Artificial intelligence agents in orthopaedics: Concepts, capabilities and the road ahead

Figures

• 

  Figure 1. Sample Architecture of an artificial intelligence (AI) agent. This figure illustrates the core components of a typical AI agent. The perception system gathers raw data from the environment (e.g., patient information, sensor readings). This data are processed and sent to the reasoning & decision‐making engine (the ‘Brain’), which uses logic and AI models to create an action plan based on the agent's overarching goals and objectives. The engine accesses memory (a knowledge base of past experiences) to improve its decisions. The resulting action plan is sent to the action system, which uses effectors to interact with and modify the external environment (e.g., updating a clinical schedule or controlling a device). This entire process forms a continuous loop, allowing the agent to learn and adapt.

• 

  Figure 2. Illustration of three common architectures for implementing agentic artificial intelligence (AI). Multiagent systems (MAS) use multiple autonomous agents working towards a shared goal. Hierarchical reinforcement learning (HRL) structures decision‐making into levels, where higher levels set subgoals for lower levels. Goal‐oriented modular architectures divide a process into specialised modules, each responsible for a specific part of the task.