Claude Moore wrote:Thanks a lot for your detailed answer. I am very excited about what you said about the superscalar speed up you can achieve on algorithms using QC:l can't stop thinking about what we could achieve in the field of Neural networks, and in AI in general...
There's an interesting aspect concerning AI which hasn't really been dealt with yet: Qubits can't be copied. They can be moved and transformed, but there's physically no way to duplicate even one.
So...
warning: far-future sci-fi speculation (just for fun)
Suppose (with your far-future-practically-scifi hat firmly in place), suppose you have a complicated quantum machine-learning system, and it took you a year to train it to do a job really well (maybe it can win championship-level games of chess and ping pong simultaneously), and its entire learned state is maintained in fault-tolerant 4 giga-qubit storage. Someone sees your AI demo, and wants to buy 100 of them.
Qubits physically cannot be copied, so you can't just save the state and stamp out more just like it, you'll have to train each one and hope it comes out the same. There's no way to back it up, so if the power goes down you'll need to start over. And if over time (and after more input) it gets worse at its job, you'll also have to start over.
Thanks to Nic, Chapter 13 in the book covers Quantum Machine Learning!
It's all very early and experimental; QML systems in the near future will be trained in a few microseconds, use the knowledge, and lose it all a few microseconds later.