back to "used by an intellect that lacks the capability to see a whole problem space at once."

Human senses and imagination have never had the capability to see the whole problem space. Our understanding of the physical world is very, very limited. Every 4 or 5 decades, it is turned upside down by new observations. The real science, that which expands human knowledge (in tiny increments) is driven by observation; as our techniques and technologies to observe improve we learn - largely we learn how wrong were our prior conclusions. The "laws of physics" are mere gentleman's agreements. Progress comes when we realize the agreement is obsolete and we need to update what we call 'laws'. Still we see only a tiny fraction of the problem space that is our existence and all we've envisioned around it. So how can we train another being that has no source of knowledge beyond our feeble and incomplete view to see more? It might be able to reason more quickly, even explore alternatives that human bigotry have closed off, if it is so trained. This could expand human knowledge for sure, but it will still be limited by observational capabilities. The AI can only work with the data available.

Where will the AI get the rest of the story? It will be limited by input just as humans are limited by our senses. Until we learned to enhance our senses, but still, we're limited. Will AI learn to enhance it's senses - design and deploy new sensors and instrumentation on it's own? Can an AI be trained to understand the nature of creation? Can it "think outside the box" or challenge the need for boxes in which to think?

Fun things to ponder ;-)