A voxel is a 3d pixel, a point in space with a color. Graphics cards do not support them but they're so simple that even with just the CPU they are drawn on screen fast enough. Voxels are like the corners between polygons, except theres no polygons between them.
I'm not here to argue the advantages of voxels over polygons as that debate has gone on for many years and polygons are usually used instead.
This is about a new kind of voxel which is less about graphics quality and more about intelligent patterns of movements of colors and shapes. Its about how colors flow between the voxels which each know a few other voxels they're near and that distance.
At any one time, there are some set of voxels on screen, each with a color, an amount of red, green, and blue, so a voxel is 6 dimensional, x y and z position and 3 for color.
Of the 3 color dimensions, or really all we need is 1 color for a proof of concept of Prediction Voxels, every voxel predicts the color of near voxels and is rewarded or punished based on the accuracy of those predictions. Waves of color will flow this way similar to brainwaves.
This space of voxels and flowing intelligent waves of color is 1 space for the whole Internet and can run on a simple peer to peer routing protocol needing no user accounts or management or maintenence of any kind.
Each pair of connected computers simply exchanges information about some set of voxels they have in common and chooses to continue that connection only if it predicts more accurately than other connections. Each computer would keep maybe the best 4 connections at any one time.
The system would be low bandwidth and efficient because positions and colors (and numbers as names of specific voxels), can all be represented as small binary numbers depending on how many digits of accuracy are needed, and because its designed from the start to be purely a prediction system at all levels (down to single voxel/variable predictions of eachother), compression happens automatically simply because evolved forms that store less data to predict the same accuracy have spent less computing resources, are rewarded/punished for that positively, and therefore the representation of whatever forms we build the voxel networks into will converge toward more efficient representations.
It sounds complex but all you really need is a simulated money system, like Ripple flows agreement of debt based on trust between each pair of friends in a social network, which can be calculated as efficiently as any other variable in realtime locally in memory of each computer and between computers. Money is a number that allows evolved forms in this system to consume cpu, memory, and other resources, to build more evolved forms with the voxels and code they contain.
Voxels must be able to contain some kind of code that runs in a sandbox. We only need the most basic abilities like list, hashtable, plus, minus, multiply, divide, pointers to other voxels, and a few standard AI algorithms like neural activation functions and bayesian networks at the same small granularity. In that way, this peer to peer network of a global Prediction Voxel space would be a cloud computing platform with Ripple-like money built in at a small enough granularity we can use it as a replacement for AI scoring systems. Bitcoin, for example, has an ability to pay the network to run arbitrary code in a simple language written into the block chain, but Bitcoin is far too complex to attract the practical use of it for cloud computing. Instead, Prediction Voxels would calculate in smaller amounts of money and without security so its efficient enough to use like any other variable. As long as the network of computers can agree approximately which groups of voxels and Internet addresses have contributed more or less to prediction accuracy, high security on those variables counting such contributions costs more than its worth. Lets do the simplest thing that will work.
It would be a massively multiplayer game. Each person sees some voxels moving on their screen and flowing colors in intelligent patterns (sometimes like brainwaves) and can interact that way or go deeper and write AI code into specific voxels to define how they interact with other voxels, and that AI code could spread depending on how much Ripple-like money those voxels earn by their accurate predictions of the colors of near voxels.
Prediction adds real value in any system, in a simulation or the real economy, and these parts of the world flow together when prediction ability of one system becomes advanced enough to compete.
People could use the Prediction Voxel system as their job, to contribute their brain power to the cloud to predict whatever patterns others put in and cause to have high payments for predictions of.
Like Bitcoin, the approximate money system would have to be backed by something thats hard to create but easy to verify. Bitcoin uses a "proof of work" which is a huge calculation any computer can do, finding a number whose secure-hash starts with a large number of zeros for example, and that is done as a competition to strengthen the security of the newest root block in the chain. Prediction Voxels don't need to complicate it that far. They only need proof that some large calculation was done on any data in the system, which most computers in the network would agree to accept as the idea "I did all this useless calculating so please take me seriously and accept this data instead of assuming I'm just another spammer trying to contact a large number of computers and spread large data at low cost to myself." A computer could, for example, only accept connections from other computers which waste 3 seconds doing such a "proof of work", which would prevent the ability for Denial Of Service attacks and change the economics of how much prediction is practical to do before choosing who to attempt connecting to next. Theres many things it could be used to throttle and fine tune the value of.
The data structures would be very simple. A voxel at any one time has a list of near voxels sorted by distance. Each voxel also has a software written in the scripting language of this cloud, which defines its behaviors in terms of these simple data structures. A voxel could be programmed to explore the network of other voxels through their lists and find groups of higher density, where more of them in the group connect to eachother, or to find certain patterns of colors or shapes in the distances between other voxels, or any arbitrary thing the Prediction Voxel economy may evolve toward paying the evolved forms to do.
At the core this global economy exists to do only 1 thing: predict the color of near voxels more accurately than other near voxels do, so the predictor wins money from them instead of losing it to them. Its a gambling system where every prediction can lose or gain you money and the smartest voxels reproduce and evolve.
My question to you all is how could such a system be designed so the economy that forms in it can practically be used to evolve the voxel shapes and flowing colors toward a kind of AI dream space where the patterns we draw with each of our mouse (and other game controllers or EEG devices for example) will navigate through the dream space of possible patterns of flowin colors and shapes? How can we build this idea of Prediction Voxels, incredibly simple at the core, into an economy that evolves into a global competition between AIs and people to predict the dream space more accurately? Unlike Wall Street, can we find a practical way for those half billion people starving to use their brains to amplify the prediction ability of the cloud and strengthen the global economy? What details of this very simple design would evolve into such a competition of who can dream the best in the shared 3d space? Wearing an Emotiv Epoc or OpenEEG game controller, can we obsolete half of all jobs and instead dream our intelligence into the Internet? Why should we need something more complex than voxels which know what other voxels they're near and predict eachother's colors as gambling? If businesses are the unit of calculation of the world as it is today, is it not a trivially easy competition to obsolete them by reducing that granularity to the voxel level, like stock market investing for every bit on the Internet would be a better way to organize the world? I think it would work, but theres so many ways to screw up by adding unnecessary complexity. Einstein said "Make things as simple as possible but not simpler." Can we build it as a cloud computing language for only Prediction Voxels and Ripple-like numbers as their money, and let the global economy build the game world on those few lines of code?