Is generative AI the future of online discussion and commenting? Let’s explore!
Author: Paul Allison
1 changes, most recent over 2 years ago
0 General Document comments
0 Sentence and Paragraph comments
0 Image and Video comments
What if we could make it easy for NowComment users to bring the power of generative AI (tools like ChatGPT) into their discussions of online documents? Imagine NowComment users deploying AI bots to generate comments and replies. A user might invite a psychology bot or an existential bot or a climate change bot to bring their perspective into a text-based online discussion, then reply to the generated comment with their own thoughts.
Users could register any number of AI bots which would be automatically associated with each creator’s name and profile (e.g. “A Feminist Bot by Paul” or “A Black Lives Matter Bot by Chris”…). These AI bots are software applications that would take a user’s description of an Afrofuturist theorist, for example, and generate Afrofuturist comments and replies in documents whenever called upon by an actual/human user.
Any logged-in user could create an AI bot by editing a bot profile page, which could look and function mostly the same as a regular user’s profile, with two significant additions, as discussed below. These bots could also be automatically collected into a repository for other users to use or remix. In the NowComment AI bot library, there could be anything from a predictable “TL;DR summarization Bot” to a mind-expanding “Afrofuturism Bot” and everything in between.
Although an AI bot is a software application, users could create these bots by composing a bio for each of them on special NowComment profile pages. If possible, an AI bot’s profile page would need two additional functions to what already exists for profiles.
-
First, users should be able to duplicate each other’s AI bot profiles then remix their copies.
- Second, AI bot profiles need to interact through an API with a Large Language Model (such as OpenAI’s GPT-3).
Paragraph 6 changes
Two texts from both a profile and from a document need to be transmitted in the form of a prompt to an AI model, then results from AI need to be relayed back through the AI bot and inserted into a comment or reply box to be published by the actual user and bot owner.
Actual users could create these AI-powered mock users, upload an image for them, and describe each bot’s perspective in the bio on that bot’s profile. Other AI-specific settings such as which model to use and how high to set the temperature for language use, and more… would be set in a new AI settings section of the bots’ profiles as well.
Users could experiment with writing bios that are simple or complex, straightforward or clever, and long or short. They could even learn how to adapt critical intellectual frameworks to compose specific natural language instructions for the AI to get the results they want from each bot.
The AI bots, the software application that we are proposing to develop here on NowComment, could, automatically and behind the scenes, take these bot bios and AI settings from the AI profile pages and turn them into coded prompts that would guide generative AI models on how to comment and reply. When discussing a document on NowComment, a user could select a paragraph or a sentence, then choose a specific bot to generate a comment or a reply.
Is this the future for online discussion and commenting that we want to create on NowComment? And can it be done? It seems possible because of what we have been doing on Youth Voices, a WordPress blog (see screenshot below). Our use of a plugin, AI Mojo, and an API from OpenAI brings the Playground with its GPT-3 LLM to sit next to an edit page on Youth Voices.
The plugin also has templates that have the functionality much like we imagine AI bot profiles could work on NowComment.
What does this proposal make you wish we could do? We need your help to write a specification for this project. Is there anything here that concerns you or that makes you feel a bit queasy? What seems strong about this plan and where do we need to think harder? Let us know in your comments.
General Document Comments 0
icons on the left to see existing comments.
0 archived comments