Pure intelligence is experimental manipulation (Machine Learning) similar to lock picking it is solving a problem. Learns from past mistakes, effectively cause and effect and probability. The limit is the items malleable by the dexterous utilities. Learning creates new toolsets that combine with each other to make meta toolsets. This way it knows different strategies will require different thinking tools. This is called bounded rationality. Given computational resources what problems can I solve and in what order of difficulty?
In the real world, it is harder to solve problems because most problems are general. Humans have bodies and must grow in a world meant for them. From birth to early adulthood you must treat it right and teach it how to learn to parent and so on until they have self-control / autonomy.
The middle path of not too hard not too easy. This is difficult because it starts at birth. The right environment and the right parenting in simulation. No one knows how to be a perfect parent. A.I. must get an o.k. parent. and an o.k. childhood. A human body is limiting but it is the best we can accompany the a.i. safety problem. No robot body that feels touch will be available for decades. Simulated A.I. will need virtual touch.
Information on the brain is known for the bounded rationality of the brain as a probability tree. All the parts are there they just need to be put together and taught like a child.