To say that the current level of human intelligence is optimal is to oversimplify. True, we could have evolved to be even smarter, but did not: we reached an equilibrium level, an evolutionary position of greatest adavantage as Eliezer Yudkowsky correctly notes.

But our average level of intelligence reached an equilibrium that was optimal for a neolithic hunter-gatherer. Our current world is vastly more complex and demanding than that. Intelligence is a greater advantage than ever before.

But there is a larger issue than that. You cannot decide optimal intelligence by only looking at the benefits - we would surely be more intelligent if it came for free. We have optimal value for money brains. And they don't come cheap: see the significant costs of the large human brain.

It is not that being more intelligent would be detrimental, but that that the additional costs would not be worth it.

Comparing us to other animals, we have hypertropied brains. We have pushed the envelope until we have run up against the bioeconomical limits of brain growth. We have reached the biological point of dimishing returns and exponentially growing costs.

Note that some of those costs do not apply to modern humans any more. For instance, few people in the civilised world are short of food. Caesarean sections could get around the birthing limitations. There is enough time and resources to care for children.

Few or none of these costs should apply to an AI.