The AI Revolution: Road to Superintelligence

[Dooz]

this is the most fascinating/important shit you're ever gonna not read, cretins

http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html#

[Torost]

read Nick Bostroms book "Superintelligence" when it came out.
It is somewhat poorly written, but he gets the message across.

Fascinating to wonder if when and how the singularity will take place.

Not in our lifetime is my bet.

[Kafein]

I have one problem with Singularity theory: intelligence cannot make a robot escape the laws of physics. With electronic computers I don't think we'll manage to have the power to reproduce even one complete human brain, which would be the condition for reaching super-human intelligence in the first place. The author doesn't understand Moore's Law, and then everything he says pretty much collapses from there.

Suppose we manage to create a computer with human intelligence capable of designing more intelligent computers. More intelligence isn't about programming but about hardware resources. Otherwise we'd be capable to do this on a small calculator except we don't. We can only concentrate so much computing power into a small space before wasting stupidly high amounts of energy into heat that needs to be evacuated, as shown with any supercomputer in the world. And that's only the beginning. Super-computers are not (and could not be) designed to be an efficient implementation of neural nets (which is the structure of the human brain and any human-like AI program). That effectively means the term "cps" in this context is a meaningless construct, as any traditional computer would be doing a lot of overhead for each "neuron operation". On top of that, exponential intelligence requires resources growing exponentially at an actually faster rate. Why? Because as I said, you can't concentrate infinite computing power in the same place, eventually you have to build larger and larger computers in order to keep the heat manageable. The fact that your computer gets bigger means the information going from one end to the other end will take more time to arrive. Signals move at around c/2 in copper wires. That's fast, but it's a serious limitation significantly impacting the design of all modern computers nonetheless.

Note though that I can totally imagine it happening with devices specifically designed to be actual networks, not multicore processors. That brings a host of other problems though, and we're technically nowhere near it.

[Corwin]

waitbutwhy is one of my favorite places on the Internet. Awesome blog, and superinteligence article is one of the most fascinating articles. Just want to say you should READ it.

[Dooz]

Do any of the points of skepticism take into account the accelerated rate of the growth of knowledge and that whole thing about not even being able to fathom what's to come, much like people in past centuries couldn't have fathomed the things of today? If everything being used to cast doubt on the possibility of this thing happening is strictly within the paradigm of human knowledge where it currently stands, it doesn't really hold up to the inevitable changes in what is considered possible constantly shifting at an ever increasing rate.

Just because we can't imagine something happening based on the current limits of knowledge, doesn't mean it can't happen. Those limits are constantly expanding and the world we think is flat will soon become round.

[Xant]

Nor does it mean it can happen.

[Tibe]

Yeah. The "when someone, somewhere will surely shake these borders" argument is not really that solid. The optimisim is nice and all but it doesnt really say anything and is useless at current stage. He's right to be skeptical, cause at this point its just a theoretic possibility based on a theoretic foundation. I predict that when Cutchulu rises, the speed of winds will increase worldwide so much that windenergy will be viable everywhere and shall replace nuclearenergy.

[Xant]

It reminds me of Pascal's wager.

[Leshma]

I have one problem with Singularity theory: intelligence cannot make a robot escape the laws of physics. With electronic computers I don't think we'll manage to have the power to reproduce even one complete human brain, which would be the condition for reaching super-human intelligence in the first place. The author doesn't understand Moore's Law, and then everything he says pretty much collapses from there.

Moore Law, electronic computers? What is this, 1980s? Moore Law today serves just one purpose and that is marketing for Intel CPUs. It isn't a thing for a decade already when it comes to consumer electronics.

Quantum computers already exist and can be bought if you have enough cash, software is an issue atm but that will be fixed in years to come. In less than a decade price of those computers will come down to under 100k and their availability will be wider. IBM and HP are already working on consumer versions of quantum computers and those should be revealed in few years to general public.

Electronic computers have hit their limits and no matter what Intel marketing department is spinning these days (we have found new material that will allow us to go bellow 10nm litography, bullshit like that), electronic computers are very near their limits. Next are quantum and bio computers. We'll probably see upgrade to electro-computers such as optic connections between chips but that will happen alongside development of next gen computing which will be nothing like what we have now.

