Machine Learning

>TFW AI will never be creative
it will always just work off a dataset

>>inb4 machine learning
machine learning is just a data set run through repetitive testing cyles.

Other urls found in this thread:

en.wikipedia.org/wiki/Cryptomnesia
twitter.com/AnonBabble

>tfw you're a brainlet who can't into statistics
>tfw you have no idea what you're talking about, but feel the need to say it anyway

try an unsupervised online ml

You think you're creative? Can you come up with a concept that isn't just a variation on things you already know?

>user asked master foo when he was taking a walk on the beach
>"How can AI be creative if it can only work off a dataset?"
>master foo calmly replied:"what is a dataset?"
>user quickly said: "a large set of numbers that represent knowledge"
>suddenly master foo stopped and drew numbers in the sand
>"do these numbers represent anything"
>"I dont know, they're just numbers in the sand they could represent anything"
>master foo smiled
>in that moment user was enlightened

Humans aren't any different though.

This
en.wikipedia.org/wiki/Cryptomnesia

How do you think humans learn, user? Our brains do the same thing.

What is human creativity? You take a massive amount of experiences and things you're familiar with, and mix them up with a bit of randomness.

Sufficiently advanced machine learning should be able to do this. See Deep Dream for a primitive example. The challenge is getting it to do it at as advanced a level as humans, and with similar tendencies as humans.

But they do it with 1/1000th the required data set size and in a few watts of power

>Deep Dream
Deep Dream has nothing to do with creativity. It pretty much tries to show what a CNN is doing. It just tries to find an input that creates the highest score possible and displays that input.

This

>1/1000th the required data set size
We have a training set the size of our lifespan, covering all senses we have. I think that's a bit bigger than anyone's used with machine learning.

They are. They have bigger datasets and more powerful hardware.

...or should I say... brainware?

LOL

>1/1000th
>He actually believes this

Sorry, that wasn't funny.

Human creativity is also based on existing shit.

>it will always just work off a dataset

what exactly do you think creativity is?

davinci didn't paint mona lisa in a vacuum.

picasso didn't just "randomly" start making abstractions.

memories and experiences are just low-fi datasets, whether you like it or not

WE HAVE A SUBJECTIVE EXPERIENCE THAT CAN ONLY BE IMPERFECTLY APPROXIMATED BY A COMPUTER


define creativity.

Well, you didn't need to express that.

I mean, you can be an asshole all you want, just be aware of it.

what the fuck

Isn't this a big problem with machine learning? How the hell will you know if the algorithm is working because you'll never be able to have a data set as large as somebody's life experience.

>you'll never be able to have a data set as large as somebody's life experience.

false

Well I guess you could be Google or something and have a lot of content but it wouldn't be continuous or in order for the network to process like we do. It also wouldn't produce meaningful info for like the first 5 years of processing the data.

> it will always just work off a dataset
That dataset can *easily* be far bigger than a human's.

You could easily derive more creativity and entertainment from existing data on the internet with the right learning & transformation.

> machine learning is just a data set run through repetitive testing cyles
Explain how this isn't human learning, too.

Yea, the software / hardware involved matters some, but at the end of the day we have a cold hard algorithm running on maybe not deterministic hardware (that nevertheless seems like something that can be modeled artificially if enough effort is invested).

Also creating a 1:1 replica absolutely is NOT required to get "creative" results. We are interested in and see "creative" patterns in all sorts of things that aren't even sentient or not human.

It has been called wetware for, oh, the last 20 years, user.

>get book about ML so I don't miss the boat
>still stuck on liner algebra section

FML why didn't I study calc instead of stats in university

Why didn't you study Linear Algebra in university

because stats was easier and i never thought I'd need it.

protip: the plan backfired when I did my graphics paper

stats is harder than linalg desu

Why isn't there a Vu phrase for this? Like deja vu and jamais vu.

memento vu

>tfw AI will never be able to fathom OP's retardation

>mfw had some pretty neat ideas
>looked them up after refining them a little
>all of them exist already
>I suddenly find out my creativity is probably non-existant
I-is this what being a brainlet feels like?

>maximum likelihood estimators
>estimators
>prediction intervals
>test statistics
>dependence and covariance
>qq plots and residuals
just fuck my shit up senpai

most truly novel stuff is already thought of user doesnot mean you can not reapply it to other areas

Well it would be interesting to see what sorts of "creativity" Google would come up with given the databanks are loaded to the brim with fake news, conspiracy stories and overall craziness. And for all the talk neither Google nor FB have been able to weed out the junk.

Intellectually speaking: "Hvor utgangspunktet er galest, blir titt resultatet originalest" a more than 100 year old quote.