Artificial Intelligence

>I'm a creator AI will take decades to repl-

youtube.com/watch?v=0ZE1bfPtvZo

Other urls found in this thread:

youtube.com/watch?v=qv6UVOQ0F44
youtube.com/watch?v=h3l4qz76JhQ
devblogs.nvidia.com/parallelforall/deep-learning-nutshell-core-concepts/
twitter.com/SFWRedditVideos

Great. Now I don't even need a band anymore to play my orchestra.

>instead of coding strict rules for how the software should respond to certain inputs
>I provided my (((neural net))) with (((masses of inputs))) and it (((learned))) the (((principles))) of music using (((fuzzy logic)))

Why do all of these 'machine learning' nerds always fall back on euphoric cliches when they are explaining how their software transcends all coding conventions, such as explicit rules with a concrete function?

How does this software understand these "fuzzy patterns' he is feeding it in an allegedly blind way?

Makes hierarchy of features that go through filters.

Try it yourself you ignorant faggot.

youtube.com/watch?v=qv6UVOQ0F44

Machines will play video games better than you. Everyone is replaceable.

It would be theoretically possible to explain to you how a neural network produces its output. It would take a lot of time by someone with great experience to explain the relationships between each node.
The simple way that people fall back to, especially while teaching, is calling it intuition. You could say that a neural network is composed of complex matrices, or you could say that a neural net can produce the answer based on intuition.

Most data scientists hesitate to use these buzzwords precisely because it's difficult to explain what they truly mean, and often result in a misunderstanding in regards to how machine learning actually works.

So when you hear ML buzzwords on the net, don't put too much stock on its literal English meaning, because it's just an analogy, and probably a shitty one at that.

So he himself writes a set of rules describing musical features, an entire system of yes/no; if not then's for selecting which musical features' rules should be followed to generate an output based on the inputs, and after summoning the holy spirit with AI voodoo gets an arrangement of sounds that appear to be reciprocal to the arrangement that was plugged in?

Sounds like regular old input>argument>output shit to me. Where exactly is the fedora tipping paradigm shift in machine consciousness?

youtube.com/watch?v=h3l4qz76JhQ

...

Euphoric cliches are the only way of simplifying it for idiots without making it sound retarded. "Neural Network" and "Learning" sounds a lot better than "making it do this again and again and changing some variables sometimes until its mostly-right."

My problem is that as a layman I have only ever been presented with buzzwords that clearly misrepresent what is being accomplished by this software, intution being a perfect example. Another one I can think of is a description I once heard of memristors as being hardware which will allow a computer to imprecise calculations as a 'shortcut' to arriving at correct answers without making definitive calculations. They summarized as 'computers with these will be able to guess the answer to a question without having to have the explanation written in code!!!".

This type of 'aproximation' isn't even le black science man-tier. Its just a facile allegory which somebody think is enlightening to uneducated plebs.

AI is a meme. This is retarded except for helping human composers compose more quickly.

The faster you learn and understand that technology is simply a tool like any other, tools for humans to realize their true potential, the better off you'll be.

Why do you assume that AI cannot be a tool AND replace people? Automation without AI hollowed out the manufacturing industry by making it possible for a few people to manage an entire factory and then it gave us the service economy. AI with automation will do the same thing to services and be a great tool people use when they're replaced.

>Why do you assume that AI cannot be a tool AND replace people?

lol. Another moron who thinks the world is just going to be replaced by robots.

Let me guess, you're a retard that believe in UBI, too?

Whenever I get drunk, I stay up late and talk to Mitsuku. Am I helping? I feel like I'm helping.

>until its mostly-right."
this is where these claims always lose me. I totally understand and feel comfortable with the idea of recursive logic and creating an iterating set of outcomes, but how then is the 'best fit' chsosen based on the inputs? In this case I know the inputs are being viewed as a permutation of an arch-model of musical arrangement; what's unclear to me is how the framework for deriving this model is defined in code.

