decades, at least 15 years.
Sorry, I thought you were using a broad definition of distorted. Nominal accuracy of GPS is 1 - 10 m, depending on conditions, in my experience. I'm not sure where you're getting this idea that it's accurate to 20 cm. That's the kind of accuracy you'd expect with something like DGPS. If this were the case, competitors in the DARPA Urban Challenge wouldn't have needed DGPS and expensive inertial navigation systems. Would you care to offer a citation for this claim?
When I used the term "distorted", I was referring to multipath errors you will get from GPS signals reflecting off buildings. This, combined with low satellite visibility makes GPS-only localization a poor choice in urban environments.
With visible land marks you don't need GPS, or in other words it wont be helpfull.
Unless you have a map with absolute coordinates of those landmarks you still can't position yourself on the globe, only relative to those landmarks. With GPS you can create a map as you explore (SLAM), and refine the absolute locations of those landmarks over time using sensor fusion and filtering (GPS, LIDAR, INS). With survey results (as you say, millimeter level precision) as ground truth, we can say this method achieves sub-decimeter accuracy in global pose.
Sorry, but not being able to pinpoint your position in a city down to less than a centimeter is fail (for a robot).
Depends on what your goal is. If your only goal is to get a position, then there are better techniques. If your goal is a mobile robot that can move down a sidewalk without $60,000 worth in sensors, then sub decimeter is enough. With a wheel base of 0.6 m and an average sidewalk width of 1.5 m, a robot traveling down the center can easily be off 10 cm either way and still succeed in its operation. Typical error I've seen using the described system is around 6 cm with a minimum of 0.12 cm. This is enough to navigated sidewalks, crowds, and even through doors.
Hm, or you don't have access to accurate maps. Here in germany you buy them or even get them for free at the land-registry (and those maps are accurate below a milimeter level)
The robot makes the map itself, as I described.
But GPS is *usually* distorted, especially as you point out next to buildings, where I want my robots to work. If I want a robot to follow a track on the ground using commodity GPS solutions, this is not possible. With GPS, and visibility to 3 landmarks, I can achieve sub decimeter accuracy.
Windows 8 always had the classic desktop and you could always switch back and forth with the press of a button.