When machines first learned to read biometrics, it took a massive server farm, with many programs working together, to decipher a fingerprint. In the beginning it was actually faster to hand the prints to a human expert.
Then the hardware shrank, and a single standalone server could do the job. Over the next few iterations the code was streamlined again and again, until biometrics entered the commercial space.
Then something remarkable happened. Engineers compressed all of the biometric algorithms onto a single chip. From that point fingerprint security could be built into phones and devices, on standalone chips that needed no connection to the internet or the cloud.
The same thing happened when machines learned to read aloud, to understand human speech and to navigate roads, all offline. There is a pattern here.
Every new capability the machines achieve is soon shrunk down to the scale of a single chip, and it doesn't take long.
Today the frontier is large language models.
LLMs are the party trick of this decade. Philosophers dreamed of thinking machines for centuries, and when the public met ChatGPT at the end of 2022, the world changed for good.
The internet is now full of instructions on how to host your own LLM. Every corporation can have its own version of this intelligence, trained on its own data and able to make leaps of understanding from it. The companies that embrace this will find themselves with a better, cheaper product, ready for market at the right moment. That is the holy grail of business.
Lawyers who use this technology will be better lawyers. In fact, every skilled, knowledge-based professional will be better at their work with an intelligent assistant. Over the next few years this too will be optimised and compressed to a scale that seems impossible today.
At that point you can picture a tiny device that sees, speaks, understands and advises anyone doing almost anything. Take it a step further and all of that sits on a single chip used in every piece of electronics, period. Every device in the world would be able to see, hear, understand, speak and solve problems, across languages and across disciplines.
Whether these machines ever become self-aware doesn't matter here. That is a discussion for another day. The potential for war and conflict is obvious, but that won't stop the advance of AI either.
Our future will be a partnership between us and AI. That partnership could destroy the world as we know it. It could also create a world of plenty, of peace, and of the rule of reason.
Looking back, November 2026
Three years later, the pattern has held. Within months of this post, Google put a language model directly on a phone. In 2024 Apple built AI into the iPhone and Mac, and laptops began shipping with chips designed specifically to run AI locally. Private models trained on a company's own data are now ordinary business practice, and professionals in law, medicine and software rely on AI assistants every day.
The last step, a single intelligent chip in every device, hasn't arrived yet. The direction is clear, though, and it is moving faster than I expected when I wrote this.
M Parak
