How to Build AI Agents 1: The Journey from Machine Learning to AI Agents

Welcome to Day 1 of my six-day journey into building AI agents! My goal is to spend an hour each day diving into the fascinating world of AI and, by the end of Day 6, create my very own AI agent. Excited? Me too!

In this article I covered
- The journey from Artificial Intelligence(AI) and Machine Learning to Large Language Model(LLM)
- What are AI Agent

Before jumping straight into AI agents, I wanted to take a step back and explore how we got here. It’s like tracing your family tree understanding the roots helps you appreciate the whole picture.

Let’s rewind to the 1950s, often referred to as the dawn of Artificial Intelligence (AI). This article from Nvidia explain the early ages of AI in 1950. Imagine a time when computers were massive, room-filling machines. Scientists back then were asking a bold question: "Can machines think like humans?" Alan Turing, one of the pioneers, proposed the famous Turing Test to measure if a machine could exhibit human-like intelligence. Spoiler: We’re still working on perfecting that!

Artificial Intelligence (AI)

AI is the big umbrella term that covers everything. Think of it as a machine’s ability to mimic human intelligence. It can reason, solve problems, and even play chess—like IBM’s Deep Blue, which famously defeated world champion Garry Kasparov in the 90s.

Machine Learning (ML)

Fast-forward to the 1980s and 1990s, where we meet Machine Learning (ML). While AI aimed to replicate intelligence, ML focused on one critical aspect of it: learning from data. Here’s an example to make it clearer:

Imagine teaching a toddler to recognize dogs. You show them lots of pictures of dogs and say, “This is a dog.” Over time, they learn to identify dogs even in new pictures. Similarly, ML algorithms analyze data (pictures, numbers, or text), identify patterns, and use those patterns to make decisions—like predicting tomorrow’s weather or recommending a song you’ll love.

Deep Learning: Machines Think Like Humans (Almost)

Then came the 2010s and the rise of Deep Learning. Think of deep learning as ML on steroids. Instead of simply learning patterns, deep learning uses a structure inspired by the human brain: neural networks.

Here’s a fun analogy: Imagine teaching someone to bake a cake. With ML, they’d memorize recipes. With deep learning, they’d learn how to experiment, tweak ingredients, and even create entirely new cakes! Deep learning allows machines to “understand” more complex data like speech, images, and videos, opening the door to exciting possibilities.

Generative AI: Machines Create

Here’s where things start getting really cool. The next milestone was Generative AI. Generative AI takes learning to a creative level. While traditional AI analyzed and predicted, generative AI creates. These models don’t just recognize patterns; they create. They can write poetry, compose music, and even generate images. If you’ve ever been amazed by AI art or tools that write essays, that’s Generative AI at work.
These models can generate text, images, music, and more.

Large Language Model (LLM)

Finally, we arrive at Large Language Models, a special kind of Generative AI. LLMs, like OpenAI’s GPT series use a clever architecture called Transformers to understand and generate text. Imagine having a machine that’s read almost every book, article, and webpage ever written it uses that knowledge to produce human-like responses.

Take ChatGPT, for example. It’s powered by a special kind of generative AI called a Large Language Model (LLM). These models use an architecture called “transformers” to understand language and respond in ways that feel natural. Whether crafting poetry, answering questions, or coding, LLMs showcase the creativity of machines.

Now that we’ve covered the history, let’s talk about why we’re here: AI agents.

What are AI Agents?

Think of AI agents as advanced AI systems that can not only think and plan using their built-in knowledge (like ChatGPT does) but also use tools to get things done. These tools could include a web browser to search for information, an IDE to write code, or a calculator to solve math problems.

For example:

1. Planning a Trip: An AI agent could use a web browser to look up flight options, calculate the total cost using a calculator, and even organize your itinerary.

2. Writing Code: It might use an IDE to debug or write code for a specific task while reasoning through the logic with its built-in intelligence.

AI agents are like LLMs on a mission. While LLMs generate text and hold conversations, agents take it a step further by:

• Using tools: They can access resources like web browsers, calculators, or even IDEs.

• Performing tasks: From booking flights to debugging code, they’re task-oriented.

• Reasoning and planning: Thanks to their underlying LLMs, they can make decisions and adapt to new situations.

• Memory: Agents retain context, allowing them to carry out multi-step tasks seamlessly.

What’s Next?

Today, we explored the rich history of AI and how it paved the way for AI agents. Tomorrow, I’ll dive deeper into the concepts and components of these agents. Stay tuned as I continue this exciting journey toward building my own AI agent—and perhaps inspire you to do the same!

References

AI vs ML vs DL vs Generative AI - https://www.youtube.com/watch?v=X7Zd4VyUgL0&ab_channel=KrishNaik