A curious mind’s take on AI

A few weeks ago, I floated a Google Form with a simple, almost laughable question: 

"Post your silliest AI-related doubts, as silly as: What even is AI?"

What I received was beautiful. 

Questions like: 

• Does AI have a brain?
• How does ChatGPT know stuff?
• Is AI just a fancy search engine?
• Is AI thinking?
• Can AI feel?

These may sound silly, but they represent something deeper, a discomfort we all feel when something seems foggy. If we can’t see how something works, we stop trying. Or worse, we pretend to understand it and move on. I remember first learning to code. Everyone else seemed to get it, but I kept thinking:

"What actually happens when I run this code?"

People said things like:

“It depends on the runtime.”

“The code compiles.”

“Functions execute.”

Of course, I understood what the code was supposed to do, the logic made sense. But what was really happening under the hood? I craved something tangible, something like how a mechanical system works.

Take a car’s brakes. When you press the pedal: 

• Your foot applies force.
• That force becomes hydraulic pressure.
• Pressure moves through brake lines.
• Calipers press pads onto rotors.
• Friction slows the wheels.

You can see it. You can picture the cause and effect. Simple. Understandable. 

So, to understand how code really works, I had to step back and ask a much simpler question. 

What happens when you press 3 + 4 = on a calculator? 

You press a button, say 3. Beneath your finger, a small dome collapses and connects a circuit for a brief moment, letting electricity pass. That current reaches a chip inside, the calculator’s brain. 

But the chip doesn’t see “3.” It sees a specific pattern of electricity: 0011. Same with 4: 0100. 

These binary numbers are passed to a tiny circuit called the ALU, the Arithmetic Logic Unit. Think of it like a little machine made entirely of logic gates. It doesn’t know math the way you do. It only knows how to flip switches in ways that, over time, we’ve engineered to behave like addition. 

It adds 3 and 4, gets 7 (0111), and sends that result to another chip that controls the display. Your screen lights up with a 7. Somehow, that made it click. It sat better in my head. Then I looked at code.

Let’s say I write: 

a = 3 

b = 4 

print(a + b)  

At a glance, it feels more abstract, variables, syntax, files. But underneath? It’s almost the same. 

Your code is just characters. The processor doesn’t read logic. It doesn’t know what a variable is. It just executes electrical impulses, flipping switches. One transistor at a time and the result appears. 

Code, just like a calculator, is electricity choreographed by logic only with more layers, more instructions, more complexity.

But with AI? It felt like that fog returned all over again. Ask three people what AI is, and you’ll get three equally vague answers.
You type:

"Summarize this article about climate change." 

And the machine not only understands it, but it also gives you a meaningful, sometimes beautiful summary. 

But you didn’t tell it how. You didn’t program the logic. It just knows.

Let’s unpack that. 

You type a sentence. Like before, it becomes binary. If your laptop doesn’t house the AI model your message travels to the cloud, to a data center filled with GPUs engineered for one thing: running neural networks. 

First, it breaks it apart, into fragments called tokens. Words, pieces of words, punctuation, like chopping a thought into LEGO bricks. Each token is then translated into a number. Not a label, not a shortcut, a position in space. Literally. The model imagines each word as a point floating in a multi-dimensional space where “king” and “queen” are close together, but “orange” is very far away.

Then, it begins its trick. Those vectors, your sentence, are passed through a towering network of layers. Billions of artificial neurons arranged in layers transform those inputs. They multiply them, weigh them, apply nonlinear functions, and pass them to the next layer. Each neuron adjusts its output based on what it 'learned' during training. It’s like pouring structured numbers into a funnel sculpted from experience, with weights shaped by patterns in books, forums, articles, conversations, and code. Every layer refines this numerical stream, amplifying some aspects, dampening others, gradually nudging the input toward something that feels meaningful.

The deeper the signal travels, the more abstract the representation becomes. Early layers might identify letters. Deeper layers recognise phrases. Even deeper ones spot relationship, like tone, intent, or contradiction. By the time the signal exits the last layer, the model doesn’t “know” what you meant, but it has calculated, based on trillions of patterns it has seen, the next most likely word.

One word. Then another. Then another.

And yet somehow, it feels like understanding.

Because that’s what intelligence often is: not rules, but patterns. Not instructions, but emergence.

You’re not watching a machine follow logic. You’re watching meaning emerge from billions of numbers reacting to your words in mathematically meaningful ways.

That’s why AI seems like it’s thinking. But it’s not. It’s predicting.

It’s a glorified guessing machine at scale.

It’s easy to say: “AI is just math.” But that undersells the beauty of it. It’s the kind of math that lets a machine finish your sentences, translate poetry, or debug your code. It’s the kind of math that turns questions into probability puzzles: 

“What would a human most likely say next?” 

That’s it. Not thought. Not meaning. Just patterns. Just logic. Just switches. 

Okay, but why is it getting so good?

Because we’re training it better. Modern AI is built on three big upgrades. Better data: diverse, curated, high-quality content. Smarter architectures: like Transformers, which allow models to “attend” to the most relevant context. Human feedback: where people rate responses, and the model learns what we prefer. 

It’s like raising a child who’s read a billion books and been corrected a billion times. That’s why today's AI feels sharp. Not because it understands, but because it’s memorised enough to approximate understanding. 

So where are we headed?

We’re now on the edge of something significantly more capable. AI is no longer just responding to prompts. It's beginning to reason, plan, and act.

We're stepping into the era of AI agents:

• Tools that take action.
• Plan multi-step tasks.
• Use other tools.
• Reflect and improve.

You won’t just query AI. You’ll delegate to it. 

And the big question becomes: can we trust it? 

This is what platforms like Watsonx are quietly working on, placing equal weight on performance and responsibility. With built-in tools for governance, transparency, and control, Watsonx is helping businesses not just build AI, but build it right. Not just powerful AI, accountable AI.

If you’ve made it this far, here’s the biggest thing I hope you take away: 

AI isn’t magic. 

It’s just math. Just probability. 

But it’s also a mirror, reflecting the scale of our own language, logic, and limitations. 

So yes, it’s okay to be curious. It’s okay to be confused. 

But the moment you look under the hood, really look, you’ll see: 

It’s logic, at scale. 

And you can understand it.  

And that’s where we are 

From a basic calculator to AI agents in the cloud, this is the arc of modern computing. Now the shift is towards building smarter systems, safer systems, AI that doesn’t just respond, but reasons, AI that doesn’t just predict, but plans, and maybe, just maybe, AI that doesn’t just impress us, but earns our trust.