because it doesn't see any practical market for the tech.

Again, the robots in the video are working in a controlled environment, handing specifically defined objects, for a limited amount of time. As soon as you start changing these variable, the robots immediately run into problems. Each new variable you add, increases the complexity of the code, which requires more CPU power, which requires more electrical powers. The more powerful the CPU, the hotter it runs, and the more electricity it needs to cool itself, and the bulkier it becomes.

Watson requires hundreds of kilowatt hours of electricity to run, the human brain manages just fine on a tiny fraction of that.

Machines are great at multiplying muscle power, but very poor at being actually intelligent.

and therefore its taking them too long to develop applications for said robots.

Seems reasonable, I do think your prediction of it not happening in this century vastly underestimates the rate of technological development, a decade, perhaps two, seems more reasonable.

and they are not trivial. Simply coming up with a power source to run the robot for a reasonable length of time is decades off. What good is a robot that can work for 15 minute, then has to recharge for 45? Or if you give it enough battery capacity to work a significant amount of time, you triple it's weight? Simulating human intelligence is about massive parallel computing, which consumes huge amounts of electricity, and then doubling that number to add a cooling system to shed the heat, which also adds more weight. Humans manage the same feat in a package weighing in at under 200 pounds and consuming a few kilowatt hours of power daily.

The same people who have been telling us AI is very close, have been telling us that fusion reactors, flying cars, and lightweight, energy dense batteries are just 5-10 years away, have been telling us that for decades. People throw out words like "quantum computing", "neural networks", etc like sci-fi writers throw out "warp drive" and Heisenberg Compensators. They sound impressive, but in reality they either don't exist, or don't really work as simply as most people think they do.

The human brain is the end product of millions of years of evolution. It is a marvel of chemical/electrical/bio-engineering. We are not going to invent its replacement in a few decades.

the same way we laugh at the inventors of "flying machines" who attempted to replicate the flapping of a bird's wings.

There is something we fundamentally don't (can't?) understand about the nature of consciousness and intelligence that makes our attempts to replicate it using bits, bytes and electrical current doomed to failure.

"Look, it can fly", how can we KNOW when a machine can think as we do? Hell, how can we prove we actually exist outside our own head.

to do it. It looks at words associated with the image, which is why many times you can type in a term and you will get pictures of things unrelated to the search term.

Watson is attempting to do what the human brain does: Look at a picture and identify whats in it by context alone. This task it fails at miserably.

Have you looked at Google photos lately?

I can type in dog and it will pull my pictures with dogs in it or tree and it will pull pictures with trees. There is no way it is using tags for that.

Google simply searches links for images that are tagged in relation to dogs. The more common the subject, the more tags to the image, the more accurate the result.

For example, type in "Miguelito Loveless", the villain from the 60s show The Wild, Wild, West, You will get a lot of photos of Michael Dunn, the actor who played him in character. However, as you look through the results you will find photos that are not Michael Dunn, or the character Miguelito Loveless, such as these:




http://3.bp.blogspot.com/_RaOrchOImw8/TAw6hije6fI/AAAAAAAAj7I/1DXkvMy497E/s1600/Linda+Hunt.jpg
http://ia.media-imdb.com/images/M/MV5BMjAyNjI0NTkyNF5BMl5BanBnXkFtZTgwMDcwNzU1MTE@._V1_UY1200_CR485,0,630,1200_AL_.jpg



http://www.famousfix.com/post/stephen-hawking-wants-to-play-a-james-bond-movie-villain-39292915

I think we can agree that Ted Cassidy, Ida Lupino, Julie Newmar, Roy Thinnes and Peter Sellers are NOT either Michael Dunn or Miguelito Loveless.

The reason they get returned is that they happen to be on a page that also refers to the search term.

If Google actually could "recognize" images, it would know that Julie Newmar is NOT Miguelito Loveless.

