Will it be a nice god? About super AI and stuff

[SilverSpook]

I've been thinking to program a simple brute force applet that fills 16x16 pixel canvas with all possible variations of all pixels of 54 colors of NES palette.

It's an interesting problem, but I would say that brute-forcing is itself not really intelligence, but application of the work of past intelligence.  When Google produces a translation from English to Japanese, it doesn't actually know Japanese, but rather aggregates billions of human intelligences translating billions of English sentences to Japanese, and finds statistical patterns in the translations. 

When it comes to art, the problem is knowing what "looks right" and what doesn't.  Google can process billions of images of faces and then look at an image and analyze the probability that there is a face there.  It will even detect the Mona Lisa, if shown it.  But this requires a human to tell the machine what to look for, which makes it not an artist but just a search function.  This is a version of Jorge Luis Borges' Library of Babel problem.  The Library of Babel contains all 410 page books possible, including books containing complete gibberish.  Somewhere in the library of babel is Moby Dick, The Bible, predictions of the future, undiscovered scientific theories.  But the vast, vast majority are utter nonsense.  Because of this, the entire library is utterly useless to its librarians, who become suicidal with despair. 

Put simply, having all of the books, or art, or 54-pixel sprites possible is useless without knowing which sprite is the right one.  And if you know which sprite is the right one, it's a waste of time to search through all possible sprites. 

[selmiak]

I thought this was an interesting look at some of the impressive things that can be done with neural networks, while also showing some of its limitations: http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html

(Keeping in mind, of course, that in each of these examples the algorithms and settings have been tweaked by humans to give the best results, and the pictures selected as the best illustrations.)

Someone ran Fear and Loathing through the videoversion of deepdream
youtubelink

another cool fractallike video

[InCreator]

Put simply, having all of the books, or art, or 54-pixel sprites possible is useless without knowing which sprite is the right one.  And if you know which sprite is the right one, it's a waste of time to search through all possible sprites. 

Isn't "right" defined by us and generally still by some common arguments? We all agree that snot and dirt is gross, three act story is exiting and one of the best ways to make fiction, brown and green do not go together well, and so on? Music is incredibly formulaic nowadays, I see zero reason why machines couldn't make trance or dubstep and tweak it to the perfection (because science of pleasurable melodies, chords and scales also exists and it's pretty universal. There was even a video few years ago where tons of pop songs are shown to have same melody). Also, with a learning machine, you can just let machine sample insane amounts of media AND it's ratings by people and draw conclusions. Think ratings on goodreads.com, imdb, amazon, beatport, imgur or even facebook likes.

One thing internet did, was letting us very precisely rate and review everything out there and we do it really, really much, basically laying guidelines for AI to learn from.

[Crimson Wizard]

Basically, the main condition to let machines create good human art is to formulate how humans create good art.

Then again, we could make a machine that uses some other formulae to define good art, and it will produce unhuman art then.

[InCreator]

Actually my thoughts are outdated.

Everyone loves Minecraft, a game with procedural worlds. All that's human there is algorithm that creates random world, but I see zero reason why good AI couldn't use even more advanced mathematics, coupled with wide (sampled) knowledge of nature to create much more immersive worlds. And amount of procedural games out there has been growing all the time. Isn't this machine art for human consumption?

[Snarky]

Back to throw some more cold water on this idea, a very interesting article rant about the current state of brain modeling: http://mathbabe.org/2015/10/20/guest-post-dirty-rant-about-the-human-brain-project/

[Stromvin]

Ok first of all, im deeply sorry, because neither did i have the time to read all the comments, nor the Patience to read the whole article... since... i didnt sound very convincing to me after the first 30 or something sentences.

The author sounds to much like a child that had seen transfomers for the first time. Im sorry i dont want to Insult anyone.
BUT i saw Little to no real proof or eveidence for the therory that we are at this Point where the curve gets so much steeper.
No matter where he read or heard this, i think it misses a very important Point. Remember the good old days of 1800? Me neither but what i know is that you got a degree in any Kind of science, even a doctor title in your early Twens as far as i remember
Today most People are in their late 20s to early 30s. Ist true, tht we have better info tech, and more People than ever can do Research, but two things seem to rarely be considered: 1. The stuff you Need to know and learn, in order to advance grows every year. Thus the time needed grows with it 2. The number of researches is high, but it has a natural Limit... So expecting it will increase the same rate it did during the last century is unrealistic.

