A Tripping Neural Network

[Nathanial.Dread]

That psychedelic squirrel picture was floating around a while ago, and I managed to dig up a blog post from the engineers at Google whom I'm pretty sure made it.

As a neuroscientist, I have a LOT of feelings about this. As a regular tripper, I have a bunch more (most of them are enthusiastic ). If you think about how the layers of the neural network work, and how altering the function of different layers creates different effects, something very similar might be going on in our brains on high doses of psychedelics. Maybe at low doses, when patterns are mostly just being enhanced and simple visuals are forming, you're looking mostly at enhancement of V1 and other low-level visual information processing regions, but as the dose increases, enhancement begins occuring at higher layers (visual association areas maybe all the way down to the FFA and related regions).

Inceptionism: Going Deeper into Neural Networks

googleresearch.blogspot.co.uk

Blessings
~ND

 

[dreamer042]

Wow! This is absolutely fascinating!

Thanks for sharing. :thumb_up:

 

[DansMaTete]

I think this thread could be merged with this one

 

[MaNoMaNoM]

Wow! This is absolutely fascinating!

Thanks for sharing. :thumb_up:

indeed!

Maybe someday there will be 'psychedelic collective conscious' options for our cameras.

 

[Nathanial.Dread]

I think this thread could be merged with this one

I was hoping we could use this thread to discuss the implications of this particular post on the cognitive and neurological substrates of the psychedelic experience (at least, the visual aspect of it).

One of the things that I find interesting is that, even if this kind of thing is a good model for the visual aspect of a psychedelic experience, it probably doesn't tell us much about the emotional or self-related side of the whole thing. We know from drugs like 5-MeO-DMT (and, at least in my case, psilocybin), that visual 'oomph' is not necessary for a really profound ego-melting experience.

Blessings
~ND

 

[hixidom]

Those images are pretty incredible. So, will this make human artists obsolete?

 

[Nathanial.Dread]

There's a whole gallery.

Inceptionism: Going deeper into Neural Networks

48 new items

photos.google.com

This is, I think, my favorite one. It looks so much like what I might see when I look a blue sky on shrooms

Blessings
~ND

 

[Praxis.]

Woah. :shock:

Suddenly everything makes sense!

Seriously though, great find Nathanial, and thanks for sharing! This is by far the closest depiction of psychedelic visuals I have ever seen. I'd imagine that this could at some point be applied to non still-images, or motion pictures? I feel they could produce a pretty awesome video clip of psychedelia with this.

One of the things that I find interesting is that, even if this kind of thing is a good model for the visual aspect of a psychedelic experience, it probably doesn't tell us much about the emotional or self-related side of the whole thing. We know from drugs like 5-MeO-DMT (and, at least in my case, psilocybin), that visual 'oomph' is not necessary for a really profound ego-melting experience.

I find that curious also. I feel like in some way that my thought process follows the same mechanisms as described here, and visual input will create visions that reflect this...which has me wondering about the implications of CEVs vs OEVs. Are CEVs just a visual representation of the process that is taking place in your subconscious? I'm by no means a neuroscientist so maybe this is a silly question.

 

[hixidom]

I'd imagine that this could at some point be applied to non still-images, or motion pictures? I feel they could produce a pretty awesome video clip of psychedelia with this.

That's a cool idea. If there were a tripping scene in a movie that made use of this effect, I think my mind would be blown. The neural network would have to be more complicated though. These neural networks only identify objects, they don't identify actions. I get all :shock: just thinking about it.

I read that the same convolution neural network configuration is used to identify time series. I think that this same effect could be used to generate music!

 

[Koornut]

I find this a little startling.
I could be wrong but they sort of look like frames taken out of a larger video, like just before the real image snaps into resolution. Which kind of makes sense when you think that the computer I'm assuming, is taking the average from a larger associated set and painting it.
It makes me wonder if that is all our brains do to process imagery more efficiently, but in real time. So none of the weirdness we see here gets through to the final frames of our perception.

 

[Nathanial.Dread]

I'd imagine that this could at some point be applied to non still-images, or motion pictures? I feel they could produce a pretty awesome video clip of psychedelia with this.

That's a cool idea. If there were a tripping scene in a movie that made use of this effect, I think my mind would be blown. The neural network would have to be more complicated though. These neural networks only identify objects, they don't identify actions. I get all :shock: just thinking about it.

I read that the same convolution neural network configuration is used to identify time series. I think that this same effect could be used to generate music!

You could apply this "filter" to every frame in the movie maybe? I wonder how dissimilar two images have to be before the computer generates wildly new images that don't flow when seen rapidly, like frames of a movie.

Praxis: I've wondered a lot about CEVs vs. OEVs actually. I don't wonder if OEVs aren't the result of some kind of feedback, similar to this. You close your eyes, and your brain is robbed of visual input and so it projects it's normal grey, shapeless phosphenes onto your visual field. Maybe information about what you're consciously perceiving is somehow taken up and a process similar to this is done on it, projected, taken up, etc. etc.

Just a thought? I have no idea though.

Blessings
~ND

 

[hixidom]

You could apply this "filter" to every frame in the movie maybe? I wonder how dissimilar two images have to be before the computer generates wildly new images that don't flow when seen rapidly, like frames of a movie.

That would work and I would pay to see it :want: , but I just meant that it might produce an effect more true to a psychedelic experience if it could do with actions what it currently does with objects.

 

[Rifle]

Woah. :shock:

Suddenly everything makes sense!

Seriously though, great find Nathanial, and thanks for sharing! This is by far the closest depiction of psychedelic visuals I have ever seen. I'd imagine that this could at some point be applied to non still-images, or motion pictures? I feel they could produce a pretty awesome video clip of psychedelia with this.

One of the things that I find interesting is that, even if this kind of thing is a good model for the visual aspect of a psychedelic experience, it probably doesn't tell us much about the emotional or self-related side of the whole thing. We know from drugs like 5-MeO-DMT (and, at least in my case, psilocybin), that visual 'oomph' is not necessary for a really profound ego-melting experience.

I find that curious also. I feel like in some way that my thought process follows the same mechanisms as described here, and visual input will create visions that reflect this...which has me wondering about the implications of CEVs vs OEVs. Are CEVs just a visual representation of the process that is taking place in your subconscious? I'm by no means a neuroscientist so maybe this is a silly question.

I think it does tell us about the "emotional or self-related side". Artificial neural networks were inspired by the general features of biological neural networks. In other works the algorithm that led to this is based on how the brain appears to function in general, not specifically just the visual cortex. So the results seen here have implications for the majority of things that occur in the brain. I can definitely see how non-visual aspects of tripping could arise out of the same processes that created these images.

As for CEV vs OEV, the blog post goes into how some of the images were made out of images of just random noise. Basically, w/r/t the ideas in the post, CEV's would simply be what your brain creates when lacking stimulation by repeatedly running what ever background random noise is present through a level or various levels of cognitive processing while using the result as the new starting point in each iteration.

Anyways, I actually can here to post about this same Google blog post, and low and behold, it had already been done.

 

[DreaMTripper]

Incredible, mind blowing...the fourth one down on the right hand side is pretty much how I see things after a decent hit of acid when the lights fading.