Two ideas about artificial general intelligence I had:
Ideas could be evaluated based on their conformance to real world observations, those thoughts that are reasoning (evolved from other thoughts) and not directly observed can be evaluated on the similarity to what has been observed. Real world observations or ideas with alot of supporting evidence would have a tendency to replicate strongly linked nodes – the replicated nodes would undergo mutation of held data and weight strengths for links to the same nodes that the parent is linked to.
A thought or idea is evaluated from a basal node, but to evaluate it one must evaluate links from the basal node to other nodes. There would be a time\hop limit to how many evaluations occur, and different problems could require varying limits. Perhaps periods of rest/sleep or contemplation would set this very high, while moments requiring direct action would have a small limit.
It might be an interesting idea to use spatial aspects in simulating the mind. All nodes having spatial attributes with their distance to other nodes influencing the evaluation time if they are linked.
A node that is evaluated as conforming to reality or the mind wants to reinforce has its neighbouring nodes pulled closer, the strength of pull is based on the weights of the links. Ideas/nodes that are not useful or are not used disperse by brownian motion – and with sufficient distance the link between nodes becomes weak and breaks. A free floating node is removed, or alternatively it remains floating around until it randomly bumps into another node and forms an adhoc short distance but very weak link.
From this mental focus could be developed. Computation and node evaluations would occur near the spatial “focus point of the mind”. This is a spatial location representing where the mind looks for new things to process and evaluate. After evaluation any node would have a refractory period where it couldn’t immediately be evaluated again. In the human brain our neurons fire every ~5ms, and many people will be thinking “But the whole benefit of thought in silico is that electronic circuits are so much faster!”, and I don’t think that we should necessarily make the refractory period as long as 5ms. I do however think temporal and spatial patterning are important in our brains and the 5ms delay might be necessary for human intelligence (although this doesn’t preclude other forms of intelligence). Think reaction-diffusion equations, but much more complex.
This also lends itself to having more than one thread or CPU running. Each could be assigned a specific region of the mind to work in such that some will be doing mind maintenance tasks, handling locomotive ability, or communication.