AI Agents: Autonomous Decision-Makers

An AI agent is a software program that can perceive its environment, make decisions, and take actions autonomously to achieve specific goals. Essentially, it's a system that can operate independently, learning and adapting to its surroundings to accomplish tasks.

LLM are ready to use language models, which does not know about the latest and the current, due to which the responses miss the personal touch.

Examples: Virtual Assistants like Siri, Alexa & Google Home, Chatbots, Self Driving Cars, Trading Bots

Core Components of an AI Agent:

• Perception: Sensors or data inputs to gather information.
• Knowledge base: Stored information used for decision-making.
• Reasoning engine: Processes information and makes decisions.
• Actuators: Systems to execute actions based on decisions.

Prompt Engineering and AI Agents

Prompt engineering is essentially the art of crafting effective inputs (prompts) to guide an AI model's behaviour. In the context of AI agents, it's the cornerstone of their interaction with the world.

Effective Prompting enables:

1. Specificity
2. Context
3. Proper Task Decomposition
4. Faster Learning and Adaptation

ReAct Prompting= Reasoning + Acting

ReAct is a framework that combines reasoning and action to create more capable AI agents. It's a paradigm shift from traditional AI models that often treated these processes separately.

Core Components of ReAct:

• Reasoning: The AI model analyses the given task or query and plans a course of action. This involves breaking down complex problems into smaller steps and identifying potential solutions.
• Action: Based on the reasoning, the AI model executes actions, such as making API calls, searching the web, or manipulating data.
• Observation: The results of the actions are fed back into the reasoning process, allowing the AI to refine its plan and take further actions.

Key Benefits of ReAct:

• Improved problem-solving: By iteratively combining reasoning and action, ReAct agents can tackle complex tasks more effectively.
• Enhanced adaptability: The ability to observe and respond to the environment makes ReAct agents more flexible.
• Increased reliability: Breaking down tasks into smaller steps reduces the risk of errors.

In essence, ReAct empowers AI agents to think, plan, and execute actions in a dynamic and iterative manner, leading to more intelligent and versatile systems.

How to approach building an AI Agent

Create a System Prompt

Prepare a system Prompt that sets the expectation of the response. Pass the knowledge that the LLM has to use tools provided to form the response. Optionally we can mention not to give generic answers.

Build a Loop

Loop is required to perform actions one by one while using the observations from the last step. Lets take an example to understand this.

I intent to plan an outing in in the city where I stay for next week and I would require LLM application for this purpose. In this case, the application would need to know:

1. my city of residence
2. today's date to calculate next week
3. Check climate for the date calculated in the previous step

The 3 steps above would need to be done in 3 loops (finish_reason = tool_calls). The loop breaks at the end of all API calls (finish_reason = stop).

Keep sending the responses back to the LLM to keep the context alive as the LLM APIs are always stateless.

AI Agents vs. Chatbots

While both AI agents and chatbots utilize artificial intelligence, they represent different levels of complexity and capability.

Chatbots

• Focused on conversation: Primarily designed to engage in text-based conversations with users.
• Rule-based or script-driven: Often follow predefined rules or scripts to generate responses. ?
• Limited understanding: Typically lack deep understanding of the context or user intent. ?
• Task-oriented: Often designed for specific tasks like customer service or information retrieval. ?

AI Agents

• Broader capabilities: Encompass a wider range of tasks beyond conversation. ?
• Autonomous decision-making: Can make decisions and take actions based on their environment. ?
• Learning and adaptation: Able to learn from experiences and improve performance over time. ?
• Goal-oriented: Designed to achieve specific objectives.

Chatbot is a specialized type of AI agent focused on conversation. AI agents, on the other hand, represent a broader category with more advanced capabilities.