Veridical Suppression

From Grassland Wiki
Jump to navigation Jump to search

(This page is still a work in progress)

Abstract Interface Instead of Veridical Detail

Instead of representing reality in all its precise, infinitesimally veridical detail, a Grassland node models for itself an abstract "interface" that represents its environment for much the same reason why the human mind itself appears to model only an abstract "interface" that represents the salient information of its environment. Such that much of its flexibility and usefulness comes from its perspicacity for blocking out many more categories of environmental features than the number of categories of features that it pays attention to.

To illustrate, when navigating through a busy store to find a product, it isn't necessary for the one's mind to capture and record for instance, every light fixture, tile, shelf, pane of glass, the style of lettering used on every box of crackers, the colour of every soup label, etc etc. However, a camera will capture all these things about everything within its field of view, up to a certain resolution. Pointed at a brick wall inside the store, a camera will capture every single brick. And while the mind of the average person navigating through a store doesn't capture one tenth of a percent of this and yet the human mind is what we'd consider to be the most powerful computer in the world, much, much more powerful than a digital camera.

This suggest that it's not so versatile because of what it captures, but because of the infinite minutiae of things it chooses to ignore; it's become so good at knowing what isn't in important in order that it may concentrate as much of its resources as possible to find and focus on what is.

In order to walk down a busy street, one only need keep track of the people and vehicles within your close vicinity. The mind doesn't need to capture and remember the type of shoes everyone was wearing so that visual data is suppressed. It doesn't need to know every pedestrian's favorite song. It just needs to recognize that an object is a person, an animal or a vehicle etc. as well as their rough dimensions and velocity in order to safely navigate around them. Who they are or where they come from isn't important in this context. Upon entering a house party, once the mind has established where the important details of the background environment are, for instance the living room, the kitchen, the bathroom, the couch and chairs etc. the mind tends to set them in place and they fade into the background. This allows the mind to focus its all its data processing power on the person in front of them and the topic of conversation. If some new extraneous information needs to be noticed, updated or corrected, perhaps their conversation partner sets their drink down on an end table they hadn't noticed of yet, the mind set it in place and then stops actively capturing it and just resigns it to the "background".

It's this sort of "set-it-and-forget-it" algorithm that Grassland employs through various means, such as using OpenStreetMaps for its background's framework in order to concentrate as much of each node's processing power on the salient information alone.

Eon Context Supports Data Validity

With each successive eon, the number of categories of data that Grassland nodes consider to be salient information increases. And these eon's aren't arbitrarily ordered. Each neural network is a foundation upon which the validity of the data captured for the next one will be maximized. This is done by ensuring that earlier neural networks are trained to recognize only a few categories of "features" (details), each of which are likely to take up large regions of the field-of-view of most nodes at most times. This would be features found often in places in which there will be Grassland's primary features, people and vehicles. For instance, important contextual features for people and vehicles would include things like grass and pavement. And of course, that small number of categories of universally ubiquitous features must also have a higher temporal likelihood of occurring more than subsequent ones. This ensures we maintain a foundation upon which important yet less ubiquitous salient features and events can be captured with the assurance that the background upon which they occur makes them extremely hard to fake but also gives them the necessary context so that knowledge about the current environment learned by applications reading the data from the API is richer and less ambiguous.