4 ms·

It makes me wonder, what if the chat bot were able to browse for information, in addition to using table look-ups or supervised training? This would allow the b

It makes me wonder, what if the chat bot were able to browse for information, in addition to using table look-ups or supervised training? This would allow the bot to be flexible in unfamiliar situations, or ones where the company policy has changed. What AI's are there that attempt information gathering and understanding? All I can think of is IBM Watson on Jeopardy.

yorwba 9y ago
If you really wanted to get rid of human customer support agents, you could just put their internal manual on the open internet. After all, nobody could ever remember all the various policies within a company, so they have to be written down somewhere. Of course, if the rules say "if the customer asks for a refund, stall for 5 minutes and then transfer to a different department, unless they threaten with legal action", you can't really do that. So I'd guess that most companies won't want to expose themselves that way.
- robk 9y ago
  This. I feel all help would already be self service if it weren't for fraud vectors like this that would be exploited en masse if publicly exposed. Redditors would devour these scenarios.
cjhanks 9y ago
There are a quite a few now. The barrier to entry for natural language processing has been lowered to a point where there has been a proliferation of chat-bots attempting to replace people. These AI models are learned by scanning the corpus of FAQ's and customer service training materials for a specific company. So far, in my experience... they don't work all that well. Although to be fair, it might be entirely psychological. When you're speaking with a human, you feel as if there is a possibility for improved understanding. When you're talking to a robot which doesn't understand you, that's a pretty hard barrier.
coldtea 9y ago
>It makes me wonder, what if the chat bot were able to browse for information, in addition to using table look-ups or supervised training? This would allow the bot to be flexible in unfamiliar situations, or ones where the company policy has changed. That requires actual intelligence to be able to interpret what it browsers -- which we are nowhere near close.

yorwba 9y ago

If you really wanted to get rid of human customer support agents, you could just put their internal manual on the open internet. After all, nobody could ever remember all the various policies within a company, so they have to be written down somewhere. Of course, if the rules say "if the customer asks for a refund, stall for 5 minutes and then transfer to a different department, unless they threaten with legal action", you can't really do that. So I'd guess that most companies won't want to expose themselves that way.

robk 9y ago
This. I feel all help would already be self service if it weren't for fraud vectors like this that would be exploited en masse if publicly exposed. Redditors would devour these scenarios.

cjhanks 9y ago

There are a quite a few now. The barrier to entry for natural language processing has been lowered to a point where there has been a proliferation of chat-bots attempting to replace people. These AI models are learned by scanning the corpus of FAQ's and customer service training materials for a specific company. So far, in my experience... they don't work all that well. Although to be fair, it might be entirely psychological. When you're speaking with a human, you feel as if there is a possibility for improved understanding. When you're talking to a robot which doesn't understand you, that's a pretty hard barrier.

coldtea 9y ago

>It makes me wonder, what if the chat bot were able to browse for information, in addition to using table look-ups or supervised training? This would allow the bot to be flexible in unfamiliar situations, or ones where the company policy has changed. That requires actual intelligence to be able to interpret what it browsers -- which we are nowhere near close.