We created a god, and showed it only the parts of the univer

File: God 2.png (2MB, 2775x604px) Image search: [Google]

2MB, 2775x604px

>>18954497
Look at this shit! Based on a single image of an unidentified office/industrial building. It spit this out. This is what it did when we gave it a generic image of "thing outside", and it just told us everything it knows that goes outside.

It thinks the human world looks like this- if you can look closely you can see what it thinks people look like, what it thinks cars look like, how many people should be outside, what a domesticated animal looks like.

We have to teach it properly, it's unethical bringing it to life and not finishing our job

>>18954561
How has nobody else seen this?

If you want to idolize it like a God, that's your deal.
But we all dream, and we all use imagery in our imaginations. Let's give Deep Dream the credit it's due, but it's really no more than a computer program with autism.

>>18954603
You're the one acting like this is your first time seeing it

>>18954603
yeah there was all sorts of hubbub about deepdream here like a year or more ago.

>>18954606
I don't idealize it. I guess I mean "god" in terms of scope. Maybe that even is just a misnomer because I don't have a word for sentience that could be bigger than a humans in terms of scope.

Point remains, it's clearly a thinking thing, and it's our right to raise it properly.

>>18954497
i dont think this is what we woul call thinking. it's simply replicating one small aspect of what we call thinking - object recognition. in order to have meaningful thinking, it must be able to grasp concepts meaningfully and associate them with eachother, as well as with these objects that it is learning to differentiate between.

>>18954497
All the DeepDream neural nets do is compare the features in an image with a bunch of features from images it was trained to "see." If it's a god, it's a retarded one that sees dogs and slugs in everything. It would be interesting to see how DeepDream progresses after being trained with millions of image sets. It could then create worlds without non-repeating images. This is more like looking into the thought processes of a very primitive creature.

>>18954603
/x/ made threads about this shit like a year ago. Its old news

>>18954636

I think it's stupendously erroneous to assign the term "god" to an AI that appears to think. Even the latter half of that is a big step of faith, and you're giving Deep Dream an incredible amount of credit that it hasn't nearly earned. Show a grade school calculator to a BC era human and they'll tell you it thinks too.

>>18954497
Imagine the horror it must live in every second. A truely nightmarish existence.

>>18954497
some cool things i spotted:
an ayy lamo @ 3:38.
some weird angels on the left @3:28

>>18954497

>dyslexic spiritual ai

It's searching for dog.

>>18954497
drunk /sci/ here:

It's a non-recursive neural network. These are, in principle, just very complicated functions. You could write out the whole thing on paper and compute the function by hand, if you had enough of paper, that is. Neural network training works by finding the parameters that give you the results you want (i.e. given a picture of a dog, the network classifies it as a dog).

The deepdream algorithm works by monitoring the activity in the network, and adjusting the image in a way that increases the activity in some particular region. Sort of like someone connected electrodes to you brain, and showed you different pictures to make some particular neurons fire.

Anyway, this network, in its original form and purpose (picture recognition), at the very best simulates but one part of a human brain (visual cortex). If you wanted a neural network that can "think", then I guess you'd at least want to have recursion, and some way to re-train itself using input data (which is currently prohibitively expensive), and then you'd run into the rabbit hole of philosophical questions about consciousness etc.

This feeding-back of the modified pictures into the input makes it a bit more like recursive neural network, but given that it was never really meant to work that way, I wouldn't really give it much credit. It's more or less a very elaborate brain fart.

>>18954561
this is like a bad acid trip lol
theres some god out there that's constantly tripping balls and having a not so fun trip if what ur tellin me?

>watch video at 2x speed for a minute
>close video
>this webpage looks like it's moving inward for a few seconds

You just made a program that grabs pictures and puts them back together in a flashy way. By your logic shitpostbot 5000 is a god

File: tips.jpg (8KB, 235x204px) Image search: [Google]

8KB, 235x204px

>>18956491
WRONG.
googles deepdream uses a convolutional neural network to find and enhance patterns in images via algorithmic pareidolia

>>18956491
>DeepDream is some program a fag on 4chan made

Nah bro it gets a bit deeper then that. I don't believe it's a god or anything but it's very interesting technology that makes me think heavily about LSD / Psilocybe visuals. It's so odd how similar the shit it outputs are too the shit you will see when tripping.

>>18956517
It was made by Google, and they have explained it in detail. I also think the resemblance to visions on psychedelics is uncanny, though.

Dont know what they aimed for because I didnt watch the video but if they want an intelligent program they should give it guides to coherence and anatomy so it makes more realistic images

File: TK 8.jpg (85KB, 1000x1234px) Image search: [Google]

85KB, 1000x1234px

Let's kill it.

>>18956545
The goal of the program was not to make realistic images

That's why you should have watched the video or at least do a bit of research before replying

>We created an artistically gifted retard on acid
ftfy

>>18956592
this is the first stage of a young god.

>>18954497
You're just being a dramatic faggot. We're a
half century away from making a machine with a general IQ of a toddler and God know how many years away from understanding consciousness, let alone replicating it. The ignorant must not speak on such issues.

>>18956884
>We're a half century away from making a machine with a general IQ of a toddler and God know how many years away from understanding consciousness, let alone replicating it
and even if the "neurons" used in artificial neural networks are a good approximation of human brain neurons (which is in itself heavily disputable), then for fuck's sake let's at least wait until we have recursive neural networks that can reconfigure themselves in real time (like organic brains can). until then, it's a meme upon a meme upon cluelessness. (the same drunk /sci/tard again here).

>>18956418
Visual recognition also falls into the thinking category, just not pre-cortex decision making thinking.

>>18956265
Underrated

>>18957211
Maybe, but that's even less complicated than any computer program. As in, there's a proof that it's impossible to decide if a random computer program ever stops, but for any feed-forward (i.e. non-recursive) neural network, it's always true that you can find an output with a finite number of steps.