As productivity increases, unemployment increases, purchasing power falls. The current capitalist system is replaced by a different socioeconomic system coinciding with abundant energy, from solar panels planted all over Africa and nuclear fusion, 3d-printing, stem-cell meat production and abundant water from Dean Kamen's ideas. Instead of having brand name clothes shipped across thousands of miles of ocean from sweatshops, they will be printed at home.
The wealthy are afraid.
http://ynaija.com/johann-rupert-worr...poor-uprising/

I dream of this utopia but you underestimate what people will do to cling to power and money as well as the ability to stir people to a revolt (even if it is just a social revolt).

People who own the machines will fine for a while. But people who own businesses cling to power by selling stuff. So it will be a systemic problem for everyone with no buyers, and a guaranteed income is the inevitable solution if we want to remain a market economy, it seems.

You mean like when we invented the grist mill?

People will stay in school longer, retire earlier and pay themselves more and more per hour while working less and less hours and doing less and less wealth generating work during those hours. They will invent things to do during those hours like tick boxes, have meetings and browse the internet.

Eventually this will become so obvious that someone will notice.

See the evolution from spade, to bulldozer, to a self-digging AI. We would still need people to point and click where to dig.

We won't build an AI that will decide on its own where to dig, without any supervision. No one will respect Eddy the Motorized Mall Cop, but such a swarm of robots may help a few security guards to keep an area safe.

Once you are talking about a fully automated world AI economy when we are decades/centuries away, you can not project today's problems onto that future. There would be no bloodbath if we can perfectly predict and beat down any brewing revolution or citizen dissent. There would be no violence when genetic engineering removes this biological urge from our newborn. There would be no economic wars with an automated world economy and diplomatic channels -- unless we build in such adversarial optimizers.

My prediction: AI will be able to emulate/combine a human "soul" and artwork to where you wouldn't know the difference.

But rest assured that there is currently not enough shared factual information on the internet to create an information intelligence that would be smarter than any human expert to have ever lived.

Further prediction: Many random worlds theory would suggest that the possibility of AGI becoming a reality, would mean that it likely already became a reality once. Then "we" build our own simulation and we could be on the brink of finding out how to (re)program our own universe.

Also "People are going to be sitting around w/ nothing to do" will probably not be true. You could emulate hundreds of worlds in increasing detail. You could play Minecraft all day. Hell, the rich kid on the block can probably emulate our entire universe and gets to play the president for a while. Hmmmm.

Stunting with loss functions and "paperclip duplicator" optimization is the major fear. I think those are infancy problems, like we had with nuclear energy and dying scientists. Let's hope we grow beyond that before the merging fully begins. Or we could be creating some horrible horrible worlds where loss of jobs is the least of our concern.

and "they" will stfu before "they" ruin it for the rest of us.

One thing we will have on the AI's is personal experience with the human condition. We will continue exploring and trying to illuminate that experience through entertainment and the arts. Anyone who wants to will be able to turn their life into a Truman type unscripted 24/7 reality show viewed on one of millions of virtual reality channels. The most popular entertainers and artists will still get rich as people spend their stipends on human entertainment.


PairTheBoard

non of this will happen as:

a) there is not enough energy
b) there is to much ppl on Earth
c) if there was enough energy for everyone to have western standard we would kill ourselves due to rising temperature (rising trend already happening) - folks with power would prevent this as this will mean the human end

the end game will be a split into high end and low end (already happening). High end will have AI and all the cool medical stuff. The low end (99%+) will have no real upward mobility opportunity do to planet energy constraints + AI squeezing them out of middle class. High end folks will thus result to keep social peace by keeping the emotional part of low end humans "happy" aka legalize drugs, keep <1% segregated using "out of sight, out of mind" through segregated communities (in the sense this is a stable Nash equlibrium so the high end folks don't need to "make conspiracy" against the rest as movies tend to make it). Low end will be used as labor payed just enough to keep their living cost covered; payed for low end jobs AI isn't profitable to take over. They will never be able to scale as there won't be enough energy...(no real energy transfer from high end to low end)

Middle class doesn't exist (no conspiracy there as Nas eq. doesn't really allow it.). AI will mostly squeeze it out of existence, and leave very few of them that will be payed slightly above comfortable wages to "over-watch" AI and consult.

