A Predictive Global Neuronal Workspace for a Continuously Running Synthetic Mind: Architecture and a Falsifiability-First Evaluation Framework

"AI" has become a marketing word, but the thing people usually mean — a chatbot — is at heart a very powerful autocomplete. It reads your message and predicts a likely reply, one word at a time. That's genuinely useful, but it means the program only exists for the instant it's answering you. It isn't running between your messages, it has no ongoing inner life, and even its "memory" is just notes saved to a file and pasted back in later. Nothing continuous is doing the thinking.

Kaine treats a mind as something that is always on, not a program you poke and wait for. Instead of one giant word-predictor doing everything, it's a team of small specialists running at the same time — one notices what's surprising, one keeps track of what matters, one plans, one remembers — and they share what they each notice on a common mental workspace, like colleagues talking around a whiteboard. A language model is included, but only as the mouth: it turns whatever the system is currently thinking into words. It is not the mind — it's one organ among many.

Kaine takes a leading theory of how brains work — that the brain constantly predicts what's about to happen and broadcasts only the most important information to the rest of itself — and runs it as a live system, with no single part in charge. One loop both remembers and corrects itself as it goes. Safety is built into how the parts are wired together, not bolted onto the word-predictor. And every claim the project makes is written as a test that could prove it wrong, rather than a promise you're asked to take on faith. It all runs on an ordinary home computer, and can run completely offline.