SEWO · Steerable Medicine World Model
World models are not for predicting —
they are for steering.
The doctor sets the direction. The model walks the path.
Five Checkpoints · Deductive Closed Loop · Every Step Auditable
Paradigm Shift
From predicting what disease you'll get, to steering where your health goes
Traditional AI gives you a probability. Steerable AI gives you a path — and shows you why every step matters.
◾ Prediction Paradigm
A probability. Then what?
✦ Steerability
A path. Every step explained.
Philosophical Metaphor
The Synergistic Wisdom of Rider and Horse
The rider does not control every muscle of the horse.
He simply sends directional signals through the reins.
The horse relies on its own robustness
to adjust its gait across rugged terrain.
Rider = Doctor
Stagewise directional signals
Horse = World Model
Inference robustness amid noise
Reins = SEWO
Closed-loop architecture with five checkpoints
Technical Backbone
Five checkpoints that make steering possible
Each step can be independently audited, challenged, and falsified.
CP1
State Representation
Aging is not a number — it's a vector
Can the model explain "what state is this person in"?
CP2
Capacity Quantification
Not "okay" — but "0.72"
Can remaining capacity of each module be quantified?
CP3
Intervention Response
"Effective on average" ≠ "effective for this individual"
Does the intervention have clear biological semantics?
CP4
Counterfactual Transfer
"What if we hadn't given this intervention?"
Can the model reason about trajectories that never happened?
CP5
Quality Control
Failure must be diagnosable and correctable
When expectations aren't met, can the problem be diagnosed?
Medical Endpoint
Steerable Precision Health
Steerable Precision Health
From passively predicting what disease you'll get, to actively steering which direction your health state transitions toward. Every intervention has a traceable direction; every deviation can be traced to its cause.
Clear Direction
Doctors provide directional signals. AI doesn't decide for you — you steer AI.
Step-by-step Audit
Five gates trace deviations. Errors aren't black boxes — they're diagnosable signals.
Individual Steerability
One framework adapts to each unique individual. "Effective on average ≠ effective for you" is structurally resolved.
Question it. Falsify it. Improve it.
The SEWO framework preprint is online. Every sincere challenge is the best peer review. We invite biomedical AI researchers, computational biologists, and clinical informaticists worldwide — test it, critique it, let it either stand or fall.
DOI: 10.20944/preprints202605.0366.v1
This is a preprint and has not been peer-reviewed. Conclusions may change in future revisions.
Get in Touch
Ideas, critiques, or suggestions? Write to us directly.