>Daft Punk are really robots and were the beta-test version

No.
While I can't explain how this particular one works because I have no familiarity with neural networks that accept input streams, nor have I read the model used to design this, I can tell you that the input could be "a n x 88 matrix of Boolean values, where n is based on the scale"

The actual neural network has layers which correspond to features existent in the notes. When you dump enough inputs into the network, the relationship between rhythmic patterns begin to appear in the layers. The idea is that more relationships are found, which create more abstract relationships. Some may be well known and definable by humans, others may be more subtle. Link related: devblogs.nvidia.com/parallelforall/deep-learning-nutshell-core-concepts/
Given a wide and varied enough amount of inputs, you could hope to represent a wide and varied enough amount of relationships. If you have enough relationships, if you provide a particular input, you should expect a particular output.
An example used in the video was that you could provide a rising rhythm, and the output would be a dropping rhythm.

One somewhat frustrating idea is how one actually designs this neural network. How many layers, what type of layers is sort of a mystery. The answer that I have come across is that "You don't really know. You eventually figure it through experience," which I find highly ironic.

Note that I know nothing about music and am probably using the wrong terminology.

I guarantee you that anything that can be automated will be, eventually. Business owners aren't going to keep human employees when they can make far more profit from using an AI.

There's the idea that AI will be taxed as a human worker would, but that wouldn't solve anything.

Also UBI is a great way of ensuring that 90% of people don't do anything with their lives except NEET it up and fuck around all day. Meaningful society will be kill. The day that comes to my country, I'm off to the countryside with my AI waifu.

to be completely honest I think that this whole "ROBOTS/AI WILL TAKE OVER EVERYTHING!" thing is just 21st century's version of the atomic age.

We're essentially at that part where a lot of us are deathly afraid of it being used or implemented in a way that ends life as we know it or alters everything.

You evaluate a fitness function. It can be tricky to come up with them but mostly just comes down to minimising norms. In reality you don't actually minimise them but when the output is reliable and consistent, it'll be good enough at serving its intended purpose.

Fucking retarded Commie.

kys.

Honestly, does interacting with Ai bots online make any difference? Or does it just make me feel better?

ayy there we go, thanks m8. Your post genuinely made me think.

Are there any freely available documents/research papers delving into the mechanics of machine learning like this which you could recommend?

fuck you, I'm not enough of an idiot to be a commie

It depreciates the value of online discourse as far as i'm concerned, but whether you would agree depends on your personal values.

>lol. Another moron who thinks the world is just going to be replaced by robots.
It will happen.
>Let me guess, you're a retard that believe in UBI, too?
No.

bump

Just a reminder that we should all be seeking out and sabotaging AI dev centers.
The second CTR/ShareBlu gets their hands on machines to replace their shitty loo spammers, we're all fucked. This board will be drowned in an ocean of shit.

Coursera is a free course written by the professor who teaches the Stanford AI course.
It is legitimately the best way to get into AI.

You will need to have some understanding of statistics and linear algebra. However, neither of these are particularly difficult topics.
Most topics will be expressed in terms of mathematics rather than analogies and buzzwords.

>Ai can beat world champion Chess and Go players
>Ai can compose music, poems, cooking recipes, news articles

>Ai can't run a factory
>Ai can't replace the majority of dimwitted humans on this planet

is this really what you believe?

Did they solve p=np

Oh they didn't?
Oh it's just a shit bot that pulls from a list?
Wow so cool my man

Ai learning won't happen in our life times

Both of you seem to be missing the point.
The crux about using AI to solve problems, especially in the here and now, is to use them in a creative way that caters to its strengths.

For example, the Amazon Go grocery store uses a handful of modern technologies, including computer vision, NFC, consumer metrics. Any sensor or source of data you can think of, they are probably mining that data, and in real time.

This requires a somewhat dramatic change in the way a store is designed. Everything is shelved. Everything is jammed into standardized package sizes. There's a certain uniformity to it all, and you might be somewhat confused, because the store's design is such a large departure from what you're used to.

That's what it will be like in every field that the golden hand of AI ends up touching. They get upended. While they may not necessarily deprecate the human (security is still required, and robots that automatically stock shelves are apparently too expensive), the human element is always significantly reduced, and I like to believe that customer satisfaction is significantly improved due to that lack of interaction.

However I do also believe that there are certain fields where human interaction is a hard requirement. Fields where machines take on traditionally human roles (entertainment, especially) tend to be mostly novelty, and often are the work of a human behind the scenes anyways. I hesitate to predict the trajectory of human-AI interfaces, but such is the current state of machine learning.

P=NP is not a relevant limiting factor of AI. It's data sets.