Blockchain is a revolutionary protocol that enables transactions to be both anonymous and secure through a tamper-proof public ledger. Despite its significant potential, blockchain faces unresolved security challenges. Blockchain systems are highly dynamic and large-scale. Therefore, many management problems, such as defense mechanisms against a wide range of attacks, cannot be handled by humans because human reaction time is insufficient for many management tasks in blockchain systems. In other words, human reasoning is required in many situations, but we cannot use real humans as managers or system admins. Meanwhile, cognitive systems, designed to mimic human thinking processes through digitalized models, have seen widespread adoption. The core component, the cognitive engine, is responsible for implementing these functionalities. This paper proposes a novel framework that integrates cognitive systems into blockchain to defend against attacks. To our knowledge, no existing framework leverages cognitive systems for blockchain security. We specifically design a reinforcement learning (RL)-based defense mechanism to counter selfish mining attacks based on the cognitive defense framework. Simulation results demonstrate that our proposed cognitive framework significantly enhances blockchain security.