ABOUT BIT BOT PROJECT
Abstract
This project is a proof of concept exploring a user interaction based context AI chatbot. The ai chatbot responds with reasonable levels of natural language chat through basic model fine tuning. There is opportunity to further fine tune the model for specialized interaction applications.
Project Details
The goal was to use an ai chatbot in a new method. As some of the options, this project showcases ui elements including button click, hover, menu change. The ai will respond to each action and the context of the action. This type of ai chatbot application could be described as an interaction context sensitive ai chatbot. This opens a new aspect of user experience as the user interacts with the ui. Receiving feedback during the ui interactions. Instead of interaction strictly through text chat input.
Methodology
- Built with: React Framework NextJS, Natural Language Processing Tool gpt-3.5-turbo from OpenAI, CSS framework Bulma
- Logic: UI element pass events containing action and contexts to the backend api, passing along to the OpenAI api, and finally returning the response.
- Model Tuning: Several prompt iterations were tested. One notable finding was strongly delineated prompt parameters (via colon per parameter) caused the responses to mirror that and also become delinated. Which broke the natural language effect of the responses. There was more successing in using natural language in the prompt to get a return of natural language.
Other use cases
This proof of concept applied the interaction context sensitive ai chatbot as a data hungry website, reacting to user interaction. Other use cases could easily be explored, such as:
- e-commerce: Upon browsing many similiar items (action: hover image or link pattern), offer suggestions or further details of item type.
- business website: When viewing services, prompt the user if they are looking for a particular service and guide to the information
- video games: 4th wall breaking NPCs responding to user interface actions
- specialized programs: Upon multiple tooltip hovers, prompt to ask what tool the user is looking for
Drawbacks
There is a big risk of creating friction or annoyance in the user experience, as unwanted information and text can cause frustration. If the user already knows how and what they want to do, obtrusive messages from an ai chatbot will likely cause dissatisfaction.
Conclusion
There is potential improvements to UX with this interaction context sensitive ai chatbot application. But a measured approach should be taken to find the right application for this tool.