There will be also a very, very small subset of society arbitraging the low cost of low end/ higher income of high end but will be constrained do to scalability issues. (High end won't target those opportunities as they are to small to scale up. )

India Legalizes Child Labor Amid Skyrocketing Rates, Activists Fight Back

I can go on and on...

The end game is not limited by Earth.

The points you make will accelerate human expansion if it hasn't already passed that stage.

But oddly, extremely powerful forms of AI may not require much more than this sort of thinking.



(Source.). So with deep learning nets that have a (# of learnable parameters) ~ (those present in a bee's brain), we are approaching human levels of performance on image and speech recognition; and already have super-human performance on narrow image domains like street numbers.

Everyone knows about Moore's Law, but exponential digitization is just as important. Current rule of thumb is that with a training set of ~10M examples you can get a deep net that matches human performance on classification tasks.

Also interesting are recent results on so-called neural Turing machines that have inferred e.g. sorting algorithms from training sets. Give this sort of thing 40 years and...who knows?

yes, you are right

but this is very, very likely only for top 1%, as the cost of energy and stable environment creation is very huge in space

most things we take for granted (environmental stability/energy input/stable energy transfer environment) are very hard to replicate. Even a hurricane environment (where most of us would die of if it wasn't a lengthy period of stability that make us prepared for it) is trivial in comparison to space...

because of that it is not cost efficient to scale this for the "rest" of the population. And it is unnecessary. There is no need to send the low end of population to space as it energy cost is higher then planet Earth.

On the other hand it could play out that due to new pricing of "environmental stability" Earth becomes a prime "housing" location and the low end does get priced out into space (or Australia).

But you can't forgo the Earth energy constrain in any case and AI pricing out the middle class in any enviroment.

rikers -

I think you're framing things strangely, since capturing ~1% of available solar would meet the energy needs of the planet...but I basically agree with your conclusions under the assumption that the future ruling class will be similar to the iron-fisted aristocrats of history.

However, decent chance that the modern ruling class will be different imo, see e.g. Larry Page as he introduces a video vignette on Internet search empowering an African farmer to find a simple solution to a pest threatening his crops...

I have beaten the estimated human performance on a similar data set with deep nets. I am a bit two-pronged about this accomplishment.

On the one hand it is a great feat: No domain knowledge, no expert rules, no months of hand-engineering features. Outperforming humans on the same task.

But on the other hand, this has already happened with chess (and distinguishing a cat from a dog seems way simpler than beating a SuperGM with the black pieces). Both are basically very fast pattern matchers. A chess machine will fail or play on like nothing happened, when a human opponent accidentally misplaces a piece. It sorely lacks creativity and psychological play. A deep net will happily see faces in a cloud, and it lacks very basic common sense about semantics, context, scenes and cultural elements.

Like Searle's Chinese Room: We now have a robot that is very accurate and fast in translating Chinese symbols to English symbols. But we are nowhere near a robot showing a fundamental understanding of both languages. Or comparing an airplane to a bird. Sure an airplane flies faster than a bird, but does it really beat the bird? Only in function. Turing would say that is enough for machine intelligence. Romanticists would disagree.

Those Neural Turing Machines are awesome though and there is no telling what will happen when we go from machine learning to machine reasoning.

'...allow the whole system to run for an appreciable period, and then break in as a kind of "inspector of schools" and see what progress had been made...' Turing on his unorganized B-type machines.

As for your prediction, I myself like the more information theoretic approach to this. Moore is too much about hardware, while machine intelligence needs new software approaches (many exciting new algorithms to invent for us this century) and learning fuel: Information. We'd need to bridge the gap between supervised learning (10 million training samples) and unsupervised learning (like how our human children are able to learn and infer without being told the label for everything.).

