Goodai has followed this route in their quest to obtain general artificial intelligence
3 Likes
@Tofara_Moyo thanks, here-s their paper https://arxiv.org/pdf/1912.01513.pdf
That approach seems pretty well considered.
Unfortunately they don’t seem to share any code, or I haven’t found it.
2 Likes
Sorry, I don’t understand the language. What do you mean by ‘spawns’ or ‘string/tree’ of object represented by’ or ‘are used to decide’?
The essence of an agent theory is of multiple entities each running its own (or the same) algorithm(s) and interacting with each other and the world. Is that what you had in mind?
There is no reason to invoke quantum theory or superposition. I don’t think it helps.
1 Like
For the lack of a biological term, I used “spawn” to describe how HTM can mimic a set of neurons that continue to learn even though it has reached a stable state with respect to learning a set of inputs. For example, given sets of parallel learning neurons A and B, both reached a stable state that prefers input set X, A stops learning for time T, and B continues until it prefers another set of inputs. But in HTM, as in code-wise, this can be implemented by spawning a new thread of computation.
B is the new string of neurons. String as in a group or thread of neurons. Sorry, I’m using computer science terms.
I meant the set of neurons as agents that process a set of inputs. Their outputs are basically representations of objects and they can have varied outputs depending on how they are configured and the sequence of inputs they ingest.
Another agent or agents are using the other lower level agents to decide which ones better represent an object for example.
Why?
2 Likes