hawkes.aiKey ideas
- The one idea — PERSYS organizes memory and attention around active concerns — persistent goals it's tracking — instead of just storing and fetching on request. Concerns act as a top-down lens: they bias what it remembers and what it notices, giving the system standing reasons to act.
- How it is shown — The same memory store queried with different concerns active — returning different memories each time, as the concerns tilt what surfaces toward what currently matters.
- The trap to avoid — Treating memory as passive storage — store, index, fetch on request. You get a search engine that never prioritizes and never surfaces anything unprompted, not a mind.
- What it sets up — Concerns give PERSYS standing reasons to act — but with many alive at once, which one wins attention NOW? it doesn't guess, it computes urgency.
AI memory is usually a filing cabinet — store, index, fetch. But a filing cabinet never reminds you of anything. Caring is what turns storage into a mind.
The one idea
PERSYS organizes memory and attention around active concerns — persistent goals it's tracking — instead of just storing and fetching on request. Concerns act as a top-down lens: they bias what it remembers and what it notices, giving the system standing reasons to act.
Most memory systems just store. You put something in, you ask for it, you get it back — a filing cabinet. PERSYS doesn't work that way. It decides what to care about, and that changes everything. At the center of PERSYS is a set of concerns — persistent, named things it's trying to do or watch: a deadline you mentioned, a bug it hasn't closed, a question left open. A concern isn't a query you type; it's a standing preoccupation, each carrying a sense of how far it is from where it wants to be. Those concerns reshape memory. Ask a plain database a question and it returns the same rows every time. PERSYS retrieves through its active concerns — they act as a lens, tilting what surfaces toward what currently matters.
How it works — the demo
The same memory store queried with different concerns active — returning different memories each time, as the concerns tilt what surfaces toward what currently matters.
The same store, queried with a deadline live, hands back different memories than it would when that deadline is closed. Concerns steer attention too. When new input arrives, PERSYS doesn't weigh it neutrally — active concerns amplify whatever touches them and mute what doesn't. A stray line in a document that bears on an open worry jumps out; the same line, with no concern to catch it, slides past unnoticed. Attention is concern-shaped. Here's why that matters. A store that only answers when asked can never act on its own — it waits, forever passive. Give the system standing concerns and it suddenly has reasons of its own: something to check, something to connect, something worth raising before you ask. Concerns are what turn a memory into a mind with an agenda.
The trap to avoid
Treating memory as passive storage — store, index, fetch on request. You get a search engine that never prioritizes and never surfaces anything unprompted, not a mind.
Why it matters — and what’s next
Concerns give PERSYS standing reasons to act — but with many alive at once, which one wins attention NOW? it doesn't guess, it computes urgency.
The trap is treating memory as passive storage — store it, index it, fetch it on request. Do that and you've built a search engine, not a mind: it never prioritizes, never notices, never brings anything up unprompted. Concern-driven cognition is the difference between a system that holds information and one that has stakes in it. So PERSYS runs on concerns, not queries — they bias what it remembers and what it notices, and they give it standing reasons to act. But with many concerns alive at once, which one wins your attention right now? PERSYS doesn't guess. It computes urgency. Next: urgency is a formula.
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 2:45 of narration
Most memory systems just store. You put something in, you ask for it, you get it back — a filing cabinet. PERSYS doesn't work that way. It decides what to care about, and that changes everything.
At the center of PERSYS is a set of concerns — persistent, named things it's trying to do or watch: a deadline you mentioned, a bug it hasn't closed, a question left open. A concern isn't a query you type; it's a standing preoccupation, each carrying a sense of how far it is from where it wants to be.
Those concerns reshape memory. Ask a plain database a question and it returns the same rows every time. PERSYS retrieves through its active concerns — they act as a lens, tilting what surfaces toward what currently matters. The same store, queried with a deadline live, hands back different memories than it would when that deadline is closed.
Concerns steer attention too. When new input arrives, PERSYS doesn't weigh it neutrally — active concerns amplify whatever touches them and mute what doesn't. A stray line in a document that bears on an open worry jumps out; the same line, with no concern to catch it, slides past unnoticed. Attention is concern-shaped.
Here's why that matters. A store that only answers when asked can never act on its own — it waits, forever passive. Give the system standing concerns and it suddenly has reasons of its own: something to check, something to connect, something worth raising before you ask. Concerns are what turn a memory into a mind with an agenda.
The trap is treating memory as passive storage — store it, index it, fetch it on request. Do that and you've built a search engine, not a mind: it never prioritizes, never notices, never brings anything up unprompted. Concern-driven cognition is the difference between a system that holds information and one that has stakes in it.
So PERSYS runs on concerns, not queries — they bias what it remembers and what it notices, and they give it standing reasons to act. But with many concerns alive at once, which one wins your attention right now? PERSYS doesn't guess. It computes urgency. Next: urgency is a formula.