hawkes.aiKey ideas
- The one idea — The Cortex is Global-Workspace cognition: a main thread plus specialist Lobes (memory recall, plan updates, clarifying questions, rule checks) run in parallel and publish typed Notes into one shared workspace, with a small router triaging mid-turn input. Unlike isolated subagents, the Lobes share a single context.
- How it is shown — While the main thread answers you, five specialist Lobes light up in parallel, each publishing typed Notes into the same shared context.
- The trap to avoid — Picturing this as the isolated-subagent swarm from Act 08. It's the opposite: the Lobes share one live context (a cache breakpoint), so they reason over the same state, not separate ones.
- What it sets up — Six minds acting on one shared state is powerful — and dangerous unless every action is accountable. so every move is written to a tamper-evident ledger.
This AI architecture borrows a theory of human consciousness — a global workspace where parallel specialists post what they find into one shared context.
The one idea
The Cortex is Global-Workspace cognition: a main thread plus specialist Lobes (memory recall, plan updates, clarifying questions, rule checks) run in parallel and publish typed Notes into one shared workspace, with a small router triaging mid-turn input. Unlike isolated subagents, the Lobes share a single context.
While the system answers you, five other minds are already working. This is the Cortex: not one thread doing everything in sequence, but a main thread plus specialist Lobes running in parallel. What looks like a single mind on the surface is a small society of them underneath, several laboring at once while one talks to you. Here's how they cooperate. The specialists are called Lobes, each with one narrow job. One recalls relevant memory. One keeps the plan updated. One forms the clarifying questions worth asking. One checks rules and constraints. Five run alongside the main thread — six minds in total — and while the main thread composes your answer, all five are already at work on their own facet. How do they cooperate?
How it works — the demo
While the main thread answers you, five specialist Lobes light up in parallel, each publishing typed Notes into the same shared context.
Each Lobe publishes typed Notes into one shared workspace — not free-form chatter between strangers, but structured, legible entries on a common board every mind can read. The memory Lobe posts what it recalled; the rule Lobe posts a constraint; the plan Lobe an update. The workspace is the shared state; the Notes are how six minds think together. And new input that arrives mid-turn, while all this is running? A small, fast router triages it — deciding which Lobe or thread should attend and directing it there, so an interruption doesn't derail the system; it gets dispatched to the right specialist. The society stays coordinated in real time, absorbing new information without rebooting its train of thought. Here's the crucial part. Remember subagents from Act eight — each sealed in its own context, passing summaries across walls. The Lobes are the opposite: they share one context, through a cache breakpoint, so they all reason over the same live state, not private copies. That's a swarm of strangers relaying messages versus one mind with many faculties. Global-Workspace cognition.
The trap to avoid
Picturing this as the isolated-subagent swarm from Act 08. It's the opposite: the Lobes share one live context (a cache breakpoint), so they reason over the same state, not separate ones.
Why it matters — and what’s next
Six minds acting on one shared state is powerful — and dangerous unless every action is accountable. so every move is written to a tamper-evident ledger.
The trap: picturing the Cortex as the isolated-subagent swarm from Act eight. It's the opposite. A swarm is many separate agents in separate contexts, coordinating by passing summaries. The Cortex is many faculties sharing one workspace and one state. Not a crowd of strangers, but a single richer cognition — parallel but unified, thinking many things at once about one problem. So the Cortex is Global-Workspace cognition: a main thread and specialist Lobes publishing typed Notes into one shared context, with a router triaging mid-turn input — sharing a single context, unlike isolated subagents. But six minds on one shared state is powerful and risky at once. Who did what, and can you trust the record? Every action has to be accountable. Next: R1's cryptographic ledger.
This is one short episode in AI: Zero → Frontier, a step-by-step climb through how AI actually works. Each episode builds only on the ones before it.
Full transcript 3:00 of narration
While the system answers you, five other minds are already working. This is the Cortex: not one thread doing everything in sequence, but a main thread plus specialist Lobes running in parallel. What looks like a single mind on the surface is a small society of them underneath, several laboring at once while one talks to you. Here's how they cooperate.
The specialists are called Lobes, each with one narrow job. One recalls relevant memory. One keeps the plan updated. One forms the clarifying questions worth asking. One checks rules and constraints. Five run alongside the main thread — six minds in total — and while the main thread composes your answer, all five are already at work on their own facet.
How do they cooperate? Each Lobe publishes typed Notes into one shared workspace — not free-form chatter between strangers, but structured, legible entries on a common board every mind can read. The memory Lobe posts what it recalled; the rule Lobe posts a constraint; the plan Lobe an update. The workspace is the shared state; the Notes are how six minds think together.
And new input that arrives mid-turn, while all this is running? A small, fast router triages it — deciding which Lobe or thread should attend and directing it there, so an interruption doesn't derail the system; it gets dispatched to the right specialist. The society stays coordinated in real time, absorbing new information without rebooting its train of thought.
Here's the crucial part. Remember subagents from Act eight — each sealed in its own context, passing summaries across walls. The Lobes are the opposite: they share one context, through a cache breakpoint, so they all reason over the same live state, not private copies. That's a swarm of strangers relaying messages versus one mind with many faculties. Global-Workspace cognition.
The trap: picturing the Cortex as the isolated-subagent swarm from Act eight. It's the opposite. A swarm is many separate agents in separate contexts, coordinating by passing summaries. The Cortex is many faculties sharing one workspace and one state. Not a crowd of strangers, but a single richer cognition — parallel but unified, thinking many things at once about one problem.
So the Cortex is Global-Workspace cognition: a main thread and specialist Lobes publishing typed Notes into one shared context, with a router triaging mid-turn input — sharing a single context, unlike isolated subagents. But six minds on one shared state is powerful and risky at once. Who did what, and can you trust the record? Every action has to be accountable. Next: R1's cryptographic ledger.