Information can have both negative and positive value, very much like energy.

Also you can consider a future where an AI pacifies humanity with superior intelligence, communication, distribution, and logic. Keeping the 1% status quo can not be the smartest thing for us humans to do. Unless we really are a big ant hill, and we need our queens and worker bees to function optimally according to our biological assets.

I think all our energy comes from the sun. There is energy a plenty (and a lot of it goes to waste right now, dissipated as heat). This view of complex dynamic systems passing around energy is very interesting. You may be interested in the works of David Wolpert.

https://www.youtube.com/watch?v=9_TMMKeNxO0
The Landauer limit and thermodynamics of biological computation

Robots building houses.
http://www.engadget.com/2015/06/25/r...share_facebook

Deep learning machine beating humans in IQ test.
http://www.technologyreview.com/view...ns-in-iq-test/

Hear it is (once again).... Earth energy constraint

to sum up: Energy usage is scarce since Earth cooling process if finite. And since all energy transformation produces heat there is an upper limit on how much energy we can use.

Exponential Economist Meets Finite Physicist

once you factor that there is an upper limit to consumption/population member (from link above) AND add AI process of automation of middle class white collar labor AND increasing Earth centralization (through globalization process) = the end result is wealth distribution

Most all of my nice things have nearly zero ongoing energy costs.

Plus, as others have mentioned, that giant yellow thing in the sky provides approximately one ****-ton of energy per day.

I think I can rest easily knowing that we can grow at 3% compounded for the next few hundred years.

Rikers is an excellent example of conformation bias, on multiple levels. Aside from misunderstanding fundamental physics and scientific methodology. And despite posting here for years he still hammers on about the same misconceptions. Fully expected. Many do this. Another example is the flamboyant and gibberish prone posters that rattle on about the second law of thermodynamics. Which usually shows they have little to no understanding of this law.

The great majority of energy is derived directly and indirectly from the sun. Earth however still has residual internal energy (both from heat and radioactive decay) from initial formation. Though much has dissipated especially during early geologic time there is still a great deal remaining.

http://www.geology.sdsu.edu/how_volc...work/Heat.html

https://en.wikipedia.org/wiki/Earth's_internal_heat_budget

ad hominem...ducy?

Worth reading (along with related links):

Cognitive Biases

Rikers, do you think AI could help develop our technology so that we can have a breakthrough in energy production that would maintain growth and prevent only the top of the food chain receiving all the benefit? I'm thinking fusion, more efficient solar and so on. And to the problem of heat, couldn't that heat somehow be harnessed (or not wasted through inefficiencies) and turned into useable energy? Isn't this how we do it now, heating water to propel turbines to create electricity?

Yes, you are such a good bad example. Never stop posting. Thanks.

yes, AI neural net domination (singularity) is the end game, but obviously a lot of problems will be solved on the path to there

but it will not be pretty for 99%

• The first thing it is used is to genetically map/remap all human genomes/bodily processes and to enhance life quality/duration (priced only for 1%) -in progress
  
• Second thing will be physical enhancements
  
• Third will be environmental stability (this includes power consumption)
  
• Forth will be total planet Earth stability - somewhere here the 99% will find out they are the cause of unstability

in the process it will price out a lot of middle class as human white color labor will become obsolete, and most of them can't shift to scientific research.

errr...no (heat cannot be completely converted into work or other forms of energy because of the entropy heat contains and you always end out with more heat at the Earth/space boundary)

ATM (today) the population is to big to sustain current western standard for all on planet Earth (not future standard). Even if you had all the energy available the temperature rise would kill us. There is no paradise of equality...

Humanity doesn't need 6 billion scientists/researchers and Earth can't sustain it for that matter. 1 big neural net will do. Rest need not apply.

note: 1%, 99% used arbitrary

You're reminding me why I dropped out of engineering school. Welp, you've identified the problems, and it will be fun* to watch them solved!

*or excruciatingly painful