에이전트 유형과 의사 결정

Welcome! In this lesson, we'll dive into the fascinating world of how AI agents make decisions. It's what makes them seem 'smart' and capable of complex tasks.

Think of an agent as having a 'brain' that processes information and chooses the best next step from a set of options, often involving tools.

At the heart of most AI agents is a Large Language Model (LLM). The LLM isn't just for generating text; it's also the agent's primary decision-making engine.

• It interprets your request.
• It considers the available tools.
• It generates a 'thought' process to decide what to do.

This 'thought' guides the agent's next action.

One of the most common and powerful agent architectures is ReAct. This stands for Reasoning and Acting.

A ReAct agent works in a loop:

1. Thought: The LLM reasons about the current situation and what to do next.
2. Action: Based on the thought, the LLM chooses a tool and its input.
3. Observation: The tool executes, and its output is returned to the LLM.

This cycle repeats until the agent reaches a final answer.

Imagine you ask an agent: 'What is the current weather in London?'

• Thought: 'The user wants weather info. I have a 'weather_tool'.'
• Action: 'Call weather_tool with city='London'.'
• Observation: 'Weather in London: 15°C, cloudy.'
• Thought: 'I have the answer. I should respond to the user.'
• Action: 'Respond: 'The weather in London is 15°C and cloudy'.'

This iterative process allows agents to tackle complex tasks step-by-step.

This Python code simulates a simplified agent's decision process based on user input and available tools. Notice how it 'thinks' and decides on an 'action'.

While ReAct is powerful, some agents need to maintain a continuous conversation, remembering past interactions. These are conversational agents.

Their decision-making isn't just about the current turn; it's also heavily influenced by the conversation history (their 'memory').

For conversational agents, the LLM receives not only the user's latest input but also a summary or full transcript of the previous turns.

• This memory helps the agent understand context.
• It allows for follow-up questions and avoids repetition.
• The decision to use a tool or generate a direct response is informed by the entire chat history.

We'll explore memory in detail in a later course!

Beyond ReAct and basic conversational agents, there are other sophisticated architectures:

• Plan-and-Execute Agents: First create a multi-step plan, then execute it.
• Self-Correction Agents: Evaluate their own outputs and try again if they detect errors.
• Tree-of-Thought Agents: Explore multiple reasoning paths before committing to an action.

Each type offers different strengths for various complex tasks.

An agent's decision-making process is influenced by several key factors:

• User Prompt: The clarity and specificity of the user's request.
• Available Tools: The functions and capabilities the agent has access to.
• Memory/Context: Past interactions that provide background.
• LLM Capabilities: The model's reasoning abilities and knowledge.

Effective agent design means balancing these elements.

Great job! You've learned how AI agents make decisions, moving beyond just text generation.

• The LLM acts as the agent's 'brain', interpreting input and choosing actions.
• ReAct agents use a 'Thought-Action-Observation' loop for step-by-step problem solving.
• Conversational agents integrate memory to maintain context over time.
• Various factors like prompts, tools, and memory influence an agent's choices.

Understanding these decision mechanisms is key to building powerful AI applications!