3 ms·

Can't remember where I read this, but the idea is that the brain compresses a vast amount of data down to a small amount of meaningful information. VR's current

by _greim_ 10y ago

Can't remember where I read this, but the idea is that the brain compresses a vast amount of data down to a small amount of meaningful information. VR's current approach is to meticulously spoof that vast datafeed in such a way that the result gets compressed down to something that seems meaningful. Which is a lot of work. Why not go in post-compression and merely spoof the meaningful bits? VR's job could be a lot easier. I don't know how it would play out, but speculating, you might render a low-poly monochromatic "skeleton" scene and then simply suggest to the brain ways to flesh it out with feeling and texture. Show the user photos of war-torn Europe in 1947 beforehand for example. Even if it means direct neural interfaces, considering that we're 20 years out using the current approach, maybe that isn't so far fetched.

exhaze 10y ago
This is the premise of a 90's sci-fi novel about VR (https://en.wikipedia.org/wiki/Labyrinth_of_Reflections https://en.wikipedia.org/wiki/Labyrinth_of_Reflections). In the novel, a character discovers a graphical sequence that puts people into a hypnotic state in which they can be shown a low-res virtual environment which their brain then fleshes out.
mxfh 10y ago
Aren't these highly compressed and encoded concepts called books and language already?
captainmuon 10y ago
I always wondered if you could use adversarial neural networks to find images that the brain recognizes as something else, just like you can fool NNs. You could use this to find low-poly representations of your objects that are indistinguishable from the real ones. Probably you'd have to calibrate this for every user though.
- usrusr 10y ago
  You might need to reset the user occasionally during training, otherwise that adversarial neural network would hunt a moving target.
nnq 10y ago
> Even if it means direct neural interfaces, considering that we're 20 years out using the current approach, maybe that isn't so far fetched. To be economically viable, the tech needs to be sold to be mass-sold directly to consumers. So it needs to be mostly safe, even when it has bugs, and usable by anyone, without months of "training your interface device and brain". Even if there was some progress with "direct neural interfaces", without a breakthrough in nanotechnology I'd be we're probably more than 20 y away from something usable outside of a hospital or a military lab that has a different cost/risk/benefit equation from the "average joe" consumer... And you probably don't realize it, but to get that "spoofing" part right, you might have MUCH HARDER problems to solve than having a ultra-high-re ultra-low-power ultra-low-latency mobile VR set...
modeless 10y ago
One way to do what you're suggesting is foveated rendering. If you have very fast and accurate and robust eye tracking, you can avoid rendering most of the pixels that lie in your peripheral vision (where your eyes have very low resolution). Future VR headsets will almost certainly need this.
pessimizer 10y ago
Because ultimately, things are compressed into a vivid narrative, or compressed even farther into a lesson learned, and we've had the technology to do that for millennia. The purpose of VR is to suck away money from investors by teasing them with a gravy train of young men sitting around in headsets. Like 3D TV, except less likely. Stories and lessons won't work for that.

by _greim_ 10y ago

exhaze 10y ago

This is the premise of a 90's sci-fi novel about VR (https://en.wikipedia.org/wiki/Labyrinth_of_Reflections https://en.wikipedia.org/wiki/Labyrinth_of_Reflections). In the novel, a character discovers a graphical sequence that puts people into a hypnotic state in which they can be shown a low-res virtual environment which their brain then fleshes out.

mxfh 10y ago

Aren't these highly compressed and encoded concepts called books and language already?

captainmuon 10y ago

I always wondered if you could use adversarial neural networks to find images that the brain recognizes as something else, just like you can fool NNs. You could use this to find low-poly representations of your objects that are indistinguishable from the real ones. Probably you'd have to calibrate this for every user though.

usrusr 10y ago
You might need to reset the user occasionally during training, otherwise that adversarial neural network would hunt a moving target.

nnq 10y ago

> Even if it means direct neural interfaces, considering that we're 20 years out using the current approach, maybe that isn't so far fetched. To be economically viable, the tech needs to be sold to be mass-sold directly to consumers. So it needs to be mostly safe, even when it has bugs, and usable by anyone, without months of "training your interface device and brain". Even if there was some progress with "direct neural interfaces", without a breakthrough in nanotechnology I'd be we're probably more than 20 y away from something usable outside of a hospital or a military lab that has a different cost/risk/benefit equation from the "average joe" consumer... And you probably don't realize it, but to get that "spoofing" part right, you might have MUCH HARDER problems to solve than having a ultra-high-re ultra-low-power ultra-low-latency mobile VR set...

modeless 10y ago

One way to do what you're suggesting is foveated rendering. If you have very fast and accurate and robust eye tracking, you can avoid rendering most of the pixels that lie in your peripheral vision (where your eyes have very low resolution). Future VR headsets will almost certainly need this.

pessimizer 10y ago

Because ultimately, things are compressed into a vivid narrative, or compressed even farther into a lesson learned, and we've had the technology to do that for millennia. The purpose of VR is to suck away money from investors by teasing them with a gravy train of young men sitting around in headsets. Like 3D TV, except less likely. Stories and lessons won't work for that.