You can use image search to try and identify an object and Google will search for images similar to the one you provide and show you what the image has been tagged as. Again, the more common the image, the more accurate the result. However, when I load personal photos, especially of ones that have never been on line, accuracy drops. I have a number of photos of Leonard Nimoy and George Takei (taken at various cons in the 70s), people easily identified by humans, but Google image search can't nail them down because my photos are candid shots that have never been online before.

Two different things.

Google photos searches my images which have no tags other than what the camera adds with no input from me.

And with no tags whatsoever it can pull pics of trees or dogs or balls or any number of things.

Take another look

or the images in the Google Photo database. Also, most photos these days, especially those taken with phones have a LOT of useful metadata embedded in them (date, time, GPS location, any tags added by social media apps, email apps, etc), which can help ID an image. Also, when you upload your photos and start sorting them, you are giving Google more data to mine for pattern matching.

But, as I explain, upload an image that has never been online, has no metadata, like that of a person, and Google is stumped. A high percentage of people in America will look at my photo of George Takei and will be able to identify him by name, or as Mr. Sulu, or the guy from Star Trek, whereas Google is baffled.

Pattern matching is not recognition. And even when you get pattern matching into the high 90 percentile, it is still way inferior to the human brain.

From your own popular mechanics link google outperforms humans in image recognition.

linked to in the article you mentioned:

http://www.popularmechanics.com/technology/a14116/microsoft-image-recognition-outperforms-humans/

Microsoft scientists have released a paper claiming that their visual recognition software is outperforming humans, a long-sought after feat. The benchmark for humans in image classification is a 5.1 percent error rate. Microsoft claims their error rate is 4.94 percent, beating out Google's 6.66 percent.
...
This recent success a demonstration of Microsoft's deep learning capabilities—though as the company points out, some of its misses were pretty glaring, making "mistakes in cases that are not difficult for humans, especially for those requiring context understanding or high-level knowledge." The researchers also acknowledge that the demonstration related to the ImageNet dataset its algorithms were given, rather than a sort of "in the wild" approach.

Thus, a routine looking for a cat—but shown a picture of the sky—will begin to pick out objects it interprets as a cat even when a furry feline isn't there.

Point is that these neural nets are outperforming humans. Ivt doesn't matter that they were given information before hand unless you want to pretend any machine designed for the task would not be given the information it needs to handle the task given to it.

Fact is the idea that machines can not identify Images is false and they are getting better at it very quickly.

You need to rethink your justification for robots not being able to replace humans and find another metric beyond image recognition.

The fact that Google just won a go competition is not insignificant and is an amazing leap from Chess.

AI is getting better and at an amazing clip there is no denying it and image recognition is certainly not a point of failure.

neual networks are outperforming humans. I am waiting to see the peer reviewed evidence. Also, Microsoft claims to outperform Google, yet admits to optimizing the data set. They also admit that when the software made mistakes, they were glaringly obvious ones that humans would not have made.

This is hardly "quibbling". The devil is, as they say, in the details.

Setting that aside, assuming you have a neural network that can pattern match better than a human. How does a warehouse full of computers, consuming megawatts hours worth of electricity, become a robot that can make burgers at McDonalds?

The 1950 supercomputer could do less than a basic handheld calculator
Inject quantum computing into Google's DeepMind, and you'll see wonders.

Also, 1950 was 66 years ago, and back then they were predicting that we would have genuine AIs by now. And yet, here we are, and a genuine AI still does not exist (also, no moon colony, no atomic reactors in the basement, no jetpacks or flying cars).

Also, it is all fine and dandy to talk about "quantum computing" but even if you manage to pull off such a piece of hardware, who will write the software? Who will create a program that can actually simulate the human mind?

No "genuine AI"? I suppose it depends what you call genuine.
But world champ at chess and Go isn't too shabby. Even before quantum computing.

but, look at the Herculean task it was to develop a machine that could beat a Chess and Go Master.

