Creative thinking.
Intuition.
Instinct for survival.
A great love of fish. (Just Joking. 
)
Tell you about their mother.
Pee.
Ethics.
Without a ‘conscience’ I would see it difficult for androids to make ethical decisions beyond a choice of numbers:
Case in point: Assuming your android can ‘eat’, it may steal the last slice of pizza, even though it knows that someone else didn’t get their last slice–which is just plain wrong. Even worse, it may drink the last beer in the fridge at someone else’ house. Go a little more real-world, and it may decide the best course of action is armed conflict even though diplomacy hasn’t had a chance to run it’s course yet.
Tripler
Come to think of it, I must hang around a lot of androids. Damned drunken robots!! :mad:
Perhaps humans are still needed as systems analysts and debuggers. That is, occasionally the AIs fall into recursive error loops in which they increasingly fail to recognize how faulty their thinking has become. (Think HAL in 2001). Maybe humans are still entrusted to hit the master “Reboot” switch; or failing that, go down to the main computer core with a stone ax.
Well, there’s something John Q. Robot can’t do.
Well, unless it’s built in Japan, I add as an afterthought.
Safely walk between giant magnets.
I was thinking about this earlier: Companies inundate the cheap labor market with so damn many, that the homeless here (and from other countries) start rallying for either better working conditions, or actually riot. Hell, they could even attempt a coup. Shoot, inundating the workplace with cheap labor would more or less stop the flow of immigrants from one place to another if there’s no work there. . .
Tripler
Just a plot thought. . .
Sweat. It could be a really hot environment, and it would be too costly to provide air conditioning to keep the droids running smoothly… but water is cheap, and the humans can work in there for hours on end.
Darn, I was going to say that. “Did you know that you’re leaking coolant at an alarming rate?”
But you didn’t say poo. I was gonna say just “go to the bathroom.” Androids can’t pee or poo. Or spit. Or blow their nose. Or masturbate.
So maybe you’re generous in your assumptions about AI, but I think the flexibility of the human mind in underappreciated in sci fi.
since the body is self maintaining up to a point, and adaptable it might be economical to use us instead of building androids for each environment.
novelty, will a well designed, “successful” android ever be motivated to improve on anything?
dare I say well adjusted? I think I dare
Use contractions. 
There’s no reason those safeguards can’t be programmed into the AI.
You need to look at what humans have always used machines to do - boring, dangerous, tedious or quantitatively complex work. Extrapolating that out, future humans would probably spend most of their time pursuing creative or artistic work or trying to stave off boredom (which could be a central theme of your story).
There does not need to be a “purpose” for humans to exist. They exist by default because they didn’t wipe themselves out. What you need to do think about why the humans need AI.
AI does not and should not just take the form of a humanoid robot. It can take the form of a camera that automatically adjusts its settings or an iron that can adjust to the fabric. More advanced AI could augment medical, legal or business teams - providing analysis, advise and instant research while the humans do the actual decision making. Those AI might look more humanoid since they would be required to work closely with humans in their environment.
And of course you would want to consider everything between human and pure AI - humans cybernetically or bionically enhanced to some degree.
Are machines or AI devices capable of creative thinking or originality? Another way to look at it is to ask what humans do NOW, in the real world? Basically, they try to survive as best they can and better their lot. They would try to do the same in any scenario, no matter the surrounding circumstances.
Hey, that’s not a bad idea. Maybe the mines for unobtanium are subjected to severe magnetic stresses or some sort of electromagnetic interference or something.
Actually, there is. Turing’s Theorem shows that a computer cannot, in general, tell when it’s gone into an infinite loop. There is much debate (a Great one, in fact) about whether this limitation also applies to human brains, but at the very least, one can say that it’s very difficult to send a human into an infinite loop. So the non-looping humans can break out the looping AIs.
I’m with MrSmith: humans would be doing the interesting jobs that they want to do. Research (basic and applied), managing (the robot workforce and other human workers), teaching, making art, doing crafts (I think ‘handmade’ would still be a valued tag), negotiating and doing political things, etc.
I mean, few humans today are ditchdiggers weilding shovels on large construction projects – that job has already been automated. Even when backhoe drivers are automated, there will still be engineers overseeing the work, project managers, backhoe designers, backhoe salespeople, project permitting attorneys, etc.
Dream of electric sheep.