Thought Propagation
The best way to solve a hard problem is often to solve an easier, analogous one first. Thought Propagation systematically identifies similar problems, solves them, and then propagates those solutions — the insights, strategies, and partial answers — back to refine the original problem’s solution.
Introduced: Thought Propagation was introduced in 2023, addressing a limitation of existing reasoning methods: they solve each problem in isolation, missing opportunities to leverage solutions from related problems. Inspired by how human experts often solve hard problems by first working through simpler analogues, Thought Propagation generates and solves analogous problems, then “propagates” the insights back to improve the target solution. The approach achieved 12-15% improvements across reasoning benchmarks.
Modern LLM Status: Thought Propagation represents an important advance in meta-reasoning — the idea that how you approach a problem matters as much as the reasoning itself. While frontier models have improved at solving problems directly, the analogy-based approach remains valuable for genuinely novel or difficult problems where direct reasoning struggles. The technique is particularly relevant for complex planning, optimization, and creative problem-solving tasks where insights from related domains can unlock solutions.
Solve Analogies First, Then Propagate
Most reasoning techniques work within the bounds of the given problem. Chain-of-Thought reasons step by step through the problem as stated. Tree of Thoughts explores multiple paths within the same problem space. Thought Propagation takes a different approach entirely: it steps outside the problem.
First, it generates analogous but potentially simpler problems. Then it solves those analogies. Finally, it “propagates” the solutions — extracting transferable strategies, patterns, and insights — and applies them to refine the original solution. This is how expert consultants work: they’ve solved hundreds of similar problems, and each new client benefits from that accumulated pattern recognition.
Think of it like a chess grandmaster who studies hundreds of historical games before a tournament. They don’t just analyze the opponent — they build a repertoire of proven strategies from analogous situations that they can draw on when the pressure is on.
When you’re stuck on a problem, a fresh perspective from a related domain can unlock the solution. Thought Propagation formalizes this: solving analogous problems first creates a repertoire of strategies and partial solutions. These solutions serve as “hints” that guide reasoning on the target problem, reducing the search space and suggesting productive reasoning paths that direct approaches might miss.
The Thought Propagation Process
Five stages from target problem to analogy-refined solution
Analyze the Target Problem
Understand the core structure, constraints, and challenges of the original problem. Identify what makes it difficult — this analysis guides the generation of useful analogies.
“Design an efficient scheduling system for a hospital ER with 12 doctors, 3 shifts, varying patient loads, and mandatory rest periods.” Key challenge: constraint satisfaction with dynamic demand.
Generate Analogous Problems
Create similar problems that share key structural features but may be simpler, differently constrained, or from related domains. The analogies should be solvable and informative.
Analogy 1: Airline crew scheduling (similar constraint structure, well-studied domain).
Analogy 2: University exam scheduling (fixed time slots, resource constraints, fairness requirements).
Solve the Analogies
Work through each analogous problem, generating full solutions. Pay attention to the strategies used, dead ends encountered, and key insights discovered during solving.
Airline solution insight: Constraint propagation works better than greedy assignment. University solution insight: Flexible pools outperform fixed assignments for handling demand variation.
Extract Transferable Insights
From the analogy solutions, identify patterns, strategies, and partial solutions that could apply to the original problem. What worked? What approach was most effective?
Key insights: (1) Use constraint propagation instead of greedy scheduling. (2) Maintain a flexible pool of on-call staff rather than rigid shift assignments. (3) Build in buffer capacity for demand spikes.
Propagate to Target
Apply the extracted insights to the original problem. Use the analogy solutions as scaffolding to guide, refine, or validate the target problem’s solution.
Refined ER schedule: Uses constraint propagation for base assignments, maintains a flexible on-call pool of 2 doctors per shift, and includes demand-responsive buffer periods — strategies proven effective in the analogous domains.
See the Difference
Why analogy-informed reasoning produces more sophisticated solutions
Direct Reasoning
Design an efficient scheduling system for a hospital emergency room with 12 doctors, 3 shifts, varying patient loads, and mandatory rest periods.
Assign doctors to shifts based on seniority, ensuring at least 4 per shift and 8 hours between shifts. Rotate weekly to distribute night shifts fairly.
Thought Propagation
Analogy 1 (Airline Crews): Constraint propagation outperforms greedy assignment for complex scheduling.
Analogy 2 (University Exams): Flexible resource pools handle demand variation better than fixed assignments.
Propagated insights applied: Use constraint propagation for base schedule, maintain flexible on-call pool, build demand-responsive buffers.
Three-tier system: (1) Constraint-propagated base schedule respecting all rest requirements, (2) flexible on-call pool of 2 doctors per shift for demand spikes, (3) historical patient load data driving predictive staffing. Strategies validated in airline and university scheduling domains.
Practice Responsible AI
Always verify AI-generated content before use. AI systems can produce confident but incorrect responses. When using AI professionally, transparent disclosure is both best practice and increasingly a legal requirement.
48 US states now require AI transparency in key areas. Critical thinking remains your strongest tool against misinformation.
Thought Propagation in Action
See how solving analogies first produces stronger solutions
A mid-size software company wants to expand into the healthcare vertical. They have strong engineering talent but no healthcare domain expertise, regulatory knowledge, or existing customer relationships.