Yet, I know several people who mastered a number of disciplines in far less time and with far less money. I follow a young woman on Youtube who is a school teacher fluent in German, English, Dutch, Spanish and French, who also writes, is an excellent carpenter, keeps chickens, and edits professional quality video. How much code would have to be written, over how much time, costing how many billions to match her skill set developed over the course of a few decades?

How about another woman who is a master violinist, dancer, composer, and choreographer?

How about someone like me, who can simply run a really enjoyable game of D&D?

with some attention to grammar and syntax. It doesn't "speak" (nor understand) the language, and is easily tripped up by idioms, metaphors dialects and context. This is not likely to change due to the fact that languages are constantly evolving.

A fun little experiment is to take a single sentence in English, translate it to language x, then take that result and translate it to language y, then take that result and translate it to language z, then take that result and translate it back into English. A multilingual human would be able to render it back exactly, a computer can't.

"Space, the final frontier" translated to Italian, then to Danish, then to Latin, then back to English comes back as "Space, end".

For another take on this, I refer you to this thought experiment:

https://en.wikipedia.org/wiki/Chinese_room

in other posts. It was part of a multi-billion dollar research project that IBM then spent six years optimizing to play Jeopardy. So, you have a a multi-ton computer that uses huge amounts of electricity, generates massive amounts of heat (which then requires a complex cooling system and its own power source) and required years of programming to create a system capable of operating (and winning) in a very controlled environment.

Hardly a device that will fit inside Commander Data's head and allow him to sling burgers at McDonalds.

I was just following the flow of our discussion
- I mentioned AI could perform at the highest level (chess, Go)
- you countered with a human speaking different languages
- I answered with Google-Translate
- you objected Google-Translate's inability to understand context
- I answered Watson
- you now mention the size of watson
- we're back to my point about DeepMind meets quantum computing

practical robots that will start taking people's jobs are not going to happen anytime soon. You came into the discussion and offered different points which I then rebutted individually. I have stipulated that yes, Watson can do context, but only in one language and only by using a machine so big, so energy hungry, and so expensive that it fails to be useful for anything other than what it was specifically designed for, playing Jeopardy. You also claimed that a computer was better at languages than humans were, and I rebutted that as well.

Booking a plane: before, you had to talk to humans.
You can now find your flight and buy your ticket online.

and agencies, still people to answer the phone and deal with problems the automated systems cannot handle. Even in my town I can still find a dozen or so in the Yellow Pages.

Plus, even at the online sites, like Expedia, Price Line, Travelocity, Kayak, Hotline.com, et al, we find thousands of people employed in running the companies, from writing the software, to maintaining the hardware, to negotiating with the airlines, to customer service.

So robots have displaced human jobs. Which was your OP contention. q.e.d.

I once had a summer job at a tomato canning factory. Robots took over some jobs which used to be done by humans. First, the sealing of the cans. Then the refilling of can tops, Progressively, the chain got more and more automated, one job at a time, between 1950 to now.

I think the difference between us is that you're talking about robots that would look like humans. It doesn't need to be that way.

running around any factories or slinging burgers at McDonalds? This was the context of the posts generated by that video and which I was addressing. Wholesale outcry that robots would be replacing humans en masse. Hasn't happened yet, not going to happen any time soon.

That was the point I was addressing, not the point that automation or some industrial robots haven't been brought in to do jobs. The company I work for has been using automation and industrial robots for over two decades. We now employ almost four times as many people as we did before we bought the technology.

Now, before you point out that we would have hired even more people if we hadn't automated, let me explain that no we wouldn't, because we would have been out of business. Automation allowed us to increase production, take on bigger jobs, and take on a wider variety of jobs. We also pay twice the minimum wage on average.

Our automation augments humans, it does not replace them. We still need people to run these sophisticated machines, and handle the greater volume of business, business we would not have had if we had insisted on hammering out metal on an anvil instead of buying sophisticated computer controlled machines.

