I feel a cognitive dissonance regarding this statement:
> Enumerate all of our beliefs and check whether the new piece of information contradicts them. This is not possible. As people - and as complex reasoning entities in general- we can not list all of our beliefs. And if we could, we would not have the time to check them individually.
Feels like it is built on the assumption that all beliefs should be stored in some unstructured fashion (aka array) → you need to enumerate all of them in O(N) time to check for contradiction. While complex entities may have more sophisticated ways of storing and retrieving beliefs (aka hash maps, similarity search, associative memories etc.). So in theory retrieving similar/or contradicting beliefs could be performed in ~constant time O(1).
For example, when human sees some new information (key), only beliefs that confirms the evidence and beliefs that contradicts it (negation of key) could be retrieved via associative memory.
> complex entities may have more sophisticated ways of storing and retrieving beliefs
For similar reasons to why they can not list all beliefs, they also can not list all the structures they used and ensure their soundness.
A self-reflexive coherent super intelligence might exist one day. If it ever does, I expect it will be much simpler than LLMs / people / group. Similar to how relational databases, logic, Bayes and the like are simple.
With set theory you can explicitly acknowledge known and unknown uncertainty, which is a massive improvement over current delusional/deceitful approaches. It adds a few seconds of overhead in communication, but that isn't the reason we do not use it.
I feel a cognitive dissonance regarding this statement:
> Enumerate all of our beliefs and check whether the new piece of information contradicts them. This is not possible. As people - and as complex reasoning entities in general- we can not list all of our beliefs. And if we could, we would not have the time to check them individually.
Feels like it is built on the assumption that all beliefs should be stored in some unstructured fashion (aka array) → you need to enumerate all of them in O(N) time to check for contradiction. While complex entities may have more sophisticated ways of storing and retrieving beliefs (aka hash maps, similarity search, associative memories etc.). So in theory retrieving similar/or contradicting beliefs could be performed in ~constant time O(1).
For example, when human sees some new information (key), only beliefs that confirms the evidence and beliefs that contradicts it (negation of key) could be retrieved via associative memory.
> complex entities may have more sophisticated ways of storing and retrieving beliefs
For similar reasons to why they can not list all beliefs, they also can not list all the structures they used and ensure their soundness.
A self-reflexive coherent super intelligence might exist one day. If it ever does, I expect it will be much simpler than LLMs / people / group. Similar to how relational databases, logic, Bayes and the like are simple.
With set theory you can explicitly acknowledge known and unknown uncertainty, which is a massive improvement over current delusional/deceitful approaches. It adds a few seconds of overhead in communication, but that isn't the reason we do not use it.