Help shape the future: AI in Cisco Community

[dbrittin]

This is a thread for public commentary on the AI experiment announced here. We’re interested in your thoughts about this project! Community managers will be following this conversation and making sure your input gets to the engineering team.  

Please remember there may be differing opinions on this subject, and that is okay. Constructive dialogue is always encouraged. 

[bill.king1]

Hi, I haven't seen many but the few that I have seen in the contact center discussion group have been less than helpful.
Here's an example of one of the posts https://community.cisco.com/t5/contact-center/renaming-the-deployed-application/m-p/4802676#M126545
Another one was here https://community.cisco.com/t5/contact-center/skills-on-my-agent-removed-how-to-find-who-did-it/m-p/4804749#M126565
but the second one may have been a different user, it looks like it was removed.

@david.macias / @janinegraves , I don't know if you've seen others?

[david.macias]

I can only think of two examples where it was obvious someone just asked ChatGPT and pasted the response without looking it over. I would hope that the Cisco Virtual Engineer is using internal Cisco information to provide better answers, right? I do think this is cool and it makes a tone of sense, but I hope it's an LLM very much internally created for the specific technology.

david

[Tyler Langston]

Hi @david.macias - yes, the Cisco Virtual Engineer is using a broad data set of Cisco data (though nothing that we would consider personally identifiable information), and is having all its generated responses vetted by an actual human engineer prior to publication.

[Tyler Langston]

Howdy @bill.king1 - thanks for sharing! Clarifications:

Your first link is for an earlier iteration of this experiment. We removed all answers provided by that bot that were not 'liked' or marked as the solution by the original poster - so nearly all of them. 

Your second link actually isn't from a Cisco team, so I'll look into that one further. 

[dbrittin]

Hey @bill.king, yes there were a few "rogue" posts early on in forums that the bot wasn't that great at. If they didn't have a Helpful vote, we removed them.  Since that time, we've also restricted the bot to only 4 forums for technology areas that it has proved the most accurate for AND all bot posts are being vetted by a TAC engineer before posting.

Take care,

Denise

[Richard Pidcock]

I would have concerns with AI generating responses for community forum questions.  I feel like AI can be unduly influenced (i.e. adaptive learning that may be incorrect).   For example, if we all start feeding AI with yoda type information (i.e. do or do not, there is not try) what kind of results is AI going to generate for us.  

[Tyler Langston]

Hi @Richard Pidcock - I think that's a fair point! Part of machine learning is making sure that it ingests proper, factual information and isn't pulled off course. That's one of the reasons why we've been working with the team involved to make sure that the scope of this experiment is narrow and focused on the technologies that the bot 'knows' best:  Network Security, Switching, VPN, and Unified Computing Systems.