Context
In Brown University’s CSCI 1951C: Designing Humanity-Centered Technologies, I explored near-future scenarios where intelligent systems—akin to today’s Large Language Model-powered products and services—become deeply integrated into our lives. Framing the project within this context of a rapidly evolving technological landscape sparked central questions:
With the accelerating advancement of Artificial Intelligence, how can we design technologies that protect what matters most to us—our privacy, emotions, memories, and our sense of self?
How can intelligent systems create moments of introspection, enabling us to reconnect with ourselves in an increasingly automated world?
For Project SeWol, I designed and developed an intelligent lamp that listens to users within an indoor space and visualizes the accumulation of emotions inspired by the Korean notion of 기운 (Giun) or 気 (Ki)—a subtle energy that permeates spaces, shaped by the actions, events, and emotions that unfold within them.
Team
Dave (🙋🏻)
Ming Dong
Timeline
a 3 week project, Nov 2024
Skills / Tools
Prototyping, Raspberry Pi, Interaction Design, Design Thinking, Sentiment Analysis
Project Ideation 💡
The project began with the idea of using technology to augment human experiences and highlight our intrinsic characteristics. Specifically, I focused on displaying emotions as entities that can accumulate within a space—or, more accurately, as reflections of the emotional build-up within ourselves—in a way that inspires introspection and self-reflection.
The images below showcase the initial ideations and form factor explorations!
Affective Computing
Understanding Our Emotions
Understanding human emotions is a complex and interdisciplinary challenge. Affective computing systems aim to achieve high accuracy in emotion detection or sentiment analysis by leveraging machine learning techniques across diverse data inputs.
When it comes to representing emotions, two primary approaches exist—discrete and dimensional emotion models.
Design
Functional Prototype
The goal of the functional prototype was to create a working model that could facilitate observing user interactions with the system as it performed similar—but simplified—tasks. For the prototype, Raspberry Pi 4 B was used to perform the input computation.
The functional prototype primarily utilized its microphone to capture conversation content, performing sentiment analysis on the transcribed text. A detailed schema of the pipeline is displayed below.
Final Prototype
The final prototype classifies sentiment into three discrete categories—negative, neutral, and positive—and uses corresponding colors to visualize the accumulation of emotions:
Red: Negative
Blue: Positive
White: Neutral
While the intended design featured subtle shades of red and blue, more vibrant tones were used in the demo for clarity and emphasis.
Understanding Our Emotions
Developing Multimodal Algorithm
Work in Progress!
I am currently enhancing the algorithmic aspects of the prototype to advance the complexity of the emotion classification model. This involves shifting from simple text-based sentiment analysis using the OpenAI API to a multimodal approach that incorporates inputs from voice, text, and heart rate readings.