The Baby's Journey - II
The Adult AI
As we discussed in the last blog, we explored how software evolved over time using the analogy of a baby’s birth. Now, let’s continue this analogy to understand the advancements that followed.
Just like humans, teenagers often believe they have figured out the world—they feel invincible, confident that they can handle anything, sometimes even better than adults. Early successes reinforce this belief, but reality soon catches up when they encounter new, unfamiliar situations.
Similarly, our teenager AI (the Machine Learning era) thrived in certain tasks but struggled when faced with scenarios it hadn’t been trained for. With limited knowledge and experience, its decision-making often faltered.
Take our email teenager, for example. It had learned to detect spam emails based on past training. But when new, more sophisticated spam tactics emerged, it failed to recognize them. Since it could only rely on what it had been taught, it struggled to adapt—just like a teenager facing real-world challenges for the first time.
Understanding the Teenager’s Struggles Through an Example
To grasp why the teenager AI struggled, let’s consider a simple analogy:
Imagine our baby AI was trained to recognize squares (only in two dimension). The creators taught it certain features of a square:
All sides are equal.
It has four sides.
With this limited knowledge, the teenager AI grew up believing it could confidently identify whether a given shape was a square or not.
Now, when shown a two-dimensional square, the teenager correctly recognized it as a square. So far, so good.
But what happens when we introduce a three-dimensional square (a cube-like representation)?
🔹 The teenager AI fails. It sees the extra depth, notices that the shape has more than four sides, and incorrectly concludes that it is not a square.
This happens because the teenager was never trained on 3D representations—it only knew squares in a restricted sense. Its knowledge was limited to what it had learned, so when faced with something new, it struggled to adapt.
Now, if we ask it to identify any shape, the teenager can only recognize squares—it has no knowledge of circles, triangles, or other geometric forms. Anything outside its training is completely unrecognized.
This limitation mirrors the struggles of early machine learning models—they could only perform well on the exact scenarios they were trained for. Any deviation, and they failed spectacularly.
The Birth of a Brain: Neural Networks
The teenager AI struggled. It could only recognize what it had been explicitly taught, failing when faced with new scenarios. The creators realized they needed a radical shift in approach.
💡 "What if we let the AI think on its own—like humans?"
💡 "What if we give it a brain?"
As these questions arose, so did the answers.
The creators invented a new system inspired by the human brain—they called it Neural Networks.
How Neural Networks Work
Neural networks consist of multiple layers of interconnected nodes (like neurons in the brain). These layers process information in stages, gradually refining the input to make a final decision.
📌 Key components of a Neural Network:
Input Layer – Takes in the raw data (e.g., an image of a shape).
Hidden Layers – These are the “thinking” layers. Each layer extracts key features and refines the understanding.
Output Layer – Produces the final decision (e.g., "This is a square").
From Teenager to Thinker: The Shape Example
Let's revisit the teenager AI’s failure and see how neural networks improve the process.
🔹 Teenager AI's approach:
It sees a 3D square but fails because it was only trained on 2D squares.
Since it doesn’t match its exact training, it incorrectly concludes: ❌ "Not a square."
🔹 Neural Network’s approach:
The input shape (a square) is broken into smaller features.
First hidden layer analyzes the edges and passes the data forward.
Next layer measures the sides and confirms they are equal.
Final layer combines all insights and correctly identifies the shape as a square. ✅
The difference?
Neural networks don’t just memorize—they break down complex data into smaller, meaningful parts.
This allows them to recognize patterns beyond their training data and make better, context-aware decisions.
Neural networks were the breakthrough that transformed AI from a rule-following teenager into a more thoughtful, adaptive system.
Note: To understand visually please do watch this video from 3blue1brown channel link
The Leap into Adulthood: Deep Learning
With the power of Neural Networks, the teenager AI evolved into a more thoughtful and capable entity. It no longer relied solely on predefined patterns but instead learned complex relationships within data.
This advanced stage of learning is known as Deep Learning—an evolution of Machine Learning that enables AI to make more autonomous and human-like decisions.
Now, the AI was no longer a teenager.
It had matured into adulthood, ready to take on more complex challenges, from understanding human language to generating new ideas.
But is adulthood the final stage? Or is there something beyond?
Stay tuned as we explore the next transformation—the rise of Generative AI.