[Dooz]

Yeah. The "when someone, somewhere will surely shake these borders" argument is not really that solid. The optimisim is nice and all but it doesnt really say anything and is useless at current stage. He's right to be skeptical, cause at this point its just a theoretic possibility based on a theoretic foundation. I predict that when Cutchulu rises, the speed of winds will increase worldwide so much that windenergy will be viable everywhere and shall replace nuclearenergy.

difference being, there's reason to think things are moving in a certain direction based on where things have come from and currently are, as opposed to completely random "theory" about cthulu

[Xant]

No, there isn't.

[Tibe]

Ctchulu has the power to control weather and generally favors cloudy and windy weather to strike fear into the hearts of mortals.

Your own theory is random, you nonbeliever.....

[Kafein]

Moore Law, electronic computers? What is this, 1980s? Moore Law today serves just one purpose and that is marketing for Intel CPUs. It isn't a thing for a decade already when it comes to consumer electronics.

That's pretty much my point, yes. Although the transistor density does still double every 18 months (because Intel is following that line), the frequency of single cores has stopped going up in the mid 00s.

Quantum computers already exist and can be bought if you have enough cash, software is an issue atm but that will be fixed in years to come. In less than a decade price of those computers will come down to under 100k and their availability will be wider. IBM and HP are already working on consumer versions of quantum computers and those should be revealed in few years to general public.

I haven't heard anything of the sort. It's not even clear whether D-Wave's computers are quantum or not, and they won't be used for practical applications until their memory size increases by huge amounts. Software is actually not a problem by the way. We already have "algorithms" for quantum computers.

Do any of the points of skepticism take into account the accelerated rate of the growth of knowledge and that whole thing about not even being able to fathom what's to come, much like people in past centuries couldn't have fathomed the things of today? If everything being used to cast doubt on the possibility of this thing happening is strictly within the paradigm of human knowledge where it currently stands, it doesn't really hold up to the inevitable changes in what is considered possible constantly shifting at an ever increasing rate.

Just because we can't imagine something happening based on the current limits of knowledge, doesn't mean it can't happen. Those limits are constantly expanding and the world we think is flat will soon become round.

This is a misinformed view. Logic is not a social construct, and not everything in science is a changing paradigm. And when it comes to computer science and electronics, most of it is provably set in stone. This isn't biology where discovering a new species in deep waters can change everything (and that is a poor image I should say). Case in point, the speed of electronic signals in cables of various materials is a critically important fact of life. The minimal amount of bits you need to transfer some message without loss is the informational entropy of the message. Pi is an absolute mathematical constant that transcends absolutely everything including any metaphysical belief you may have. The halting problem is undecidable. My criticism of the short term singularity stems from these kinds of things mostly, due to the practical problems involved in making computers that are more powerful. Hell, the first rule that many people seem to forget is that you *need* bigger computers to make machines that are more intelligent.

[Dooz]

Hell, the first rule that many people seem to forget is that you *need* bigger computers to make machines that are more intelligent.

Do people also forget that computers which could only do a minuscule fraction of what our phones can now used to be the size of a room? According to my misinformed view, I think your views are misinformed. But it doesn't matter. Whatever's gonna happen or not is gonna happen or not. And our views will have nothing to do with it.

[Leshma]

I haven't heard anything of the sort. It's not even clear whether D-Wave's computers are quantum or not, and they won't be used for practical applications until their memory size increases by huge amounts. Software is actually not a problem by the way. We already have "algorithms" for quantum computers.

General purpose quantum computer won't be a thing for quite some time. D-Wave is something akin to accelerator chip, needs conventional supercomputer to finish calculation, but speeds up solving optimization problems by big margin, making problem smaller (engrish). It doesn't give you result, but it tells you what isn't much faster than any super computer would. That is why big corporations are paying tens of millions for it. Second gen runs much cooler (1st is cooled at near absolute zero because it gets too hot, 2nd gen runs 40% cooler) and much more precise (main issue with q-computers is variability of results).

No one but D-Wave people can claim it is true q-comp because:
- it is not available to most researchers (it is very rare globally)
- most researchers don't understand how it works, it is very "black boxy"

HP put its future on next gen computing, it is do or die for them. IBM has always been the one to reveal the future, they pulled out of manufacturing of e-computers. They will be first to make this tech "available" (research labs on universities, 10-100k price point per machine).