Inspiration
While more and more technology is integrated into our everyday life—via chatbots, voice assistants, and automated support features—one question lies relatively unscathed: Do people treat AI kindly, and does perceived AI empathy affect behavior? This project seeks to deconstruct human-AI interactions and determine if and how users interact with AI warmly, angrily, or indifferently, and how AI's responses influence the same.
What it does
The AI Empathy Test conducts human-like conversations with a simulated AI agent and examines user sentiment, tone, and response patterns. The major features are:
Real-Time Sentiment Analysis – Identifies user feelings and classifies responses as positive, negative, or neutral.
Adaptive AI Responses – The AI changes its tone according to sentiment detected, answering empathetically or neutrally based on conversation.
Behavior Tracking – Captures patterns of how users interact with AI across many interactions.
Data Visualization – Presents information about user behavior patterns, enabling researchers and developers to comprehend how humans interact with AI.
How we built it
Backend (FastAPI) – Processes API calls, sentiment analysis, and response generation.
Frontend (React & TailwindCSS) – Offers an interactive user interface for chatting with AI.
Gemini API – Renders responses based on user sentiment and conversation history.
Sentiment Analysis Model – Identifies user input to determine the classification of emotions and accordingly tune AI responses.
Data Logging & Visualization – Tracks and displays patterns in user-AI interactions.
Challenges we ran into
Making Natural Conversations – Tuning prompts so that AI replies seem natural and compelling.
Sarcasm & Rich Emotions Handling – Sentiment analysis found it difficult to pick up on sophisticated emotions such as sarcasm and passive-aggressiveness.
Patterns of User Behavior – Deciphering the various user demographics and their interaction patterns with AI necessitated round-about testing.
Processing in Real-Time – Refining API calls and response time to provide a seamless interaction.
Accomplishments that we're proud of
Successfully created an AI that adapts its responses based on user sentiment.
Designed an intuitive frontend that makes interactions feel natural.
Implemented behavior tracking to gain insights into how users treat AI.
Overcame challenges in sentiment detection, improving the AI’s emotional awareness.
What we learned
Empathy in AI is a multifaceted problem – More than sentiment analysis is needed to create responses that are human-sounding; context is key.
Individuals interact with AI in varying ways depending on perception – Others are friendly with it, while others probe it.
Emotion adaptation in real-time enhances interaction – Users reacted more positively when AI expressed sensitivity to their feelings.
Fine-tuning prompts is crucial – Minor adjustments to AI commands radically impact the user experience.
What's next for AI Empathy Test : How do people treat AI
Greater Emotional Intelligence – Making AI more able to comprehend advanced emotions beyond the level of basic sentiment analysis.
Conversational Memory – Leveraging long-term memory to generate more personalized interactions.
Comparative Studies – Compiling comparisons in user behavior in response to interacting with empathetic versus neutral AI.
Multimodal Inputs – Investigating voice-based interactions as a means for a more experiential interface.
Ethical Considerations – Studying how AI empathy may affect human relations and trust between humans and AI systems.
Log in or sign up for Devpost to join the conversation.