I think an optimized robot would find it advantageous to retain the ability to extract nutrients and energy from biomatter instead of relying solely on electricity. Robot bodies might retain some organic parts for a digestive tract, or they might have digestive tracts made of soft, synthetic tissues that did the same things as human cells.
The first human-level AI we create will need supercomputer hardware that is vastly more powerful than a human brain. This would be in keeping with the pattern of the first examples of any new technology to be inefficient and barely functional. Consider steam engines. The first, commercially successful one was the Neucomen Engine, and it was huge, inelegant, and had a terrible energy efficiency of 0.5%.
However, it’s also a truism that technological efficiency improves over time. So someday, an AGI with human levels of intelligence will fit on a portable device like a laptop or something even smaller, like a smartphone, and they will be cheap. It’s hard to contemplate a place for humans like us in a world where intelligence is so ubiquitous that it can be thrown in the trash.
Millions of years of evolution have equipped the human brain with an exquisite ability to recognize human faces and their smallest details since we are social animals who must quickly “read” the people around us for the sake of survival. However, that ability only weakly carries over to other species, and many of them “all look the same” to us (think of squirrels or seagulls). Of course, those animals are able to tell each other apart and to recognize subtle indicators of emotion and intent, so there must be a way. Better AI is the obvious solution to this problem. Someday, machines will be able to recognize individual animals as well as they recognize humans, and to understand and communicate with animals much better than we can.
Thanks to satellites and huge numbers of robots, someday our maps of the Earth might be so comprehensive that even individual trees in forests will be cataloged and counted.