Why does AGI not yet exist?

The frameworks being used to model Artificial General Intelligence (AGI) are primarily interdisciplinary in nature. This means that people use knowledge informed by only a handful of fields. There has not yet been an attempt based on the transdisciplinary synthesis across all domains, nor has there been a synthesis across transdisciplinary, big picture, unification frameworks themselves.

Another way of putting this, is that attempts at making AGI do not have a unified framework that accounts for, describes, and puts into natural arrangement universal classes of what exists and universal classes of how intelligence grows. One cannot make an AGI that operates similarly as a person unless one understands the total scope of what is possible for a person, and can model it. This is what we believe we’ve done – well enough to proceed with prototyping an AGI.

Our approach

While current approaches to building AGI are almost entirely through neural networks, our approach treats neural networks as one type of intelligence along a larger trajectory of intelligence growth. Our approach is twofold.

First, we are converting mathematical expressions of arches into code. Arches are an outcome of archdisciplinary studies – proposed non-arbitrary universal laws for how cognition (and possibly how the universe) works. Universal computability, complexity, integrative levels, and ruliad are examples of arches. For example, universal computational properties like static (entities), dynamic (relations), and multinamic (systems) are a requirement for cognition and the universe to function at all. Or if we removed complexity from reality, we couldn’t have things existing at the same level, or have entities as building blocks to combine into a new level. We are using these proposed laws as the fundamental units for building the simulator.

Second, we are converting mathematical expressions of an archtheory called architectonic simulation into code. An archtheory is a theory that synthesizes big picture, transdisciplinary, theory of everything, unification metatheories into a higher order theory. Architectonic simulation synthesizes across these big picture frameworks and synthesizes what the smartest people in the world have all came to similar conclusions about. In architectonic simulation, these classes are called simulation classes. such that the architectonic simulation framework accounts for, describes, and puts into arrangement all known simulation classes by their natural relation. It does so through a fractal heterarchical taxonomy of simulation classes from most universal to most unique. It is fractal because there are a self-similar patterns that underpin and pervades across knowledge and intelligence, it is heterarchical because it is partially hierarchical and partially flat (n-dimensional), and it is a taxonomy because it organizes classes of simulation by genetic inheritance of characteristics.

There are no black boxes in our framework. The means in which the simulator coordinates information into increasingly complex structures is always observable in this AGI approach. Furthermore, training the simulator will look differently than one might expect – we’ll be training it similarly to how a person learns. Just like how a person is born with all the capacities of knowing and doing yet needs to develop them in an order, so too will this system.

In simple terms, this is a general theory of simulation, with a well-organized, non-arbitrary, unified framework for representing, relating, and systematizing universal simulation classes of knowledge and intelligence along universal patterns of nature and cognition.

Orders of simulation

This fractal heterarchical taxonomy consists of three orders of simulation: computational simulacra, ratiocinative simulacra, and instantiation simulacra.

• Computational simulacra is a self-referential system that orients, transforms, and iterates encapsulated complex patterns. Though neural networks are included in our approach, we base our approach on a fractal phase calculus, a ruliadic representation of all possible states and transitions that any given simulation can undergo while representing and transforming data.

• Ratiocinative simulacra is the means in which computational simulacra computes, and consist of universal architectural, processual, and calculatory classes. Architecture classes are universal simulation classes of what things are. When we simulate things in our minds, we do so using universal types. Process classes are universal methods of how things do. When we relate and transform ideas in our minds, we do so along a trajectory of stages of intelligence. Calculatory ratiocination is where architectures and processes are combined simultaneously. Neural networks and symbolic computing are expressions of process ratiocination, and are fitted together in the design in the same way they happen analogously in people.

• Instantiation simulacra is the content and occurrences that ratiocination calculates. Instantiations can be anything, including architectonic simulation itself as an instance.

Goal

Goals in the design are as follows:

• Architectural coherence: Identifies and organizes data in terms of their mapping in the overall whole of knowledge, and makes suggestions for relevant knowledge for a given task.

• Processual coherence: This is the “general intelligence” part of our approach to artificial general intelligence. Our framework accounts for, puts into natural relation, and systematizes all forms of intelligence. Performs and identifies the general intelligence i.e. behavioral complexity of users, scaffolds them up transitions from one order of behavioral complexity to the next – fitting to where users are at developmentally.

• Universal calculation: Assists individuals and groups in mostly any endeavor. While it could directly solve problems on it’s own, for most cases it would be implemented for human-computer joint problem solving towards scaffolding people through direct experience so that people learn how to think and act in more complex and integrated ways.

• Singularity coordination: As a self-referential system, the design inherently seeks to create equilibration across all it’s data and processes, which informs how it relates to users across the system.

• Interface: Users engage with the AGI like one would engage with a person. We’d like to design an interface that gives users some freedom to design their own tools for specific use-cases.

• Data & machine processing: We think this is going to take up vastly less storage and computational power by several thousand magnitudes over current LLMs, and would run natively on average power machines. The reason is because neural networks are here mostly only used for lower stage tasks, and symbolic computing is used to do the more intensive tasks. Further, we have a method for compressing data fractally, that, for whatever reason, no one seems to be using.

Ethics

Our ethics is informed by works found in Kohlberg, Carrol, Commons and Sonnert, Loevinger, Cook-Greuter, and Freinacht, among many others. In other words, we look across a multiplicity of developmental psychology frameworks jointly to measure how complex and integrated ethical decisions are.