Humanity's search for artificial intelligence has been in service of these two qualities.
The axes so far
Two levels are named. Anything deeper still carries a handle — TC.1.1.2,
SA.2.1 — shown on the spoke and listed under the wheel as it appears, so the
structure stays arguable before the vocabulary is fixed.
- Task Complexity
- Actor Capability — Observability · Decision Authority
- Environment Complexity — Horizon · Dynamics
- System Autonomy
- Data Availability
- Self-supervising
Names attach to position, so renaming one never disturbs the rest, and an unnamed axis keeps working underneath a named parent.
Two methods on it
VLA takes pixels and words to motor commands, learned from human demonstrations. RL takes a reward and a rollout budget and finds the behavior by trying. They are close to opposites on this wheel: VLA reads a rich, partially observed world far better than RL does, and RL supplies its own experience where VLA has to be handed every hour of it.
A reading is a claim about how much of one axis a method absorbs, and it is made
at whatever depth the wheel is split to — Observability is a sharper claim than
Task Complexity. Drag a handle and the axes underneath it inherit the new
value; split further and you can disagree with yourself in more detail. Nothing
is saved: copy readings exports the whole set, ready to paste back over
METHODS in autonomy-radar-data.js.
Problems (loco-manipulation) come next, on the same wheel.