There is some very interesting research in recent years that offers some insight into why this is. The nuts and bolts of it is that tractable intelligence for a construct of our size can only exist in an exceedingly tiny part of the algorithm phase space, despite its extraordinary value from an evolutionary standpoint. Ideal intelligence is extremely intractable, and usable approximations make a very narrow target. The basic components are pretty simple theoretically, but there is relatively little wiggle room for implementation.
From an evolutionary standpoint, once the basic component was discovered it was exploited like mad in nature until it eventually produced humans despite its high cost in biology. The evolution of intelligence in animals was extremely quick on the timeline as such things go. Ideal intelligence has some severe limitations in the mathematics and the human approximation, while not as efficient as it could be, is a reasonable facsimile.
Intelligence is not an exceptionally complex thing, but one has to hit pretty close to the target or it really does not work at all as a practical matter for esoteric theoretical reasons.
Interesting...any links to hand?