SignalGraph← kjames.dev

SIGNALGRAPH

AI is fast. Most places are not.

Published October 14, 2026 by James Kamugisha

Four people examine a cola bottle in a still from The Gods Must Be Crazy.
A still from The Gods Must Be Crazy (1980).

In The Gods Must Be Crazy, a cola bottle falls from a plane into a Kalahari community. People discover many uses for it. Soon the one bottle becomes a source of quarrels. The film is a fable, not a guide to real communities. But it leaves me with a question: what happens when something useful arrives before anyone has worked out how it fits into everyday life?

AI is moving from screens into physical life. A model can be impressive in a demo. Whether it helps someone depends on the systems around it: reliable power, connectivity, usable data, skilled people, maintenance and clear rules.

NVIDIA says its Cosmos world models can generate simulated environments to help develop robots and autonomous vehicles. Those worlds may help a machine rehearse conditions it will later face. Simulation cannot show that it will work safely on a particular road, or tell us who will repair it when it fails.

Imagine a delivery robot in a neighborhood without consistent addresses. It might navigate using maps and landmarks. But reaching the right person is another problem. Who knows the route, handles a blocked path, keeps it powered, repairs it and takes responsibility if something goes wrong? Those are not details to add after the model arrives. They are part of the solution.

A yellow rover with binocular eyes and a paper map strapped to its side, stopped on a red dirt road through hilly farmland at sunset.
Illustration: a delivery robot navigating a rural road.

Public spending shows the same dependence on surrounding systems. AI may help flag unusual procurement patterns when records are available. But someone must check what it finds, and institutions must be willing to act. A flag alone cannot produce accountability.

None of this means waiting for perfect infrastructure. It means starting with a problem people name locally, perhaps moving produce after harvest, and asking what the whole solution needs. Build the skills, maintenance and rules alongside the tool. Test it in the place it will serve.

A model does not arrive in an empty world. That is what the bottle makes me ask: what must be built around the next technology, and who gets to decide?

What would physical AI need around it to help where you live?