Arbor Artificial Minds

Arbor AI Theory

Back to Index

Arbor Project

The main idea behind Arbor is to uncompromisingly pursue the goal of building a true artificial mind; one which can pass the Turing test in various ways, with initial emphasis on language processing.

Theoretical basis

Main article.

A part of the research community asserts that we will someday achieve true intelligence by building myriad separate solutions to subproblems of AI, and these separate solutions will naturally grow together into one grand implementation of intelligence. (See also: Is Strong AI possible?, alanturing.net). I reject that notion, and therefore have invested all my effort into directly tackling all major facets of intelligence and cognition at once, aspiring to build one unified theory of mind from the ground up. This means that whenever I find an aspect of intelligence or human functioning which is incompatible with the Arbor model, or which violates its principles in some way, I redesign the model to incorporate the new insight. This has been an arduous process and has taken roughly 3.5 years of fulltime research and development to achieve. It bears clarifying that this does not mean Arbor tries to be equal to a neural model. It does mean to be equivalent.

Implementation progress

Main article.

Arbor is now at a point where enough major features are in place to produce significant results. The key feature of Arbor that makes these advances possible, is that virtually no part of the algorithm works in a top-down manner. Everything is based on single core unit, which I've dubbed the dendron, which supports a key set of basic properties. Every dendron is defined by a varied set of connections with other dendrons. A distributed algorithm sits on top of this, ensuring that dendrons can influence each other in powerful ways, and maintaining the constraints that govern the knowledge network.

The result is a sort of idea language which can store any type of abstract information in an organized and meaningful way. The algorithm allows for dynamic problem solving within this abstract model, one example of which is the interpretation of natural language to its abstract counterpart, and vice versa. This means that the application can understand the meaning of language, and produce meaningful language in return. It is not only language-agnostic, but even data-agnostic. In other words: the algorithm works irrespective of which language and more importantly: the algorithm does not require any knowledge of the nature of language itself to be hardcoded, because the algorithm is general enough to allow for this. Arbor could therefore be expanded to solve any problem of cognition with the same algorithm. It can then also combine information from different "senses" to form new ideas and connections between them. It follows naturally from this that Arbor can also perform meaningful operations on completely abstract information using the same algorithm.

Future development plans

Main article.

Autonomous learning is a top priority for the project, because it will open a lot of doors. It is a complex process to incorporate fully autonomous learning into the existing algorithm, even though much of the groundwork has already been done. The process is complicated by the distributed nature of the Arbor algorithm, which is more arduous to debug.

After several years of rushing development on Arbor to reach a prototype to prove that the theory is solid, the code could benefit from a round of dedicated restructuring and optimization.

External Links:

-