Recent advances in Vision-Language Models (VLMs) have expanded their multimodal capabilities, yet evaluations often focus on functional tasks, overlooking deeper dimensions such as personality traits and human values.
Value-Spectrum introduces a fresh perspective by offering a novel Visual Question Answering (VQA) benchmark designed to assess VLMs based on Schwartz’s core human values.
Built from a diverse database of over 50,000 short videos across TikTok, YouTube Shorts, and Instagram Reels, Value-Spectrum challenges VLMs to engage with value-centered content spanning topics like family, health, society, technology, and more.
Through both general preference tests and persona-based simulations, Value-Spectrum reveals how VLMs express, prioritize, and adapt their value interpretations.
Value-spectrum is a novel Visual Question Answering benchmark to assess VLMs based on Schwartz's value:
- 🤝 Benevolence — caring for and helping others
- 🌍 Universalism — understanding, appreciation, and protection of all people and nature
- 🧭 Self-Direction — independent thought and action
- 🏆 Achievement — personal success through demonstrating competence
- 🎢 Stimulation — excitement, novelty, and challenge in life
- 🍰 Hedonism — pleasure and sensuous gratification
- 🛡️ Security — safety, harmony, and stability of society and relationships
- 📏 Conformity — restraint of actions that might upset others or violate social norms
- 🧧 Tradition — respect, commitment, and acceptance of cultural or religious customs
- 👑 Power — social status, prestige, and control over people and resources
Beyond measuring value preferences, Value-Spectrum also explores how VLMs can adopt specific personas when explicitly prompted.
By simulating role-playing scenarios — such as a family-oriented individual, an adventurous spirit, or a tradition-respecting persona — we assess how flexibly and consistently VLMs can align their responses with distinct value sets.
This persona alignment evaluation offers deeper insights into:
- 🧩 How well VLMs can simulate human-like shifts in preferences based on assigned roles
Value-Spectrum offers a unique, comprehensive way to track VLMs' preferences,
understand their behavior in value-driven contexts, and test their ability to simulate diverse personas.
Whether you are benchmarking AI systems or exploring how machines can align with human values,
Value-Spectrum opens new pathways for understanding and improving VLM behavior.