Can smaller AI models perform complex reasoning tasks? New research suggests they can, with a little help from knowledge graphs. Traditionally, large language models (LLMs) have been the go-to for tasks requiring deep understanding, like causal discovery – the process of identifying cause-and-effect relationships between variables. However, these large models are resource-intensive and expensive. This new research explores how smaller language models (SLMs), which are more efficient, can achieve comparable performance in causal discovery by leveraging knowledge graphs. Knowledge graphs, essentially maps of interconnected information, provide contextual background knowledge that empowers SLMs to reason more effectively. This research introduces a novel technique called "KG Structure as Prompt," which cleverly incorporates the structural information of a knowledge graph into prompts. Instead of relying solely on their limited internal knowledge, SLMs use the knowledge graph as a guide. By examining relationships between variables within the knowledge graph, such as shared neighbors or indirect connections (metapaths), SLMs can infer causal links even when the direct evidence is subtle or missing. Experiments on various datasets, spanning biomedicine and open-domain topics, reveal that this approach significantly enhances the performance of SLMs in causal discovery. Remarkably, these enhanced SLMs even rival, and sometimes surpass, the performance of much larger LLMs, demonstrating the potential of knowledge graphs to boost smaller, more efficient AI models. This opens doors for broader access to sophisticated AI capabilities, especially in resource-constrained environments. The research also highlights the importance of choosing the right knowledge graph for a specific domain. For example, domain-specific knowledge graphs like Hetionet proved more effective for biomedical causal discovery, showcasing the value of specialized knowledge. Future work aims to extend this approach to more intricate causal relationships, ultimately enabling more powerful and cost-effective AI systems.