Nice sentiment, but are you using chatbots now and having a positive experience with them?
But it's not the bots are bad.
The bot experience = $_investment * tech efficiency.
I've been in the chatbot industry for the last 6 years.
If you disable event past-gen NLP bots, millions of people will never be able to talk to a human agent or get any support at all. Unless companies invests 10x-100x more in customer care.
Even companies like Apple, in a premium segment of margins, use community forums and bots to manage their level of support.
LLM bots will bring us to a different reality when you can have 100x better customer support with the same level of investments.
And btw, that's why google is also moving this way.
I'm souring quickly on the concept
I plopped for the chatbot, then I had a problem. My apartment is wired weirdly, long story. The chatbot was very clear that it was a bot, it worked well, and it only took a few minutes of chatting before it decided to escalate to human support. The human support called me, so I didn't have to pay any phone bill, and they did it within a few minutes of bot making the decision to escalate.
All in all it was pretty satisfying. Comcast's version sounds ... less satisfying. The arXiv paper is really the cherry on top.