I’m currently using the Discourse Automation plugin and I’d like to know if it’s possible to trigger an AI-generated reply when a user creates a new topic or post. Is there any way to integrate an AI service (Discourse AI) with the plugin for this purpose?
I have a PR in progress to add this concept to personas, my main concern though is that this can add lot of noise to topics if we are not super careful.
Can you expand on your exact use case?
Our primary goal is to enhance the community experience by introducing dynamic and personalized AI interactions. We want to create AI personas like fortune tellers, scientists, engineers, athletes, and culinary experts, each capable of responding with insights relevant to their area of expertise. This would not only diversify the conversation but also provide users with more meaningful and engaging interactions based on the content they post, whether it’s a written topic or an artistic creation.
Given that Discourse AI already supports persona creation, would it be feasible to achieve this vision by enabling automatic responses from these personas when users create new topics or posts?
My English isn’t perfect, so I hope I’ve conveyed my ideas clearly.
Yes, automatic responses are nice option, but those work already that way using mentions. If a category creates an automatic respond it will never be quiet. If those topics are pure Q/A that will be wanted, but it will create a lot of noise too.
Otherwise, on meta-level, you have very romantic idea how capable AI is
Interesting…… we started some very simple tests with automating replies using the automation plug-in, but it feels limited as you need to be very specific (unless we are doing something wrong)
Being able to do more with AI, especially in our support categories could be extremely useful.
The key to building these systems is ensuring they will produce the value you are expecting.
The great news is that you don’t need to wait on anything at all to validate the idea.
Just create a persona in discourse-ai and prompt it to see if it is able to provide the value you are expecting it to provide.
You will need to gather a bunch of local examples and then create conversations with the persona to confirm it behaves as you expect it to.
The first problem to solve with the “prompt” with @joo’s use case would be triage
- Create a system prompt that “figures out” who should respond, that answers the name of the persona or “not response needed”
- Then pass in 20-50 examples to see if it performs the way you want to.
Once you validate this by hand then we can start looking at automating, there is not a lot of work needed to automate this kind of stuff, but the key is first proving that this would be valuable