Ayu Batish
Research Mentor: Araceli Cruz
Mentor Department: Political Science, LSA
Author(s): Cesi Cruz, Ayu Batish
Session: Session 1 (9:00 AM – 9:50 AM)
Presentation Type: Poster 52
Abstract
The current political scene sees arguments without informational basis rampant, and our chatbot hopes to help individuals understand their own perspectives better. The chatbot addresses widespread misinformation, political polarization, and sociological division. While most chatbots focus on fact-checking and moderation, our chatbot relies on DeepSeek as a foundation with a custom system prompt to encourage conversations with feedback on thinking instead of the actual facts. Data was collected from three peers who interacted with our chatbot, reflecting on feedback of boredom and annoyance. Ten research papers on topics ranging from therapeutic AI to prompt engineering were collected in literature reviews to aid in the improvement of the system prompt based on the user feedback, pushing it to be more descriptive in its responses, less rigid, and more open. I continued to analyze various research papers for their implementations of tools such as Retrieval Augmented Generation for improved memory and the application of such tools to therapy-like chatbots, such as the Social AI chatbot centered in the project. Working with the team, we refined iterations of the system prompt. While the research is still ongoing, we have improved the chatbot’s prompting and responses. Now, the response gives answers longer than one sentence to encourage longer conversations without feeling awkward, and it deviates from the strict, methodical approach as needed to discourage boredom. We are continuing our research to help those with different native languages to encourage worldwide usage, especially in the Philippines. The intended goal of the research is to broaden horizons and decrease polarization and misinformation, and the chatbot’s conversations reflect good progress towards encouraging this goal.