Again, the post was in response to the video that was being offered as proof that robots were about to replace our workforce and thus industrial robots and automation were bad. My counter was that the robots in the video were great in the lab for making Youtube videos, but impractical in the real world. Automation was great for boring, repetetive tasks, and/or dangerous tasks. They also allowed greater precision and higher production which meant more business, which meant more decent paying jobs.

Is this true in every case? No, certainly companies like Amazon would love to get rid of as many people as possible, and pay what remains a pittance. But Amazon is NOT a manufacturing company, which is what we are discussing. Also "free trade" agreements like TPP pose a much greater threat to manufacturing jobs than robots.

AI or robots do not need to look like humans. Watson doesn't need to look like a robot. The machine that replaced me did not look like a person. Yet both AI and robots can and do replace humans in jobs, freeing human time for better purposes.

And I'm not sure it's a bad thing "free trade" agreements like TPP replace manufacturing jobs. manufacturing jobs were mostly mechanic tasks better left to machines, save for soem very high skill, high value added jobs (mostly in luxury or high tech goods)

I encourage everyone interested to read this lenghty but informative article
http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html
http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html

but I argue that it is for the wrong reasons.

We have a VERY stupid military that wants to give weapons to software driven machines, software that is woefully incapable of the task it is being delegated and will make for very bad outcomes. Human beings fail miserably at making ethical and moral decisions (see Donald J. Trump, presidential candidacy of). Software, and lets be honest, that's all this is, a program, is the ultimate in "just following orders". There are NO ethical or moral dimensions for software unless it is programmed in, and even then, we have made a conscious decision to NOT implement Asimov's Laws of Robotics, so a bloodbath will be the ultimate outcome.

Yes, computing power is increasing at a steep rate, yet we are now starting to hit the wall of physical limits. ICs can only get so small before they start becoming unreliable due to quantum effects. The clock speed of CPU chips has pretty much stagnated for over a decade because we can't run the chips any faster without elaborate, energy hungry, cooling systems. Also, the more complex the hardware, the more complex the software needed to run it, and humans must write that software.

I see AI the way I see sustainable fusion power, something that has been perpetually 5-10 years away, that stays 5-10 years away.

Fast food is the easiest to imagine. How many hundreds of thousands of people work in fast food in our country?

It's probably not even just fast food, as actual sit down restaurants can utilize robotics to replace their human employees and risk associated with them.

I can foresee grocery store check out clerks, butchers, gas station attendants, and a myriad of other jobs getting replaced as well.

It's probably not even just fast food, as actual sit down restaurants can utilize robotics to replace their human employees and risk associated with them

while humans pace of evolution is rather slow. It isn't hard to see a point where the two in terms of intelligence meet and cross. If the op-ed writer thinks 10 to 20 years it isn't possible... well in 1000 to 2000 it seems almost certain. The future generally arrives before you know it.

There would have to be multiple breakthroughs in a variety of sciences before robots will pose a threat to human manufacturing at the level people are worried about. And after you figure out the science, then you have to work out the economics. Then there are a host of legal and regulatory issues to deal with, then the social obstacles. These type of robots have been predicted as just 5-20 years away for the last sixty years. They will eventually come, but probably not for another 50+ years at best.

The point of this post was to address a lot of doom-laden posts that were proliferating of late, generating fear, uncertainty and doubt. When people are afraid, they are easy to manipulate. My contribution to the discussion was "scientia est potentia", knowledge is power

Unless one has a pacemaker, of course.

But the distance between Einstein and a dumb human should be greater.
I'd trust some chimps over some humans.

People often use the word "robot" as a catch-all term for all types mechanical things controlled by computer programming. Repetitive work is already being mechanized and performed by single function automated devices in many types of factories that are trying to cut costs by replacing human labor. This process will expand as advances in technology find ever increasing uses for more high tech machines. With wider use comes increased demand which will make this automated equipment even more cost effective than hiring workers.

