There are about several thousand active posters on /sci/, with

get a gurl prego

What are restricted Boltzmann machines

>>8753578
Two questions:
1. Is the AI allowed to count?
2. Is defining the addition operator considered "providing it with a pattern"?

Define "pattern".

Use an EA that has its fitness bound to how well the end set is sorted. Pretty easy desu. Unless the guidelines for the fitness count as a pattern. Clarification would help.

>>8753638
1. A pre-defined rule of having it somehow evaluate higher values, or "count", yes, but not with a directly programmed number comparison as that's obvious interference from your side
2. Yes

>>8753647
The set {1, 2, 3} - a pattern with an addition of +1

>>8753666
I would venture that it is impossible to devise an unsupervised learning AI

If you're not allowed to "teach" it addition, there is no way for it to organize numbers into a continuum of ascending or descending values. This also would prevent it from checking if two variables were equal.

I could elaborate further.

>>8753676
And yet, humans having unsupervised intelligence is a proof that such things do exist.

>If you're not allowed to "teach" it addition, there is no way for it to organize numbers into a continuum of ascending or descending values.
Maybe that's the problem to begin with - it's supposed to somehow realize that they have to be organized and organize them itself without our interference, maybe something towards this >>8753636 direction?

>This also would prevent it from checking if two variables were equal.
Our universe has differing quantities, the sums of which are measured by gravity and (considering we're not a simulation) without any predefined rule, so there should exist a way

>>8753578
>>Come up with a way to make an AI discover patterns without providing any kind of a pattern beforehand (unsupervised learning)
can you prove that humans are not provided with patterns beforehand from birth?

>>8753704
Those patterns should've been formed at some point in our evolution, and the method of formation was obviously unsupervised.

>>8753715
I guess I can grant that
I got nothing

>>8753715
You could argue that natural selection provided the training, i.e, whether various "ingrained" perceptions or insticts reflected the real world accurately enough to keep the organism surviving, which would require basic perceptions of bigger than / less than / etc.

>>8753578
Not possible. You need to define the pattern before and make it search for that. All methods of unsupervised learning require a more or less specific set of assumptions.

>>8753694
humans do not have unsupervised intelligence
everyone was taught by others
those precious few skills that the first hominids possessed were not taught by other hominids, but were innate; more advanced skills must be taught

the objects in our universe do not actually exist as collectives. They may be the "sum" of many parts but it is these individual parts that do all the actions. Each atom in the Earth pulls you slightly. So does each atom in the Sun. The universe doesn't measure your potential energy relative to some reference point, and then give directions to your constituent particles and tell them where to go. You're an incredibly complex balancing act. Everything is.

So I disagree that there's some way to construct an AI without supervision. Human intelligence is the result of 14+ billion years of "supervision".

>>8753724
>>8753726
>>8753731
Sure, we can argue that the Universe itself is providing positive reinforcements towards our supposed "pre-defined pattern" of survival and maybe entropy reduction. But how do we apply it to a virtual environment then?

>>8753738
Here's an idea

Instead of building a computer and running software on it, throw a suitable amount of hydrogen into a suitably large room and then wait a suitably long time for it to develop intelligence.

>>8753731
His argument is that if humans learned from one another, then AIs would also be able to learn for themselves or from one another.

To OP:

The answer is that we created what a pattern is. If an AI were to define what a pattern is, it will either line up with our definition or it won't. If it were to, it would have done so through knowledge gathered through us. If it were not to, we wouldn't be able to identify it as a pattern even though the AI would.

>>8753750
or just reproduce with a female lol

nerd

>>8753578
Do what nature did. Start with little to nothing and brute-force it for a gorrilion years.

>>8753770
See: >>8753738

In order to go with that, we should first define what the expected goal of the Universe is and what are we getting "rewarded" with for being good boys. The only case of positive reinforcement in our Universe is gravity, where the more it "consumes" the more reinforced it is to consume more, ultimately leading to a Black Hole, which again has a sole purpose of devouring as much as possible.
If gravity is the positive reinforcement then what exactly is its functionality, and how can we replicate that in any beneficial way in a virtual environment?

