Knowledge Graph — Coursera Notes › Organizations › Amazon › Amazon Web Services › Amazon Bedrock

AI agent

concept · part of Amazon Bedrock

A generative AI-powered system in Amazon Bedrock that can plan, reason, and interact to carry out complex tasks.

It operates through a perceive-think-act cycle: perceive (gather information via sensors), think (process information using algorithms or models), and act (execute actions to affect the environment).

The perceive-think-act cycle is the fundamental loop governing AI agents.

AI agents are used in healthcare (assisting doctors in diagnosing diseases and personalizing treatment plans), finance (monitoring stock markets and making split-second trading decisions), and customer service (virtual agents handling inquiries, reducing human intervention in call centers).

An AI agent is a software entity that observes its environment and takes actions to achieve specific goals autonomously. It can range from simple reactive systems to advanced models capable of learning and planning.

Autonomy refers to the ability of an AI agent to operate without direct human intervention. An autonomous agent perceives its environment, makes decisions, and takes actions based on those decisions while pursuing specific objectives. Example: a self-driving car operates autonomously by gathering data from its surroundings and navigating traffic without human input.

In digital environments, this could involve collecting data from APIs; in physical environments, it might include cameras, microphones, or lidar. Example: a robotic vacuum cleaner uses sensors to detect obstacles; a stock trading bot uses virtual sensors to collect market data from financial APIs.

AI agents are designed to act rationally, striving to achieve their goals by selecting the best possible actions from available options. They aim to maximize success, often based on utility functions or reward mechanisms, and also minimize risk. Example: an autonomous vehicle makes decisions that maximize passenger safety and minimize accident risk, guided by a utility function balancing safety, efficiency, and traffic law compliance.

Many AI agents have the ability to learn from experience, improving over time through ML algorithms that adapt behavior based on past interactions, feedback, or rewards. Example: a recommendation system on a streaming service learns from user behavior to provide more personalized suggestions over time.

In multi-agent systems, each agent has its own tasks and must communicate and coordinate with others.

An autonomous agent in a warehouse simulation has attributes like position and task, and methods to measure distance, assign tasks, and mark completion. Each robot operates independently but coordinates with others.

class Robot:
    def __init__(self, warehouse, position):
        self.warehouse = warehouse
        self.position = position
        self.task = None
    def distance_to(self, location):
        return abs(self.position - location)
    def assign_task(self, task):
        self.task = task
    def complete_task(self):
        if self.task:
            # perform retrieval or restocking
            self.task = None

Inside AI agent (7)

• Multiagent System — A multiagent system (MAS) consists of multiple interacting agents, each with individual goals, perceptions, and actions.
• AI agent architecture — The architecture of AI agents includes core components: Perception (sensing) collects data from the environment; Knowledge base stores information about the environment, goals, and learned experiences; Reasoning and decision-making processes input to decide actions; Learning (optional) improves behavior over time; Action (actuators) executes decisions; Communication interface handles interaction with other agents, systems, or humans.
• Goal-based agents — Goal-based agents take future goals into account when making decisions, prioritizing actions that move them closer to achieving those goals.
• Intelligent troubleshooting agent — An intelligent troubleshooting agent is a specialized AI agent that assists users in diagnosing and solving problems autonomously.
• Model-based agents — Model-based agents maintain an internal representation of the environment to inform decisions.
• Simple reflex agents — Simple reflex agents make decisions based purely on current input, without internal state.
• Utility-based agents — Utility-based agents optimize actions based on a utility function, aiming not just to reach a goal but to find the best way to do so.

Connections

• Uses Foundation Models
• Uses AWS Lambda
• Uses Knowledge Bases
• Uses Guardrails
• Alternative to Bedrock Agents
• Related to Amazon Bedrock
• Related to Foundation Models
• Related to AWS Lambda
• Related to Knowledge Bases
• Related to Guardrails
• Related to Bedrock Agents
• Related to Bedrock Agents

This is the text view of an interactive 3D knowledge graph — open this page with JavaScript enabled to explore it visually.