TLDR: This research introduces a system that allows users to design new social media spaces using spatial metaphors (e.g., ‘a personal bar’). The system translates these metaphors into platform features and generates interactive simulations with AI agents. A study showed that users easily conceptualized distinct ‘open’ and ‘closed’ spaces, with the system effectively capturing many intended social dynamics, though some nuances in features like recommendations and networking proved challenging. The work highlights metaphor-driven design as a promising way to create more diverse and user-centric online environments.
Social media platforms have become central to our daily lives, yet their designs often prioritize engagement and scale over meaningful human interaction. This narrow focus has led to a limited range of online experiences, often failing to support the diverse ways people wish to connect and interact. Researchers have long sought alternative visions for online spaces, but prototyping these ideas within existing platform constraints has proven difficult.
A recent research paper, Prototyping Digital Social Spaces through Metaphor-Driven Design: Translating Spatial Concepts into an Interactive Social Simulation, introduces an innovative system that uses metaphors to help users imagine and explore new social media environments. Authored by Yoojin Hong, Martina Di Paola, Braahmi Padmakumar, Hwi Joon Lee, Mahnoor Shafiq, and Joseph Seering, this work proposes a novel approach to expand the design possibilities for future social platforms.
The Metaphor-Driven Design System
The core of this system lies in its ability to translate users’ spatial metaphors into structured platform features and then generate interactive simulations populated with AI-driven agents. Instead of starting with technical specifications, users begin by describing a desired social space using a familiar metaphor, such as “a personal bar where people share their daily lives” or “a ballpark with an energetic, cheerful, and joyful atmosphere.”
The system then processes this metaphor through a three-phase pipeline:
- Metaphor to Social Attributes: The initial metaphor is expanded into eight distinct social attributes. These include the atmosphere (e.g., intimate, vibrant), reason for gathering (e.g., shared interests, friends), connection style (e.g., spontaneous, routine), duration of participation (e.g., drop-in, long-term), communication style (e.g., dialogue, broadcasting), identity type (e.g., public, pseudonymous), interaction goal (e.g., relationship-building, informational), and control over participation or visibility. This step helps clarify the user’s intent before translating it into concrete features.
- Attributes to Platform Features: These eight social attributes are then mapped to a comprehensive set of social media features. These features are categorized into a three-level taxonomy: Network Structure (e.g., feed-based vs. chat-based timelines, chronological vs. algorithmic content order), Core Interaction Mechanisms (e.g., nested vs. flat comments, like-only vs. expanded reactions, real-name vs. anonymous identity), and Advanced Features & Customization (e.g., ephemeral content, public vs. private visibility, topic-based vs. popularity-based recommendations). This modular approach allows for diverse platform configurations.
- Generation of Interactive Simulation: Finally, an interactive social media environment is generated based on the selected features. This environment is populated by LLM-driven agents, each initialized with a social role (e.g., Influencer, Lurker, Moderator) and behavioral traits (activity, engagement, updating behaviors). These agents simulate realistic user interactions, allowing participants to experience and evaluate how well the generated space aligns with their envisioned values and social expectations.
Exploring Open and Closed Social Spaces
To evaluate this metaphor-driven approach, the researchers conducted a study with 32 participants, asking them to imagine and interact with two contrasting types of social spaces: open and closed. Open spaces were characterized by high contextual openness, implying broad accessibility and loose structure, while closed spaces were intimate, private, and selective.
The study revealed clear distinctions in how participants conceptualized these spaces:
- Open Spaces: Often described as expansive, vibrant, and spontaneous, with metaphors like “a playground” or “a job fair.” They emphasized dynamic interaction among diverse participants.
- Closed Spaces: Characterized by intimacy, trust, and privacy, using metaphors such as “a neighborhood bar” or “my bedroom with close friends.” These spaces focused on sustained, deeper engagement.
Participants’ feedback on the system-generated features highlighted that Messaging Types, Reactions, and Identity were consistently well-aligned across both open and closed settings. However, features like Recommendation, Networking controls (e.g., block/mute), and Connection Type often showed misalignments, indicating that the system sometimes struggled to capture the nuanced relational dynamics implied by the metaphors.
Experiencing the Simulated Environments
When interacting with the simulated platforms, participants noted qualitative differences in how content, messaging, and social connections felt. In open simulations, content flow was ambient and distributed, with many people talking simultaneously. In closed spaces, content felt slower and more sequential, reinforcing a sense of individual or niche community engagement.
Messaging in open environments was perceived as casual and general, while in closed spaces, it was more personal, coherent, and anchored by shared context. Social connections in closed simulations fostered familiarity and continuity through visible profiles, whereas in open spaces, interactions were expected to be fleeting and anonymous.
Also Read:
- Understanding Smart Environments: When “What If” Explanations Help Most
- AutoMaAS: A Self-Evolving Framework for Multi-Agent AI Systems
Future Directions and Impact
This research demonstrates that metaphor-driven design is a powerful tool for eliciting value-driven expectations early in the design process, allowing users to explore fundamentally new social architectures beyond the limitations of current platforms. While the system successfully generated coherent representations of metaphors, challenges remain in capturing subtle social dynamics and generating truly realistic AI content.
Future improvements could include supporting richer social complexity, enabling dynamic modification of social space features during use, and enhancing visual and contextual representations to make simulations more immersive and emotionally resonant. By continuing to refine this approach, metaphor-driven social design can foster broader participation in shaping the future of digital social platforms, moving towards environments that better support diverse and meaningful human connections.
