My Ideas

[Tom D]

Hi all,

Like the founder, I have also been pondering how to implement AI for a long time. Here is my idea as it stands in my head.

My implementation centres around a semantic hypergraph; similar to that of the OpenCog project. The idea is to store everything in this graph (with each mutually exclusive concept/object referenced by a separate node) and operate on it using agents (which themself is programmed into the graph). These agents are responsible for reasoning.

The strength of the model presumably from the way relations in the graph are indicated. Whereas openCog has a large, ever-growing list of relationships, relationships between entities are nodes themselves. This allows relationships which cannot be described (due to lack of information or otherwise) to be represented, and by examining the neighbouring nodes and their attributes and their nodes, similarities can be detected, clustered, and potentially merged.

This approach is in lieu of a neural network approach, as I believe neural networks are too fine-grained and have too many nodes to be practical.