According to me, just like any other machine, AI is also designed for solving a specific problem. Like driverless cars, Alexa, Alpha Go, etc., are all good at solving some specific problem. By One AI, you are referring to a machine that combines all forms of intelligence, I presume. Maybe something that understands everything like a human, but is better at everything? Although, such a machine doesn’t seem viable in near future, but if it is then it should be having a specific purpose.
Assume that you made such a machine that is super intelligent but has no specific purpose. What will it do? Make a purpose of its own, I guess. Now, its new found purpose may turn it into a great humanitarian or the source of human extinction. This is a chance we are not ready to take, or else we may end up in the Matrix.
Which is why an AI must be limited by its purpose, and if you wanna make something that is not, go make babies!!
As to why we can never entirely give up the concept of such an AI, I think because that will be true pinnacle of human intelligence, something that every human aspires for.