Analogy 1: Fintech entering insurance. Solved: Partnership-first strategy with established insurers reduced regulatory risk. Key insight: domain expertise acquired through strategic hires, not built organically.
Analogy 2: Gaming company entering education. Solved: Started with non-regulated adjacent market (corporate training) before tackling K-12 compliance. Key insight: beachhead in low-regulation adjacent space builds credibility.
Propagated to target: (1) Start with wellness/fitness apps (lower regulatory burden) to build healthcare-adjacent credibility. (2) Hire a Chief Medical Officer and compliance lead before building product. (3) Partner with an established health system for pilot programs rather than going direct-to-consumer. (4) Use the partnership to learn HIPAA compliance in practice, not just theory.
Result: A phased market entry strategy informed by proven patterns from analogous vertical expansions, reducing risk at each stage. Note: Always verify AI-generated business strategy with qualified domain experts.
Design a pedestrian bridge over a busy highway that must span 60 meters, support dynamic loads from foot traffic and wind, and minimize visual impact on the surrounding neighborhood.
Analogy 1: Cable-stayed footbridge (30m span). Solved: Single pylon with radiating cables. Key insight: cable-stayed designs distribute load efficiently but the pylon creates a visual focal point.
Analogy 2: Covered garden walkway (aesthetic priority). Solved: Tensile membrane roof with minimal structural members. Key insight: translucent materials reduce visual mass while providing weather protection.
Propagated to target: (1) Use a cable-stayed design with twin low-profile pylons instead of a single tall one (from analogy 1’s load distribution insight, modified to reduce visual height). (2) Apply tensile membrane cladding (from analogy 2) to create a light, translucent enclosure that reduces perceived mass. (3) The combination achieves structural efficiency from the cable-stayed approach while maintaining the visual lightness from the garden walkway solution.
Result: A hybrid design combining structural strategies from two analogous projects, each contributing a proven solution to a different aspect of the original challenge. Note: Engineering designs require professional structural analysis and verification.
A state government wants to implement a comprehensive AI transparency regulation for public agencies. They need to balance accountability with practical implementation, avoiding both regulatory capture and innovation-stifling overreach.
Analogy 1: Environmental impact assessments (EIA). Solved: Tiered review process based on risk level. Key insight: not all projects need the same scrutiny — risk-proportionate review prevents bottlenecks while ensuring high-risk cases get thorough analysis.
Analogy 2: Financial audit requirements. Solved: Independent third-party auditing with public disclosure. Key insight: external auditors provide credibility that self-reporting lacks, but audit standards must be specific and testable.
Propagated to target: (1) Implement tiered AI impact assessments (from EIA analogy) — low-risk AI systems get self-certification, medium-risk gets departmental review, high-risk requires independent evaluation. (2) Require third-party algorithmic audits (from financial audit analogy) for high-risk deployments, with publicly disclosed results. (3) Define specific, testable transparency standards rather than vague principles. (4) Build in sunset clauses for reassessment as AI capabilities evolve.
Result: A risk-proportionate regulatory framework drawing on proven governance patterns from environmental and financial regulation. Note: Policy recommendations should be reviewed by qualified legal and policy experts.
When to Use Thought Propagation
Best for novel problems where analogies from related domains add value
Perfect For
Genuinely hard or unfamiliar problems where direct reasoning struggles — analogies from related domains can unlock entirely new solution paths.
Planning and optimization tasks where proven strategies from similar domains can inform better solutions than reasoning from first principles.
Problems with structural similarities to known solved problems in other fields — the technique excels at transferring insights across domain boundaries.
Tasks where cross-pollination of ideas adds value — the analogy generation step naturally produces diverse perspectives on the problem.
Skip It When
Problems that don’t need analogy-based reasoning — the overhead of generating and solving analogies adds no value to straightforward tasks.
When speed matters most — solving multiple analogous problems before addressing the target adds significant processing time and token usage.
Problems with no useful analogies — when the problem is so domain-specific that solutions from other fields would not transfer meaningfully.
Use Cases
Where Thought Propagation delivers the most value
Strategic Planning
Solve analogous market entry, expansion, or pivot challenges from other industries before designing the target business strategy.
Product Design
Study how similar user experience challenges were solved in related products before designing new features or interfaces.
Research Direction Setting
Analyze how breakthroughs were achieved in analogous research fields to identify promising approaches for the target research question.
Complex Negotiations
Study successful negotiation strategies from analogous multi-party disputes before entering a high-stakes negotiation.
System Architecture
Examine how similar scalability, reliability, and performance challenges were solved in analogous systems before designing new architecture.
Curriculum Development
Analyze effective pedagogical approaches from analogous subjects and learning contexts to design more engaging and effective curricula.
Where Thought Propagation Fits
Thought Propagation leverages analogies that other techniques overlook
The power of Thought Propagation depends entirely on the quality of analogies generated. Guide the model toward analogies that share structural similarities (same type of constraints, similar optimization goals) rather than surface similarities (same domain but different problem types). A scheduling problem in airlines is a better analogy for hospital scheduling than a medical diagnosis problem, even though the latter shares the healthcare domain.
Related Techniques
Explore complementary reasoning techniques
Propagate Better Solutions
Apply analogy-based reasoning to your challenges or explore other advanced techniques.