Navigating the AI Frontier 2024

failure or errors and limited skill sets. By distributing tasks among multiple agents, MAS could increase both efficiency and capability. In theory, multi-agent systems are highly adaptable, as agents can be dynamically added or removed, allowing the system to respond to changing environments and requirements. This scalability is crucial for applications that need to grow or evolve over time without extensive re-engineering. In many ways, multi-agent systems can be considered as a future type of system that could coordinate agent actions among multiple users or organizations through human-comprehensible language or to-be-determined AI agent protocols. FIGURE 5: The structure and relationships among the AI agent, AI agent system and multi-agent system Advanced AI agent EXAMPLE: AI agent infotainment systemEXAMPLE: Fully autonomous vehicleEXAMPLE: Connected smart city coordinationAI agent system Multi-agent system Sensors Learning AI agent 1 Orchestration AI agent 3Other AI agentAI agent 2AI agent system 1AI agent 2 AI agent 3Other agentsEffectorsControl centre Model Decision- making and planningMemory managementTools 1. 2. 3. Example of a multi-agent system: Smart city traffic management with vehicle-to- everything (V2X) communication Source: World Economic Forum In a smart city, a multi-agent system (MAS) manages traffic flow in real time, using vehicle-to- everything (V2X) communication, enabling vehicles to interact with other vehicles, pedestrians and road infrastructure.27 Each traffic signal is controlled by an AI agent system that communicates with nearby signals, public transport systems, emergency services and parking services to check availability. Vehicles, equipped with their own AI agent system, share data such as speed, location and road conditions, allowing for coordinated actions to enhance road safety, traffic efficiency and energy usage. For example, if an accident occurs, AI agents can reroute traffic, adjust signal timings, notify emergency services and communicate with vehicles and pedestrians to avoid the area, all with minimal human intervention. This system optimizes traffic flow, improves road safety and reduces energy consumption by dynamically adapting to real-time conditions. For instance, if a parking lot is full, the system can direct vehicles to available parking further away, even if it conflicts with the driver’s and the onboard AI agent’s preference for proximity.3. Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents 15