Now one or two senctences to the Topic: Understanding the brain? really? We understand basically the function of Neurons, yes. We know what part of the brain we Need to cut off to impair a certain function, yes. But to say we understand the brain, let alone the higher functions, like consciousness is far from the truth. At least last time i checked. Unfortunately i found brain bio kinda creepyand didnt like taking Brains outof Baby mice, so i switched to algaea

But heres the Thing. When i was thinkin about the brain and ist nature i came up wiht a rather phylosophical idea maybe.
How do you put a box insinde a boxe of the exact same size without damaging it?
Because remember, ist the brain we are talking about. OUR very organ of understanding. Coputers might at some Point help yes, but Computers are also products of human Brains.
So, as far as i understand it. We have Kind of a chicken egg Problem.
To make a Computer AI that ca surpass us, we Need to make it more capable than our brain. To do that we Need to fully understand our brain. To fully understand our brain, we needa supercomputer more intelligent than us.

Maybe we can find a backdoor, quite likely we will. But as far as i see it, not with curren Technology and not in the next 50 years.

greez Stromvin
PS: Maybe im just completely wrong, im open for discussion.

[MiteWiseacreLives!]

The author sounds to much like a child that had seen transfomers for the first time. Im sorry i dont want to Insult anyone

Are you referring to the author of the article posted by Snarky? This doesn't make much sense, she seemed to be sticking to the facts of where research is at this time (and it seems you agree with her ) To try and calculate when a breakthrough will happen, based on history of technological advancement, is the same process that told us we should have flying cars and personal androids by now. Building an AI system that's pretty darn good with ones and zeros is a lot more possible than simulating the human/anything brain. As far as I know, not a scientist here obviously, we are not able to even replicate/build a single cell, a little under gunned to take on the brain. But, who am I to say you can't dream or do the impossible even.

[Snarky]

Are you referring to the author of the article posted by Snarky?

I think Stromvin has jumped back to the first post in the thread, and is talking about the article linked there.

Ok first of all, im deeply sorry, because neither did i have the time to read all the comments, nor the Patience to read the whole article... since... i didnt sound very convincing to me after the first 30 or something sentences.

Hi Stromvin. Maybe you should read the thread; a number of us have raised similar objections.

But heres the Thing. When i was thinkin about the brain and ist nature i came up wiht a rather phylosophical idea maybe.
How do you put a box insinde a boxe of the exact same size without damaging it?
Because remember, ist the brain we are talking about. OUR very organ of understanding. Coputers might at some Point help yes, but Computers are also products of human Brains.
So, as far as i understand it. We have Kind of a chicken egg Problem.
To make a Computer AI that ca surpass us, we Need to make it more capable than our brain. To do that we Need to fully understand our brain. To fully understand our brain, we needa supercomputer more intelligent than us.

This is an old argument, usually stated as “If our brains were simple enough for us to understand them, we'd be so simple that we couldn't.”

There are several counterarguments to that:

1. Because of repetition, a lot of the complexity of the brain can almost certainly be simplified. (You don't need to understand how every neuron works individually; it's enough to understand one, or a few representative examples, and then generalize. The same applies to other levels of complexity.)
2. You could similarly argue that since it's near to impossible to memorize thousands of pages accurately, no author could write a (coherent) book of that length. However, writing (and other kinds of memory storage) allow us to externalize and "fix" knowledge so that we don't have to keep all the details in our brains at once.
3. It's not necessary for any one person to understand the brain completely, if you have a team of experts working together, each understanding one component.

You could also object that it's not necessarily true that we need to fully understand the brain in order to build a computer as intelligent, or more so, than ourselves.

[Stromvin]

Hithere
Sorry again if i irritated anyone. I was indeed referring to the author of the article linked in the first post. And yeah sorry i didnt have the time to read all the text, maybe i will once i have it So ist kinda likely and expectable that my ideas have already been mentioned. Btw. Snarkys last link is indeed very interesting ad its about what i heard last time i had a lecture about this, plus some new interesting things.
I am not saying we/mankind will never achieve this! But i am getting a Little angry if i see or hear People talking about this Topic as if they were so Close to actually doing it, while we are still decades away. So as a discussion Topic its very interesting, but indeed expecting this in the nex 10 years is ridiculous in my opinion. And every Person that is stating something like this on tv or whatever seems highly dubious to me.
Aside from that, the idea of living in a pc for all eternity is pretty pseudo religious.

One last thought, because its getting late. A friend of mine(former physics Student, now physicist() )
Once tried to explain to me why we Need Quantum Computers for real randomization. Yeah, i didnt really get it but i only had to check if the talk ws suitable
And as far as i understand, with something like this we actually would be a whole step closer to creation of true AI.
Im off,
greez Stromvin

[Billbis]

I found this article particularly enlightening:
http://timdettmers.com/2015/07/27/brain-vs-deep-learning-singularity/
It is rather a long-read, but it worth it in my opinion.
It explains, from a biological point of view, why we are nowhere near to completely stimulate a human brain.