Also, the company I work for added automation years ago. Back when we started automating we had about 30 employees. With automated systems we now employ almost 4 times that number, so I don't see the problem. We still need people because automation can only do so much.

It's hardly, I suspect, automation cet par driving increased headcount.

but automation allowed us to handle bigger and a wider variety of jobs that would could not have gotten without the equipment. The end result was still more jobs, better pay. Also, we produced better quality parts which has resulted in some customers bringing work back into the country from Asia because we have fewer defects and put parts on their dock faster.

I design control systems for manufacturing processes, automated production lines, and specialty machines.

We won't eliminate the need for labor in the foreseeable future, but generally speaking, the trend is for fewer and fewer people required for many manufacturing processes. It's safe to say the trend is essentially permanent. Advances in materials and technology, including microprocessors, controllers, servos, sensors, etc. are making automation easier and cheaper every day.

Good, bad, or indifferent, that's what's happening.

We use a lot of automation in our plants, but it has only resulted in us being able to handle more jobs, which has meant hiring more workers.

Unfortunately, the future of robots is ED-209, not C3PO.

then shit will hit the fan.

The abilities of computer tech and AI are accelerating at an exponential rate, and before anyone is prepared for it, their abilities will leap frog us.

CPU power has hit a wall. You can only make chips so small, before they become unreliable. Also, the smaller they are, the hotter they run, meaning we now have to dedicate resources and energy to cooling the system. Watson weighs in at a couple of tons and eats hundreds of kilowatts hours of power, and yet it can't identify a picture of itself, something a three year-old can do.

We have had 64 bit processing with multiple cores and multiple threads for over a decade and we have yet to find an operating system that uses more than a fraction of those capabilities. Why? Because of the leviathan task of writing (and debugging) such an OS.

As I explain in other posts, I work with a company that very much jumped into automation, with great success. The end result over 20 years has been constant: We hire more people (and at about double the minimum wage).

Robots can do manual labor tasks, which could replace enough people overall to cause unrest.

which means they must either be tethered to a plug, or haul around a battery which limits their operational time. We can make batteries bigger, but that means greater bulk and weight. Robots work fine in controlled environments with flat floors and predictable, well-defined tasks. Change any of these parameters and failure rates climb.

Again, go back to the article I mention above. The most sophisticated computer on the planet, weighing thousands of pounds can't look at a picture and tell you what it is, it can only guess with varying degrees of (un)certainty. Technology is NOT going to solve these problems anytime soon. We have been trying for half a century, and while we have taught computers some neat tricks, we are still light years away from anything actually simulating even the most basic mind.

While some simple tasks a computer can do, are impossible for us. Until we see a true artificial intelligence (if ever), robots will remain clunky automatons that can perform simple tasks that are dangerous for humans to do (I'm thinking about robots that are bomb sniffers).

Take multiplication. Computers basically cant do it, they simply do addition REAL fast. They can sort data in microseconds, but can't UNDERSTAND any of it.

is vital to their performance and right now that is impossible.

I'm sure he could use a laugh about now.

(at a cost of billions) can now beat a human at Go. And Watson can beat a human at Jeopardy.

Whatever will we do with the massive unemployment on the game show contestant and professional Go circuit?

But carry on regardless and have fun!

I have been working in the computer field for three decades and the last eight in a manufacturing environment with robots and automated systems.

I have have a solar array on my roof, drive an electric car, and wired my house with a computer network, and automated it, for fun. And even the most simple systems I am working with are not reliable enough that I would bet my life on them. Every sensor I add, every switch, every light, transmitter, etc, adds a layer of complexity, and therefore a point of failure.

I am hardly naive about tech, nor a Luddite. I love tech, but understand its limitations.

I am constantly amazed at how impressive technology is, and how stupid it is. More impressive is the blind faith many people put in technology, even to the point of disparaging the millions of years of human evolution that resulted in the brain that creates the technology.

