Navigating the AI Frontier 2024

with the physical world (in robotics), executing a software function or providing recommendations and decisions to human users. The learning component is intrinsic to the model and enables the AI agent to improve its performance over time as the model gathers more input, using machine learning and deep learning techniques as mentioned in section 2.1. The application layer surrounds the control centre, models and other components, acting as the interface between the AI agent and its environment. It interprets the outputs from the control centre and adapts them to specific tasks or domains. For example, in a healthcare AI agent, the application layer would translate model outputs into diagnostics, treatment recommendations or medical alerts through an appropriate user interface. In summary, when the varying components of an advanced AI agent come together, they represent the agent’s ability to model the environment, maintain memory or knowledge storage with beliefs and preferences, as well as inherent abilities to learn, plan, make decisions, perceive (sense), act (interact) and communicate with the agent’s surroundings. Example of an advanced AI agent: AI agent infotainment system An AI agent in a car’s infotainment system acts as a smart assistant, activated through voice commands to manage navigation, entertainment, climate controls and other vehicle settings. It processes live traffic, weather and driver preferences to optimize routes, suggesting alternatives around delays or hazards. The agent personalizes entertainment based on user habits, recommends nearby stops such as restaurants or fuel stations and proactively provides updates such as low fuel alerts or optimal recharging points for electric vehicles – all while ensuring the driver remains focused on the road. AI agent system 2.4 An AI agent system is an organized structure that integrates multiple heterogeneous (e.g. rule- and goal-based agents) or homogeneous (e.g. goal- based only) AI agents.20 Each agent is typically specialized, possessing its own capabilities, knowledge and decision-making processes, while sharing data to collaboratively achieve the goal of the system. Several designs are possible, such as: –Mixture-of-agents, where each agent is called sequentially, with agents processing the outputs from each previous agent21 –Central orchestration, which coordinates calls of agents and manages the inputs and outputs accordingly The AI agent system is designed to ensure that each agent contributes to the overall objective, whether it involves managing complex real-time processes such as autonomous driving, optimizing industrial processes or coordinating activities; for example, in smart city infrastructure. By dividing the workload among specialized agents, the system can handle dynamic environments and adapt to changing conditions, ensuring optimal performance. Example of an AI agent system: Autonomous vehicle AI agent system A human user gets into an autonomous vehicle (AV). The AV is comprised of an AI agent system that includes agents for perception, path planning, localization for finding its specific place on the road and control to steer and brake. The perception and localization agents are dedicated to continuously mapping the environment through sensors, the global positioning system (GPS) and cameras. The planning agent calculates the optimal trajectory by factoring in real-time traffic, weather and road conditions. The control agent handles the vehicle’s core mechanics, such as braking, accelerating and steering.22 The AI agent infotainment system serves as the interface with the passenger, and handles elements such as processing voice commands and adjusting routes, climate, entertainment or other in-car settings based on user preferences.23 All agents work together in a coordinated and centralized manner to ensure the vehicle reaches its destination safely and efficiently, prioritizing both passenger comfort and safety.241. 2. Navigating the AI Frontier: A Primer on the Evolution and Impact of AI Agents 13