Intelligence is the most misused or abused word of the decade. Here are few AI achievements: 10 years ago, every one of them would have seemed impossible.
Clear an exam, write code….
Translate across more than a hundred languages in a jiffy
Summarize a five 500 page report before you finish coffee
Recall almost every fact on earth
Hold a fluent, confident conversation on any subject
Now rate the achievements on 10, not for how impressive it looks, but for how much real intelligence it actually takes. Every item on that list is the same thing underneath. It is recall and pattern and prediction, done at a scale no human could ever match. Astonishing, useful, and still the same thing. It is memory and mimicry, wearing the costume of thought or thinking tokens.
Now look at what is not on the list:
A child reading the parents voice without words
A salesman seeing a deal go cold three sentences before the client says no
A mother knowing which cry means hunger and which cry means pain
No AI model can do these things.
You can taste a meal and judge it without knowing how to cook. You can sit in a plane at 40 k feet without knowing a thing about aeronautics. You can ask AI a question without knowing how it works. As a taster or a passenger, you are fine. Stay comfortable.
But the moment you do more than consume, everything changes. If you cook for others, you have to understand heat and salt and timing, and what an ingredient does under pressure. If you fly the plane, you have to understand the engine, or people fall out of the sky.
Right now, every leader is being told to stop being a passenger and start flying the plane. Build with AI. Deploy it. Put your hiring, your customers and your strategy in its hands. If you are going to fly this plane, you cannot treat the engine as a mystery. So let us open the engine.
We spoke long before we wroteStart with something we forget, because it sits too close to us to notice. Human intelligence was not built on text. It was built on voice. We spoke for thousands of years before anyone wrote a single word. Speech is so old it is wired into our biology, into the shape of the throat, into parts of the brain that do nothing else. Writing is barely five thousand years old. On the timeline of our intelligence, writing arrived yesterday afternoon.
A child shows you this every day. It understands the feeling in a voice long before it can read a word. It answers your tone before it answers your sentence. It knows love and fear and safety purely as sound. We do the same even now. In a conversation, you respond to the tone.
Text is a magnificent invention, but it is a compression. It keeps the words and quietly drops the voice that carried them. This is the problem sitting at the centre of modern AI. Every model we have built was raised entirely on text. We fed our machines the transcript of human thought, and we threw away the recording.
So look at the engine itself. In 1943, two researchers, Warren McCulloch and Walter Pitts, did something bold. They took the biological neuron, one of the most intricate objects biology has ever produced, and wrote it down as a simple equation. Add up the incoming signals. If the total crosses a line, fire. If not, stay quiet. In 1958, Frank Rosenblatt took that equation and gave it the power to learn. He called it the Perceptron. Every neural network since is a descendant of that Perceptron. Every model, every chatbot, every system we now call AI. Eighty years, one idea, scaled.
But look at what happened in 1943. A real neuron does far more than add and fire. It lives in time. It carries a charge of its own history. And it does one thing no equation has ever done. It is affected by what passes through it. Something happens to it. To fit the neuron into mathematics, McCulloch and Pitts had to leave that part out. They kept the half that calculates. They let go of the half that feels. That was the right call in 1943. It is the flaw nobody examines in 2026. We have spent eighty years scaling the half they kept, and almost no time on the half they let go.
Feeling is not a soft layer that sits on top of intelligence. Feeling is what drives it. We feel, and so we act. A jolt of fear pulls your hand off the stove before any conscious thought arrives. Curiosity, doubt, care and unease are not noise around intelligence. They are what sets it moving. Emotion is the outcome of feeling, never the cause of it. Take the feeling away, and an intelligent response has no reason to exist.
If that sounds like philosophy, here is something solid to hold on to. There is a rare condition called congenital insensitivity to pain. The people born with it cannot feel pain at all. Their nerves work. The signal arrives perfectly: this pan is hot, this cut is deep. The information is flawless. Only the feeling is missing. Knowing was never enough. The feeling was doing the real work. That is the gap. And it is the exact gap between what AI does and what intelligence is.
Does a mathematical neuron feel?A number goes in. A number comes out. Nothing happens in between. The equation has no inside. It cannot feel, because there is no one in there to do the feeling. The biological neuron is a different kind of thing. It does not just carry a signal. It undergoes one.
This is what philosophers have circled for centuries, under 2 heavy words, sentience and qualia. It is the plain and staggering fact that experience feels like something. That being you is not dark and empty on the inside. That there is a light on.
In the mathematical neuron, there is no light on. There is no inside at all. You cannot scale your way to an inside. No amount of arithmetic ever adds up to a feeling. So if intelligence runs on feeling, one practical question follows. Where does feeling live, and where could a machine actually find it? Not in text. By the time a feeling has been flattened into words on a page, most of it is already gone.
Feeling lives in voice. Listen to how someone says "theek hai". 2 small words. They can mean yes, genuinely, all is well. They can mean I don't agree, but I am done arguing. They can mean something is wrong, and I will not tell you what. The meaning was never in the words. It was in the tone, the pause, the rise and the fall. That is a person, feeling out loud, in real time.
This is the recording we threw away when we chose text. It is what a 4-year-old hears without effort, and what the largest model on earth cannot hear at all.
It is why, at Pradhi, we are making one specific bet. The next frontier models will not be built on ever bigger piles of text. They will be built on voice. Voice is not data about intelligence. Voice is where intelligence actually lives.
We built AI from half a neuron. The calculating half. And even that was enough to astonish the world. The other half, the half that feels, has not even been switched on. Half a neuron took us this far. Imagine what the whole one will do.