Universe of Thoughts: Enabling Creative Reasoning with Large Language Models

Abstract: Reasoning based on LLMs has garnered increasing attention due to outstanding performance of these models in mathematical and complex logical tasks. Beginning with the Chain-of-Thought (CoT) prompting technique, numerous reasoning methods have emerged that decompose problems into smaller, sequential steps (or thoughts). However, existing reasoning models focus on conventional problem-solving and do not necessarily generate creative solutions by ``creative reasoning''. In domains where the solution space is expansive and conventional solutions are suboptimal, such as drug discovery or business strategization, creative reasoning to discover innovative solutions is crucial. To address this gap, first we introduce a computational framework for creative reasoning inspired by established cognitive science principles. With this framework, we propose three core creative reasoning paradigms, namely, \textit{combinational}, \textit{exploratory}, and \textit{transformative} reasoning, where each offers specific directions for systematic exploration of the universe of thoughts to generate creative solutions. Next, to materialize this framework using LLMs, we introduce the \textit{Universe of Thoughts} (or \textit{UoT}, for short), a novel set of methods to implement the aforementioned three creative processes. Finally, we introduce three novel tasks that necessitate creative problem-solving, along with an evaluation benchmark to assess creativity from three orthogonal perspectives: feasibility as constraint, and utility and novelty as metrics. With a comparative analysis against the state-of-the-art (SOTA) reasoning techniques as well as representative commercial models with reasoning capability, we show that UoT demonstrates superior performance in creative reasoning.

• The paper introduces the Universe of Thoughts framework enabling LLMs to perform autonomous creative reasoning across combinational, exploratory, and transformative paradigms.
• It details a modular prompt pipeline evaluated on tasks like one-lane bridge, electricity tariff, and social cohesion, demonstrating superior performance over existing models.
• Empirical results highlight enhanced feasibility, novelty, and utility in AI-driven creative problem-solving, paving new directions for autonomous innovation.

Enabling Creative Reasoning with LLMs: A Summary

Introduction to Creative Reasoning Framework

The paper "Universe of Thoughts: Enabling Creative Reasoning with LLMs" (2511.20471) presents a novel computational framework aimed at equipping LLMs with the ability to autonomously perform creative reasoning. Traditional LLMs excel in structured problem-solving but falter when addressing ill-defined problems where the solution space is vast and ambiguous. This gap is especially pronounced in domains such as drug discovery or strategic business planning, where conventional approaches may yield suboptimal solutions.

The Universe of Thoughts (UoT) Framework

To address this challenge, the authors introduce the "Universe of Thoughts" (UoT), a suite of methods that implement three core creative reasoning paradigms: combinational, exploratory, and transformative reasoning. Each of these paradigms derives from Margaret Boden's cognitive science framework, tailored to systematically explore the "universe of thoughts" to generate creative solutions.

Creative Reasoning Paradigms

• Combinational Reasoning: This paradigm synthesizes novel solutions by combining existing thoughts in unconventional ways. It identifies thoughts from analogous domains and integrates them into the target solution space.
• Exploratory Reasoning: It expands the thought palette by identifying and incorporating novel thoughts not previously part of known solutions. This approach broadens the target solution space and enables the generation of innovative solutions beyond the existing thought boundaries.
• Transformative Reasoning: The most profound form of creative reasoning, it involves altering the foundational rules of the solution space. By relaxing or redefining these constraints, it opens up entirely new solution spaces that were previously inconceivable.

Implementation and Evaluation

To operationalize these paradigms, the UoT framework is implemented as a structured pipeline of modular prompts that leverage LLM capabilities. The framework is evaluated on three novel tasks designed to test creative problem-solving capabilities:

1. One-Lane Bridge Task: Designing a policy to minimize average vehicle delay while maintaining unidirectional traffic flow on a single-lane bridge.
2. Electricity Tariff Task: Creating a tariff and demand response program to reduce peak load on a residential feeder without expanding physical capacity.
3. Social Cohesion Task: Developing an intervention to strengthen cross-group cohesion within a community while ensuring privacy and inclusivity.

The framework introduced an evaluation benchmark that assesses creativity from three orthogonal perspectives: feasibility, utility, and novelty. A canonicalization process ensures consistent evaluation across varying solution granularities.

Results and Implications

Experimental results highlight the efficacy of UoT methods, particularly T-UoT, which consistently demonstrates superior creative reasoning performance compared to existing state-of-the-art techniques. Notably, UoT implementations surpass proprietary models in some instances, showcasing the potential of structured creative reasoning frameworks to drive innovation in LLM applications.

The ability to systematically generate creative solutions suggests significant implications for AI research and application in domains requiring innovative problem-solving. The structured approach to reasoning presented in this paper provides a foundation for developing AI systems capable of addressing complex, ill-defined challenges autonomously.

Conclusion

The "Universe of Thoughts" framework propels LLM capabilities beyond conventional problem-solving by embedding cognitive science-driven creative reasoning methodologies. This advancement holds promise for AI systems to autonomously discover novel solutions across diverse domains, marking a significant step toward more innovative and adaptive AI applications. Future research directions include exploring additional methods to further enhance the proposed creative reasoning framework, potentially expanding its applicability and efficacy in increasingly complex and dynamic problem spaces.