We "learn" to "see" things. And we call that "remembering" them. Two people walk through the same forest, but they never see hear, or feel exactly the same forest. What we truly excel at is pattern matching, we see patterns, but no two people see exactly the same patterns. It really is epic, what biological perceptual systems can do, and no von Neumann architecture machine will ever be able to do it fast enough, it takes massive, mind-boggling parallelism to do it. Such systems are grown, not made or designed.

Last edited Tue Mar 22, 2016, 12:19 PM - Edit history (1)

The recent debut of so-called "driverless trucks" is what I am referring to.

Sure, they could operate a tractor trailer on an open stretch of interstate, but until they are capable of detecting and properly reacting to the myriad variables I encounter on a routine basis, they won't be taking me from behind the wheel anytime soon.

There was a thread on DU recently about the Freightliner/Mercedes Benz prototype driverless heavy truck. Many of the posters claimed that this technology is right around the corner and it will be the death knell for truckers.

Let me see this truck detect and properly avoid the construction zone worker who steps into the narrow lane created by the cones I am driving next to. Let me see that software do it at night with the truck - in heavy traffic moving at the posted speed limit - give that worker every single inch of extra space available and do it in the often fraction of a seconds warning I get.

Software can take off, fly and land a passenger airplane. But it can also crash one just as easily.

Technology can do a LOT to help reduce accidents and correct mistakes drivers make (automatic breaking, lane following, lane departure warning, adaptive CC, "look ahead" sensors, back up cameras, etc), but there are just too many variables to eliminate people from the equation. Self-driving cars are "possible", but only practical when you upgrade the road infrastructure to support such a system (embed sensors in the road, provide two-way communication with all vehicles about traffic/road/weather conditions, etc).

I see automation and "AI" being used to augment, not replace humans.

Plus, the legal system hasn't caught up with the Internet age yet, never mind the age of driverless vehicles.

My firm had several trucks installed with such a device, and all our trucks have camera systems that have triggers (motion sensors, primarily) that log "events" such as hard, heavy braking, very rough patches of road, etc. Some are more sensitive than others. Some of our units will log when you simply cross railroad tracks. Others seem to need to be pushed off a cliff before they will log the event.

Anyway....about these automatic braking systems;


We have had incidents where the truck got rear ended because the device applied the brakes in heavy traffic. They use a device that detects vehicles in front and are capable of determining how fast that vehicle is moving, how far you are from it and your closing or opening speed. The problem is, in rush hour traffic for instance, I (and every other competent driver) puts a space between himself and the car in front. If you see a space in front of a heavy truck, that driver put that space there. Cars will very often change lanes in front of me, filling that space. As long as that car carries on with the flow of traffic, it is a simple thing for me to slightly lift and put my buffer back in. The device however, sees that incident as an imminent hazard and will apply the brakes, with very little to ZERO finesse.

Also, we have those "lane departure" warning devices on many of our trucks. The drivers HATE THEM WITH A PASSION because for the most part, they tell you what you already know. Not only do they detect when you are covering the lane line, they warn of a vehicle ahead. OK...fine....it's telling me what I already know! Sure, it goes off because a car changed lanes in front of me, but I bloody well let him in! Not to mention they go off constantly in traffic and if you have a co-driver in the bunk, it keeps him awake.

If they were tied to a sleep detector (they exist also, a device that monitors the eyes of the driver for fatigue) and went off ONLY if the driver was tired and that's why he strayed across a line or was too close to the vehicle in front, that would be different, but they aren't. And as such, they're a pain in the ass.

and that is "unforeseen consequences". Automatic braking is great tech, but not without problems when you turn it loose on a road where 99.9% of the other vehicles DON'T have it. And you just put your finger on the trouble with lane departure sensors.

I have heard of testing on system to monitor for sleepy drivers, but that is still a few years away at best, but probably will come with its own problems when deployed.

