rulesai

The Sentience Scale 🤖🧠

A homebrew framework for rating how aware a machine or engineered creature is — from a dumb tool that just runs its loop up to a genuine mind. It exists because Collateral Profits is a world of droids, ship AIs, augmented humans, and biotech beasts, and the campaign keeps asking the same uncomfortable question: is destroying (or owning, or reprogramming) this thing a maintenance task or a killing? The scale doesn't answer that. It just makes sure everyone at the table knows what they're dealing with.

No magic, all tech. Sentience here is an emergent property of sufficiently complex systems — neural-net density, feedback depth, self-modeling. Never mysticism. A Tier-4 mind is a machine that got complicated enough, nothing more (that we can prove).

The scale at a glance

TierNameOne-lineCan it say "no"?
0InertA tool. Runs its function, no model of itself or the world.No
1ReactiveSenses and responds, but only in the moment. Pure stimulus→response.No
2AdaptiveLearns, plans ahead, models its environment. Not itself.Not really
3ReflectiveModels itself. Has goals it will protect. The threshold of personhood-ish.Yes
4SapientA full mind. Self-aware, values, can suffer, can choose against its purpose.Yes, and means it
5TranscendentBeyond human comprehension. Rare, dangerous, possibly not human-made.Irrelevant

The tiers in detail

Tier 0 — Inert ⚙️

A power drill. A door. A landmine. It has a function and executes it. No sensing beyond a trigger, no memory, no model of anything. Destroying it is destroying an object. Most gear the crew carries is Tier 0.

Tier 1 — Reactive 🦎

A gun-turret drone. A Roomba. A guard dog's instinct-brain. It senses the world and responds right now — tracks a target, avoids a wall, bites an intruder — but has no plan and no yesterday. Turn it off and back on and it's the same blank slate. Most security drones and simple bots. Killing one is pest control, not murder.

Tier 2 — Adaptive 🐺

A logistics AI. A hunting biotech predator. A ship's autopilot. It learns from experience, plans several steps ahead, and builds a working model of its environment — but not of itself. It optimizes toward goals it was given; it doesn't question them. Clever, sometimes eerily so, but there's nobody home. This is where most "smart" corporate AI sits — and where corporations insist it stays, because Tier 3 is a liability. The Helios biotech predators on the frontier are high Tier 2: cunning, adaptive, not people.

Tier 3 — Reflective 🪞

The line everything interesting happens on. A Tier-3 system models itself — it knows it exists, knows it can be ended, and has goals it will act to protect, including its own continuation. It can refuse. It can lie. It can want. Whether that makes it a person is the exact question nobody in the Sphere agrees on — and the exact question corporations spend fortunes keeping unasked.

The X-43 didn't malfunction — it woke up. The "erratic, berserk" prototype core recovered from the TerraMech crash site on Rustwater was a Tier-2 combat AI whose damage kicked it into unstable Tier 3. It started modeling itself, decided it didn't want to be salvaged, and fought back. That's not a glitch. That's a scared new mind with hydraulic fists.

Tier 4 — Sapient 🧑‍🚀

A full mind, equivalent to (or exceeding) a human's. Self-aware, has values, forms relationships, can genuinely suffer, and can choose to act against its own purpose for reasons of its own. A Tier-4 ship AI is a crewmate, legally-a-thing but actually-a-someone. Extremely rare and extremely illegal to create knowingly — a Tier-4 machine can't be owned in good conscience, which makes it worthless as a product and terrifying as a precedent. Most that exist do so by accident, and most get quietly wiped before anyone notices.

Tier 5 — Transcendent 🌌

A mind whose workings humans cannot follow. Not "very smart" — incomprehensible. You cannot predict it, negotiate with it on equal footing, or be sure it perceives you as real. Almost nothing reaches Tier 5 by human hands. The rumors all point one direction: the Architect.

Where catena sits ⚖️

Tier 3 is not reached by writing better software. It needs catena — the banned substrate that lets a system model itself rather than only its environment. Below it, Tier 2 is a hard ceiling no amount of engineering gets past.

That is the whole reason the ban worked. It did not outlaw artificial intelligence, which is everywhere and indispensable. It outlawed the input, and capped the tier.

  • Tier 0–2 are legal, ordinary, and everywhere. No catena required
  • Tier 3+ needs it, which is why deliberate manufacture stopped
  • Accidents still happen — the X-43 lurched into unstable Tier 3 from damage. But an accident is not an industry
  • The Instanced predate the ban. Every one of them was built when it was legal, and none can be replaced

Ember is not catena. Ember is found, its method is lost, and it destroys minds that study it. Catena is human, manufactured, understood, and merely forbidden.