At home we’re eloquent speakers. We’ve got the gift of the gab. Quick thinkers, fast talkers.

That’s what always amazes me when I’m with my family in New Zealand.

Being well-spoken is considered a measure of intelligence. It might seem only natural that AI would choose language as its starting point.

But it didn't. Many attempts were made but failed to deliver the sacred ingredient: a real thinking being in a box.

Until quite recently. Oh, someone thought, what if we built a system that could predict the next syllable, the next word, that would occur in response to a question? Bypass those high-brow notions. Just throw the entire internet into a prediction engine.

And it works. Creates science papers, computer code, tells you who to date. Very much like there’s a real thinking being in the box. But there isn't.

We’re a small software company. To code, we use coding agents. A specialised AI for software.

A coding agent has read all the code on the internet. Well, I should put that in quotes, “read”. Ingested its patterns. Like all language-based AI, it’s a giant pattern recognition machine.

It’s incredibly helpful. It speeds up coding immensely, I couldn’t say by how much, ten times? Twenty? But to say it's intelligent?

Myself, I spent 25 years learning how to code. But more importantly, WHAT to code.

An example: my first real project. Where I worked, there was narrow-casting system. Basically a slide deck on a tv screen.

The person in charge of the system was set to leave, I volunteered to take over. But also to bring it to the screens everyone was sitting at. An intranet.

I started by moving through the company, looking at how people were working, talking to them. When I started coding, I did it bit by bit, first facebook-like, later adding more.

The point being, I was learning WHAT to build by moving through the company.

Movement is the starting point for a radically different approach to AI by Jeff Hawkins and his team at Numenta.

Their inspiration is the human brain, not the language produced by it. Specifically, part of the human brain called the neocortex.

Within the neocortex, hundreds of thousands of cortical columns each create models of the world. Models that have as input our senses: smell, sight, hearing, touch. Models that predict the movement of the body in relationship with the environment, the movement of bodies in that environment. If reality doesn’t match, the models adapt. Learning.

Fed by thousands of multi-modal models working together, what we call intelligence is a picture, or maybe better, a continuously self-learning movie. Intelligence that expresses itself in many forms: music, dance, design, art. And yes, in language and computer code. Like my first project.

Again, we all know language-based AI cosing agents are extremely useful. But it'll never learn WHAT to code. For that, you need true intelligence.