From what I've seen of AI so far, it's utter rubbish. We have it at work for certain very routine operations, and it was initially touted as the best thing since sliced bread, since it would supposedly take away the need for us to do most of those humdrum repetitive tasks, and we'd be able to reduce staff. Well guess what, it can only operate succesfully between two very narrow parameters of simplicity. Anything even a tiny bit out of the ordinary, leads to that now all too familar note in the case diaries: "robotics failed". Scans of these are being dished out to staff for them to clear manually.
With any sort of chat function, such as on a bank website, the "Hello, I'm Robbie the robot, how can I help today" type "assistance", can only ever handle the very simplest of enquiries. Most queries a customer has, it cannot even hope to comprehend. So inevitably you get transferred to a live agent.
As far as algorithms, if they're anything like the ones on facebook, for example, they get stuck on the difference between a legitimate dictionary word used in its proper sense, and the same word as racial abuse, for example. Using the term "a chink in the armour" gave me a 3 day ban for race hate - the offending word being "chink". It's pathetic.
I dread to think what might happen if TPG is given over to AI.