Google's neural networks see things that aren't there - awesome computer acid art

Mr. Tea

Let's Talk About Ceps
I'm sure a few of you must have seen this already - could just as easily have gone in Art or Thought, but anyway:

http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html

Basically Google researchers have used neural net programs trained to pick out certain features, such as animal faces or buildings, and put them to work on an image containing none of these things in conjunction with an algorith to gradually turn parts of the image that look a tiny bit like the sought image into that image, and then iterated it many times. The results are incredible:

building-dreams.png


skyarrow.png


What I find amazing is how incredibly similar they look to some of the stuff you see when you're tripping. In fact it seems to back up a hunch I've had for some time: that a lot of the visual effects of psychedelics arise because they turn up to 11 the circuits in your brain that deal with pattern recognition and detecting streams of meaning in otherwise noisy signals, like how the random dirt on the wall in a nightclub toilet suddenly appears to be alive with slowly morphing faces.

Cool stuff happens even when they just train the program to pick out and exaggerate edges:

seurat-layout.png
 

droid

Well-known member
What I find amazing is how incredibly similar they look to some of the stuff you see when you're tripping. In fact it seems to back up a hunch I've had for some time: that a lot of the visual effects of psychedelics arise because they turn up to 11 the circuits in your brain that deal with pattern recognition and detecting streams of meaning in otherwise noisy signals, like how the random dirt on the wall in a nightclub toilet suddenly appears to be alive with slowly morphing faces.

Thats well known isnt it? A combination of the that and entoptic phenomenon.


All good stuff, though the Seurat is easy pickings if youre looking for edges.
 

Mr. Tea

Let's Talk About Ceps
Thats well known isnt it? A combination of the that and entoptic phenomenon.

Is it? I mean it may seem subjectively obvious to anyone who's taken psychs, but I have no idea if this was just a widely-held hunch or is well supported by neuroscience.

All good stuff, though the Seurat is easy pickings if youre looking for edges.

Damn droid, you're a hard man to impress!

I wonder what would happen if you fed the program an Alex Grey painting? Maybe it'd turn it into a photo from a Next catalogue.
 
Last edited:

luka

Well-known member
I've never had a hallucination I feel cheated I want my world to be an Alex grey painting
 

Mr. Tea

Let's Talk About Ceps
To paraphrase Hunter S. Thompson, "Where're all these acid flashbacks I was promised?"
 

Mr. Tea

Let's Talk About Ceps
Hardly upset, just slightly amused and commenting on the fact that you're obviously a highly verbal/textual/lyrical person.
 

Mr. Tea

Let's Talk About Ceps
Interesting paper, but it's surely a major oversight not to discuss or quantify the apparent fractal nature of geometric hallucinations?

You know, I'm not entirely convinced the apparent similarity between fractals and some of the form constants seen while tripping isn't a coincidence. Or rather, if it's not a coincidence, if the causality is the obvious way around.

Consider that 'fractals', as the public understands them (representations of the Mandelbrot and Julia sets with a very garish colour scheme chosen) started becoming well known in the late 80s/early 90s thanks in large part to James Gleick's Chaos. In other words, exactly the time acid house and rave culture were kicking off, ecstasy was becoming the club/party drug of choice and there was also a resurgence in the use of LSD.

So I'm wondering if people making images of certain fractals were inspired to use colour schemes that gave 'trippy' results, rather than there being something inherently fractal about the patterns you can see when you're tripping.
 
Last edited:
Possibly, but it's been established in the past decade that the neuronal organisation within the brain is fractal. This paper pre-dated that though, so fair enough. Neuronal architecture is self-similar on many, many levels, ranging from dendritic branching to distribution of relative voids in the grey matter. So it's not much of a stretch to speculate that geometric hallucinations owe some of their repeating aspects to fractal patterns.
 

Mr. Tea

Let's Talk About Ceps
That's an intriguing possibility.

I wonder if Roger Penrose is aware of that paper droid posted above, and what he made of it if so.
 
Top