But on the other hand, there's also negative reinforcement - entropy - and the more an object degrades the less likely it is to stop degrading into lower structures. The positive reinforcement then becomes fighting the negative attrition - fighting entropy. If that is the case, how can we replicate that so we can create an environment where growth of unsupervised (by us) intelligence is promoted?

>>8753793
>The universe is only gravity and entropy

>>8753806
Matter and energy, both of which fall in the "accumulate more so you accumulate more" category of gravity
Well-ordered structure of matter which uses its order to keep itself ordered for longer times (hint: life), which is under the entropy category

Literally what else can possibly exist?

>>8753578
>monte carlo:

>random state configurations
>search for states which give the same outcome for each set of identical inputs
>it has learned a pattern: it knows that this input will give the same output for every time it is processed.

If a set of inputs gives a different answer for each identical input, then it is not a pattern, unless it is an output cycle, which may or may not be infinite.

>>8753820
I don't see how gravity and entropy are supposedly the two major aspects of our universe. Sounds like you are just putting buzzwords together to fit in your theory. Gravity is a fundamental force, entropy is a state variable. I don't see how those are supposed to fit together in a single context without including other shit like pressure, the electromagnetic force or whatever else.

What is even a pattern, that's so vague.
Literally anything could be a pattern, if you pump enough information into the system. Just look at the conspiracy theories that seem to hold water with all the "evidence".

Just let the program generate patterns

>write an AI that can learn without instruction

so.... you want something to do complex data analysis while leaving it with the IQ of an ant

great idea senpai

>>8753839
Yeah its pretty dumb.
There is only "thing" and "no-thing" in a duality.
Thing derives all chaos, no-thing represents total order.
The struggle between existence and void is the root of the quantum behaviour, the struggle between position and momentum,
between energy and time.

>>8753877
Dude, that's just aesthetic, pseudo-philosophical crap. Don't respond, I'm not interested.

I could post my own AI ideas and progress on an internet forum for the fuck of it just because some fag asked me to, or I could continue to work on them myself and swim in cash for the rest of my life if I end up succeeding
hmmm....

>>8753878
yeah it is ying yang doink. superposition. I wouldn't expect you to try to comprehend the singularity event, so go eat a bag of dicks.

>>8753878

>what is duality?
>what is polarity?
>what is chirality?
>what is the uncertainty principle(lol)?
>what is quantum vaccum

Clearly /sci/ needs to get out its own ass and past freshman year.

>>8753578
It'd be cool to organize research teams consisting of /sci/ autists for various subjects. They could just be simple but useful problems too. Or at least a /sci/ study group might be fun

>>8754063
This

If the combined autism of /pol/ can get blompf in power then everything is possible

>>8753578
>>Come up with a way to make an AI discover patterns without providing any kind of a pattern beforehand (unsupervised learning)

Prove to me that humans can do this first.

what about creating a computer that starts with the most basic of functions, and program it to "build" or improve on its own software?.

>inb4 get the fuck off of /sci/ fag

I'm not a damn genius level intellect, but I'm allowed to express simple ideas, so don't be a fucking dick if it's a stupid idea.

>>8754117
No, you're not stupid, that's exactly how it's supposed to be done.

>>8754137 oh.

>>8754117
This is a fine idea, but making an effecient implementation is the difficulty.

>>8754227 is it feasible to slowly give it information to add over time? how about when starting, implement a simple sorting program, that way the computer will gradually develop a routine and learn?

>>8754257
That's the general idea, allocate disordered resources and let the program order them. The problem is, how? How do you achieve that without human interference? And what exactly is it even supposed to do or output?

>>8754267 I think the answer might lie in transferring information through energy, since that's the primary ingredient in creating life. like, connect the computer to a source of electricity with the encoded information and "jolt" it per se (that probably sounded way more autistic than intended to be but I'm not very good at expressing thought)

>>8753666
You provided an example of a pattern.
Not the definition.

Give us a clear definition of what you mean and perhaps we'll make an AI that spots patterns.