I am not so much interested in cars which drive themselves, but ones which talk to each other and track each other. When I approach a green light I want to know about the car heading perpendicular to me that hasn't slowed down and may blow through the light into my side. I like to know about the car stopped beyond the crest of a hill, or around a curve where I can't see it.

Augmentation seems more profitable than automation.

All kinds of jobs mostly try to suppress any of the really useful talents humans have. Our characteristics are a liability, not a benefit.

I believe I used the word "evolution" and "marvel".

As to jobs stifling creativity, that is another discussion entirely.

Take GM as an example. Back in the day, GM made practically every part that went into a car. These days they source a lot of that work to sub-contractors. That is where all the jobs went, to where the labor was cheaper. Why pay union wages to a guy in my plant to build a starter motor, when I can buy them from a factory in Mexico cheaper and with less overhead?

Again, the kinds of robots everyone is fearful of are not practical for a variety of reasons, chiefly due to engineering/materials constraints as well as economic and legal ones.

Here is my reason why robots will not replace humans.

I heard, this week, they are trying to replace restaurant employees, of Carl Jr., with robots. Unless we are really stupid, this trend will not happen. The plus, for the owners, is the short sighted saving of money. I say short sighted because of the down side for businesses, and the public, of this catching on.

Let's think this through. People own businesses to make money. If they do not make money then they go out of business. Now imagine that robots catch on big time. McDonald's uses them, Walmart, etc.....What will happen? There will be a lot less jobs. When there is a lot less jobs then there is a lot less money. When people have a lot less money they are cutting back and not going out to eat or buy things at places like Walmart. The businesses do not make enough money and the stores close down.

Worse than this is the actual impact on our economy and our people. People lose their jobs and theirs houses, possibly their cars. There are more homeless which means less money going into our system. At that point there are also so many people riding our social net that it is stretched to far.

A business cannot survive without consumers. By robots taking over we either need to go to a socialist system where everyone is being clothed and fed or a barter system. It is hard to barter with robots and defeats the purpose of having a business to make money.

I do not know if they have thought through the economic collapse this would bring to our system.

which are also quite imposing. Though I would mention a few other point that would have to be overcome first.

Who will write the health regulations concerning robot food preparation? What about liability laws when your robot contaminates food, or serves spoiled food, or impales a drive thru customer who leans forward at just the wrong time?

How will a Dominos Pizza Bot survive a delivery to a frat house?

There's a hell of a lot machines already do far more accurately and efficiently than humans, and the qty of "hell of a lot" is far more likely to increase than decrease.

and again, one that works well when you have specific objects pre-defined.

As I discuss in other posts, there are many non-trivial obstacles to be overcome to develop "robots" the way people fear them, such as power, processing, mobility and software.

Once you solve enough of those problems to actually create a viable economic device, the next set of obstacles become the legal and regulatory system, then social resistance. These are formidable and time consuming.

we excel at others.

In addition, factories and warehouses can become heavily controlled environments for robots, many already are. A UPC scanner/camera on a robot would allow it to sort about 95 percent of consumer goods in a warehouse. Assembly lines are another area robots excel in, repetitious work, and the work area can be so well controlled, they don't even need eyes.

There are a lot of jobs that don't require flexibility, agility or humans being clever.

and a manufacturing floor. One can be VERY controlled, the other, not so much. At least that is my experience in a plant that makes lots of different things, as opposed to a single product, where there is much variation. Also, while we have automated quite a bit over two decades, the automation has led to our ability to handle greater amounts of work, and increasing our work force almost four fold.

People always underestimate technology!

we were promised a lot of things that have NOT materialized. Not too many people driving their Atomic Cadillacs around town, or zipping down to the spaceport with their jetpack, to catch the moon shuttle.

Lots of things work great on paper and in the lab, yet never work out because they aren't practical.

Also, in 1905 man had already be flying for over a century. In fact, the LZ-1's maiden flight was in 1900: