When I look at the frenzy around Artificial Intelligence today, I’m reminded not of Newton, but of Kepler.
That may sound strange, but stay with me. It explains why I believe our current neural networks— transformers, backpropagation, and all — are still trapped in the Keplerian era of pattern observation, not the Newtonian era of insight and causation.

🌌 Kepler’s Gift: The Patterns of the Sky

Johannes Kepler didn’t have Newton’s mathematics.
What he did have were Tycho Brahe’s meticulous planetary observations. Out of them, he extracted patterns — elliptical orbits, equal areas, harmonic ratios — elegant, empirical, accurate.

Kepler described what planets did, but not why.
It took Newton, decades later, to reveal the law that made Kepler’s equations inevitable: gravity. Newton transformed observation into understanding.

🧠 AI’s Keplerian Moment

Modern AI, despite all its computing power, is still at Kepler’s stage.

Our neural networks — whether used for image recognition, language models, or even demand planning — find intricate patterns in data.
They tell us what has happened and what might happen next with uncanny precision, but they don’t know why.

Just like Kepler’s orbits, they are superb approximations.
But like Kepler, they lack the universal principle beneath the curves.

⚙️ Transformers Aren’t Newtonian

The transformer architecture is often hailed as the breakthrough that made “reasoning machines” possible.
But in reality, it’s an engineering triumph, not a scientific one.

Transformers excel at discovering correlations in vast data — sequence attention, positional encoding, contextual weighting — but their “understanding” is statistical, not conceptual.
Backpropagation adjusts millions of parameters to reduce error, yet the model never forms an internal theory of the world.

It’s Kepler’s mathematics scaled to planetary-sized datasets — accurate, but not explanatory.

📊 Pattern Matching in the Real World

In my own work with demand planning and inventory optimization, I use neural networks to forecast patterns of consumption, production, and substitution across thousands of SKUs.

They work — sometimes brilliantly.
But I’m under no illusion that they understand demand. They recognize patterns — seasonality, promotions, lead times, elasticity — and extrapolate them.

They don’t grasp causation — why a sudden festival surge changes substitution behavior, or how geopolitical shifts ripple through supply chains.
That still needs human reasoning, domain judgment, and deterministic optimization layered on top.

AI today is an aid, not an oracle.

🔍 The Missing Newtonian Leap

Newton’s revolution was not about computation — it was about unification.
He found a principle that applied equally to an apple and a planet.
That’s what true intelligence requires: the ability to derive laws from patterns, to compress reality into generalizable understanding.

For AI, that would mean:

• Moving from correlation to causality.

• From data fitting to concept formation.

• From statistical prediction to theoretical reasoning.

• From energy-hungry simulation to efficient comprehension.

This is the Newtonian revolution intelligence still awaits.

⚡ Why We’re Stuck

We’re not short of brilliance; we’re short of incentive.
Keplerian AI scales with money — more data, more GPUs, more cloud credits.
Newtonian AI demands patience — theory, abstraction, interdisciplinary research across neuroscience, physics, and computation.

One can be industrialized; the other must be discovered.

Until that shift happens, we’ll keep calling better telescopes “smarter brains,” when all we’re really doing is observing more stars in higher resolution.

🧭 Closing Thought

Kepler helped humanity look upward; Newton taught us why the heavens move.
Likewise, today’s AI helps us see patterns; tomorrow’s must teach us why those patterns exist.

When that day comes, we’ll finally move from neural networks to laws of mind — from Kepler’s astronomy to Newton’s physics of thought.

Until then, what we call AI is not intelligence.
It’s the observation of orbits — elegant, useful but incomplete. And it took science 79 years to move from Kepler to Newton. Here challenge is far more profound and make even take 790 years. But till then we should give our best talent not to datacentre capex but original research because its not scaling but paradigm change (may be many of them to reach final goal).