AI versus simulated Intelligence
We have made no progress at all on AI since 1946, all progress has been narrow areas of simulation and so called "Expert Systems". Much language in the AI field has been redefined. Computer Neural Networks are nothing of the sort. "Learning" and "Adaptive" is nearly a misuse of English. Successful language translation has almost abandoned grammar / parsing / semantics / context to use a brute force "Rosetta Stone" approach.
Speech recognition is nowhere near as good as Audio Typist, never mind a personal assistant. It just needs less "training" that it used to.
We don't yet have an adequate definition of natural Intelligence, so how can we define the program requirements much less write one? True natural language interaction rather than the Artificial Stupidity of phone response "robots", Siri and Cortana is very far away and probably needs true AI. It's not a question of computer power, or it could be done slowly.
Neal Stephenson in "The Diamond Age" asks is in fact real AI even possible with a "turing machine" (i.e. ANY computer). The "book" in the story certainly isn't possible today [ignoring issues of communication and charging] without even a real time team of humans, rather than the one person. The hardware and software of the "Book" is certainly feasible, though it's more like something implemented with eInk plastic paper than OLED or LCD.
In a way Project Xanadu is the first attempt at the "book" and while earlier than Apple's Hypercard (before HTTP/WEB) and HTTP/Web tries to solve some basic limitations of web pages.
The hardware of C3PO or R2D2 was possible even in 1977. Though power supply and balance for C3PO a problem then, now solved by e.g. Honda's robot (though I suspect power / running time is an issue).