I think part of the paragraph I read the AI bit in mentioned that a learning AI would be possible simply because of the way QCs work.
QC's read and write to atoms. This means 1 bit can do thousands of operations in nanoseconds. Multiply that by 1 million and you have virtually limitless processing power. That easily supersedes the brain, then on-top of that have virtually limitless memory for everything it learns. Or so it can store everything it ever does.
Now give it a program to start off learning, give it the ability to detect mistakes it's made in similar situations it's encountered before and then try again until it's successful. You could teach it basic things this way, until it builds up a history of tasks and ways to accomplish said tasks.
Though I'm just speaking off the top of my head, that's the basic way a learning chess computer works, only this could do things much faster, and remember every single thing it does.
I'm sure given enough time, an efficient program that learns would be developed. One that can adapt to various tasks. And potentially read, write and delete it's own programming, further developing it's capabilities.
