In the days of yore, we had first level operators who got an initial description of what the caller was talking about and routed calls. A handful of companies still do that; Barracuda is my go-to example, and their phone support is one of the best parts of their service.
Then, we got menus - sales, press 1, support, press 2, billing, press 3, etc. Not great, but it helped weed out the support calls from the billing questions.
Then, we got voice prompts, where we *said* 'sales', 'support', or 'billing'...and that's when things got messy. For starters, the always-listening system mistook traffic for a person speaking, giving "I'm sorry, I didn't get that" vibes, and made navigating the menu take twice as long.
And then, it continued to get worse, with the "in a few words, tell me what you're calling about". It got even worse, because it's like getting to a bash prompt for the first time, with no 'help' or '?' option...so now we had to distill the description of a problem into a few words, hoping one of them is a keyword...God help you if the issue is "I can't get the app to show me my current balance" - obviously a support issue, but "current balance" is more likely to be a keyword to send to the billing department. Oh, and systems vary as to whether they'll listen the whole time, or if they'll ignore you until they've given their whole spiel. Frequently, with long annoucements that aren't relevant to the situation at hand.
Also, there's a special place in hell for whoever decided to inject advertising into hold music.
As a counterbalance, I *will* give some credit to my cable company, who really went out of their way to make the automated functions actually-helpful. It detects the account based on the incoming phone number, checks for outages in the area, and can reboot the modem and do a connectivity test right from the IVR. Does it take six minutes to get into the queue? Yes. Is that annoying? Yes. Can I appreciate that "reboot the modem and router and do a connectivity test" solves the majority of technical issues for the majority of people, and that streamlining the process to do that is helpful for both the ISP and the customer? Yes, I can.
So, let's move the football down the field and discuss the AI element...In certain areas, it probably *could* be helpful. Tier 1 tech support is probably a great application for it. At our office, we take turns calling Intuit for support, because they seem to be trained in being infuriating, and even our lowest tier techs don't deserve that kind of torture. Would I rather talk to an AI when calling Intuit for support? Yes.
However, I could see areas where this would be bad. Insurance carriers would be my perfect example of this - there's a *need* for both human judgment and accountability when dealing with insurance claims. Having each statement from the phone system conclude with a paragraph-long ChatGPT disclaimer would be insufferable, and they'd all amount to "I'm just a chatbot, you need to verify this information with a CSR..." "then let me verify this information with a CSR" "Before you verify this information with a CSR......", it'd easily devolve into being unproductive...but if 'fool the AI' is the name of the game, some enterprising troublemakers will get the phone system to agree to do some massive payout, which will then make it even more impossible for end users to get their claims sorted properly.
Ultimately, there are indeed places where an AI system can be helpful in a phone system. If the goal is for it to be helpful, I do think it can be. However, if the phone system is intended to be a barrier to customer service, rather than an enabler of it, AI will look great this quarter, and terrible every quarter.
Of course, things will get *really* interesting when an enterprising developer with a grudge and a GPU cluster gets so pissed that he writes his own AI who can call customer service with the express intent of doing what he wants...six hours over the phone and ultimately finding an exploit? Sounds like an unstoppable force meeting an immovable object...