>>8753578
Sorry SENPAI, we are all smart but lazy.

>>8753877
this nibba gets it

tho you sound fucking pretentious and annoying as fuck, you won't really convince anyone who wasn't already believing that.

Also why do you think its the root of quantum behavior? Whats the mechanism?

We can already do this.

Take, for example, the concept of addition. Say that I wish to create an intelligence capable of discovering the 'pattern of addition.' Take a feed-forward neural network whose inputs are pairs of integers in a range and whose outputs are integers in twice that range. Create a 'virtual environment' which 'selects' for intelligence capable of adding by generating outputs, observing the accuracy of networks in the current generation, and `reproducing' (by averaging network parameters) accurate networks [with mutations] until we obtain a satisfactory network (this is known as a genetic algorithm).

>>8753877
“What we imagine is order is merely the prevailing form of chaos.”

http://www.thedailyliberator.com/liberty-discord/
>The Aneristic Principle is that of APPARENT ORDER; the Eristic Principle is that of APPARENT DISORDER. Both order and disorder are man made concepts and are artificial divisions of PURE CHAOS, which is a level deeper that is the level of distinction making. With our concept making apparatus called “mind” we look at reality through the ideas-about-reality which our cultures give us. The ideas-about-reality are mistakenly labeled “reality” and unenlightened people are forever perplexed by the fact that other people, especially other cultures, see “reality” differently. It is only the ideas-about-reality which differ. Real (capital-T True) reality is a level deeper that is the level of concept. We look at the world through windows on which have been drawn grids (concepts). Different philosophies use different grids.

>>8754267
I'll tell you how I do it, I input as much shit as I can into my mainframe to see what sticks, when I detect repeat info, I store more of it, order through chaos.

Also, why has gravity never been demonstrated in a smaller scale, why not create a massive lead sphere and try to get a dust to stick to it, never been done, never will because (((G)))ravity is a spook

>>8753578
I'm just here for the memes, dude

>>8753578
Construct a random number generator. Use the random number generator to create billions of collections of data which could be interpreted as programs. Assign billions of mathematics problems to the billions of programs. Eliminate 90% of the programs, keeping those whose responses are closest to the correct answers. Duplixate the remaining programs, but with some randomized errors. Continue testing.

If you can store a random number generator and a metric fuckton of memory inside your AI, then congratulations, you have invented a pattern finding AI with minimal preprogramming, just like nature is presumed to have done it.

>>8753731
>humans do not have unsupervised intelligence
>everyone was taught by others

In fact purely unsupervised learning from infant onwards causes irreparable brain damage.

>>8754049
>Sometimes there are pairs of things, which means everything must be

Is unsupervised learning the same as fluid intelligence in humans?

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

>>8753578
Use the pattern provided as a training set for a given ML method and then return the training success rate.

>>8754500
>Eliminate 90% of the programs, keeping those whose responses are closest to the correct answers

This is not unsupervised

This is retarded. By building any sort of AI, would you not have to give it an organized structure?

Would that organized structure not be instilling it with a pattern in the first place?

>>8753578
Well, I could hard-code a fuckload load of algorithms in a processor, will that do?

Just a thought...

asking an A.I. to learn something it has no idea how to learn is the same as asking a blind person to see.

Maybe there simply has to be an underlying architecture that allows it to build an understanding.

Or, it could also be that learning happens through mistakes. Organisms evolve because of mutations. Our understanding of sight is dependent on our eyes. Our eyes are the product of mutations that had a positive influence on the survival of its species. In other words they weren't intended to happen but they did because of faulty transcription of DNA.

A.I. on the other hand could also possibly begin to learn new things through bugs. (assuming the A.I. entirely built upon 0's and 1's). Then again computers are not biological and may not ever react to accidents the same way a biological organism would.

>>8753578
If I write a letter, it's a pattern. Literally everything is a pattern. kys nubbbie wubbie high iq master race swag nig nog gg word up senpai

>>8753878
>being this arrogant

You are what is wrong with science today. Go watch black science man and masturbate to the god delusion.