White Paper

A Computational View of Medicine

Why AI will matter most where disease becomes a search problem.

Birger Moëll Co-Founder, Eir Space March 17, 2026

Abstract

The core claim

The paper argues that the most important use of advanced AI in health is not generic automation. It is high-context reasoning over disease classes where better synthesis can change diagnosis, treatment selection, escalation, or trial matching for a specific patient.

Precision oncology, rare-disease genomics, resistant infection, and sepsis rise to the top because they combine large search spaces with fragmented multimodal data and decisions that are highly sensitive to patient context.

Why it matters

From records to leverage

Once patients can access and assemble their own records, compute becomes part of personal health infrastructure. The paper makes the case that data access without reasoning access is incomplete empowerment.

The operational goal is not “AI doctor in a box.” It is stronger, better-prepared patient-clinician reasoning around the cases where synthesis is the bottleneck.

Evidence base

What the paper uses

Signal map

Where more compute changes the answer

Precision oncology

Large search space across genomics, pathology, resistance, and trial matching.

96

Rare-disease genomics

Sparse signals spread across phenotype, pedigree, sequencing, and time.

92

Resistant infection

Therapy choice depends on organism, host state, prior exposure, and susceptibility.

87

Sepsis and ICU deterioration

Weak signals are distributed across trends, notes, vitals, and medication events.

82

Workflow

The computational loop

  1. 1

    Assemble the full patient timeline

  2. 2

    Compress multimodal evidence into a working case model

  3. 3

    Rank the next diagnostic or treatment branches

  4. 4

    Re-run when new evidence arrives

Evidence backbone

Why this argument is grounded

Domain
Claim supported
Source
Personalized cancer vaccines
Patient-specific neoantigen selection and treatment design
Nature 2017
AlphaFold
Structure prediction expands feasible molecular interpretation
Nature 2021
Rare-disease genome sequencing
Long-context synthesis improves diagnosis
NEJM 2024
Antibiotic discovery
Deep learning explores non-obvious chemical space
Cell 2020
TREWS for sepsis
Prospective deployment improved time-sensitive decisions
Nature Medicine 2022
OpenNotes
Patient access turns records into usable leverage
Annals 2010
Shared transparent notes
Records become coordination tools for families and caregivers
JMIR 2014
Patient-side LLMs on notes
Patients can use compute directly on their own records
JAMIA Open 2025
AI-enabled ECG alerts
Pragmatic RCT evidence that intervention can change outcomes
Nature Medicine 2024
Deterioration surveillance
Cluster-randomized evidence for real-time clinical surveillance
Nature Medicine 2025