AI cannot violate Natural Intelligence
Green chemistry principles hold
Biomimicry is the key.
My nature guide and critic Yogita Shukla sent me the book “nature intelligence”. I used Ai to summarise it. But it got me thinking.
Natural intelligence represents the intelligence of life itself. It is the art and the science of leaving the Earth healthier, wealthier, more vibrant and more viable than before.
Whales 🐋 cool the climate. Fungi 🍄 make rain. Termites 🐜 green deserts. The species that endure are not the strongest or the smartest, but those that contribute to the greater whole – they are world-makers. This speaks to the fact that living systems create the opposite of entropy: they create more order, more integration, and more capability.
The American model of ai is contrary to the basic principles of natural intelligence. They would have us believe that the only way to do it is to build 1 or 5 GW data centres which cost Rs 5 lac crores a piece. And probably won’t generate even a fraction of this as revenue in their lifetime. Over eager state government and under educated industrialist want to copy these in India. That too in a capital starved economy. Most equipment has to be imported. This doesn’t work.
We can set up a data centre or two to learn. But it just doesn’t make sense to become a meal ticket for Nvidia esp since we know Nvidia challengers are around the corner. And that the photonic fabric data centres will consume 95 or even 99% lower energy. Bigger is not better. And rarely more sensible. We don’t have to compete with the Global north on everything. Most certainly not on losing money.
India has to re invent ai data centres. And use biomimicry for it. Everything in nature works at room temperature and pressure. Systems are distributed, not centralised.
The Nvidia /Open AI model is flawed. We have to wait it out till the flaws get revealed and then get resolved. We should not waste scant national resources on importing GPUs. We have to wait for an indigenous GPU to be available.
We currently have three contenders – Shakti from iit Chennai, iQ from iit bombay and Ko from iit delhi. All three take a different approach. And no one has been able to think of a fourth approach.
Shakti is a computing chip – regular cpu – low power, high GHz, narrow tracks (5 nm). It needs colocated live memory (we have a global memory shortage looming). Shakti is a multipurpose chip so maybe we can make laptops with it as well.
IQ takes old 45 nm technology and integrates it with compound semiconductors to get photonics on board inside a 3D package. The low cost high performance architecture has major potential. But is 18 months away.
Ko uses silicon photonics. Something intel struggled with but failed. It violates basic laws of physics. But who knows.
All we know is that by August 2027, we should have an indian gpu solution. How good and how expensive – one will have to wait.
