Technology Category
Cognitive Modeling & Cognitive Systems: Computational CreativityHumans and AI: Human-Aware Planning and Behavior PredictionMultiagent Systems: Adversarial Agents
The Spark Effect: On Engineering Creative Diversity in Multi-Agent AI Systems
🤖 AI Summary
Multi-agent creative systems often suffer from output homogenization due to LLMs’ inherent uniformity, hindering their ability to meet brand- and art-specific diversity requirements. To address this, we propose “Spark”, a persona-driven agent architecture that employs a persona-conditioned system prompt library to induce behavioral differentiation among agents. Spark integrates a multi-agent collaborative workflow with an LLM-as-a-judge automated evaluation framework, calibrated against human expert annotations (gold-standard references). Its core innovation lies in the explicit embedding of persona modeling into system prompt design, enabling controllable and scalable regulation of creative diversity. Experimental results demonstrate that Spark achieves an average improvement of 4.1 points on a 1–10 diversity scale, narrowing the gap with human expert ratings to only 1.0 point—significantly bridging the performance divide between AI-generated and professional creative outputs.
Technology Category
Application Category
📝 Abstract
Creative services teams increasingly rely on large language models (LLMs) to accelerate ideation, yet production systems often converge on homogeneous outputs that fail to meet brand or artistic expectations. Art of X developed persona-conditioned LLM agents -- internally branded as "Sparks" and instantiated through a library of role-inspired system prompts -- to intentionally diversify agent behaviour within a multi-agent workflow. This white paper documents the problem framing, experimental design, and quantitative evidence behind the Spark agent programme. Using an LLM-as-a-judge protocol calibrated against human gold standards, we observe a mean diversity gain of +4.1 points (on a 1-10 scale) when persona-conditioned Spark agents replace a uniform system prompt, narrowing the gap to human experts to 1.0 point. We also surface evaluator bias and procedural considerations for future deployments.
Problem
Research questions and friction points this paper is trying to address.
Addressing homogeneous outputs in multi-agent AI systems
Engineering creative diversity through persona-conditioned LLM agents
Bridging performance gap between AI systems and human experts
Innovation
Methods, ideas, or system contributions that make the work stand out.
Persona-conditioned LLM agents diversify outputs
Role-inspired system prompts create behavioral variation
Multi-agent workflow calibrated via LLM-as-judge protocol
🔎 Similar Papers
No similar papers found.
Art of X | HFBK Hamburg
