Phase III — Integration and Scale

Years 15–30. Full AIHS prototypes for select clinical applications. The phase where the integration question stops being theoretical — either the buckets work together at scale and deliver clinical value, or they don't.

Years 15–30

Risk profile High

Decision gate Integration value

Strategic objective

Phase III brings full AIHS prototypes online for select clinical applications. The capability envelope by end-of-phase: AIHS systems providing meaningful coverage for trauma, sepsis, select cancers, and regenerative medicine indications. Coverage is incomplete — Phase III does not deliver universal capability — but it is integrated, autonomous within validated scope, and clinically deployed.

The strategic question this phase answers is whether integration produces clinical value beyond what specialised systems alone would. If specialised diagnostic AI plus specialised therapeutic platforms plus expert clinical teams are equivalently effective to integrated AIHS systems, then integration is not justified at this scale. If integrated systems clearly outperform the unintegrated alternative, the AIHS architecture vindicates itself and Phase IV proceeds.

This phase also bears the brunt of regulatory maturation, public adaptation, and equity infrastructure work. Capability without those is undeployable.

Bucket A · Diagnostic priorities

A2 (whole-body molecular mapping) reaches clinical readiness for at least partial implementations. Full single-cell whole-body mapping remains aspirational. The Phase III deliverable is whole-body imaging at organ-and-tissue resolution combined with single-cell sampling from accessible compartments — sufficient for AIHS-grade diagnosis in most contexts, even if not the full specification.

A1 (in vivo single-cell readout) extends to deep internal tissues. The technical maturity reached in Phase II for accessible tissues now extends to liver, kidney, lung, brain. By end-of-phase, in vivo cellular profiling at clinical scale exists for most major organ systems.

A5 (distributed pathogen and damage detection) matures into routine clinical capability. Engineered sentinel cell platforms cover broad pathogen classes plus damage detection. Phase III deliverable: comprehensive surveillance capability competitive with the native immune system for known threat classes.

A3 and A4 are largely mature by Phase III start and continue incremental improvement rather than fundamental advance.

Bucket B · Interpretation priorities

The Phase II gate determines what Phase III Bucket B work looks like. The assumption here is that B3 yielded at least partially.

B3 (causal disease modelling) reaches deployment maturity for the disease classes where it has yielded. The Phase III work is extending the causal-modelling capability to additional disease classes — likely a long-running effort that continues into Phase IV. By end-of-phase, causal counterfactual reasoning should be available for major disease categories representing the bulk of clinical need.

B4 (digital twins) approaches whole-patient scope. Multi-organ integrated digital twins with intervention-simulation capability become available for major disease categories. Whole-patient twins remain aspirational but partial whole-patient twins (covering most clinically relevant systems) are achievable.

B2 (real-time epigenomic state inference) reaches clinical use, leveraging mature Bucket A in vivo readout capabilities. By end-of-phase, real-time chromatin state monitoring is available for at least cancer surveillance and possibly for broader cellular state assessment.

B1 and B5 are largely mature by Phase III start.

Bucket C · Therapeutic priorities

C3 (architectural tissue reconstruction) reaches first clinical applications. The Phase III deliverable is in vivo regeneration for at least one organ type in routine clinical use — likely kidney, liver, or peripheral vasculature given the current research trajectory. Complex organs (heart, lung, brain) remain experimental.

C4 (neural reconnection) reaches first clinical applications for peripheral nerve injury. Spinal cord and CNS applications remain partial — functional improvement in some indications, full restoration in none. This is consistent with the bucket-C assessment that C4 is the longest-horizon advance.

C1 (multi-target coordinated intervention) matures to platform status. By end-of-phase, multi-target coordinated therapy is the default mode of treatment for complex diseases rather than the exception. The architecture of platform-orchestrated combinations is established clinical practice.

C7 (accelerated healing energy supply) sees serious work in Phase III. The thermodynamic constraints identified in Phase II yield to a combination of metabolic substrate optimisation and metabolic-rate modulation. Phase III delivers accelerated wound healing capability — not the science-fictional rapid regeneration, but a 2–3× speedup over normal healing rates for specific applications.

C2, C5, and C6 are largely mature by Phase III start and continue refinement.

Integration milestones

Full AIHS prototypes for trauma, sepsis, select cancers. By end-of-phase, AIHS systems meeting all four criteria from the definition are in routine clinical use for at least three major indication categories. These are partial AIHS in coverage (not all diseases) but full AIHS in capability (all four criteria met within their scope).
Established regulatory frameworks for autonomous closed-loop care. FDA and international equivalents have mature, predictable approval pathways for AIHS-class systems. New AIHS approvals proceed through established processes rather than novel case-by-case review.
Equity infrastructure for AIHS deployment. Deployment models that serve poorly-resourced settings exist alongside commercial deployments. Specific design choices in earlier phases bear fruit here — open data standards, federated infrastructure, public-sector deployment options.
Cross-jurisdictional governance frameworks. The first international agreements specifically addressing AIHS-class autonomous medical systems. Comparable in significance to the early agreements on human gene therapy or recombinant DNA research.
AIHS capability for major chronic diseases. Stretch goal: extension of AIHS scope beyond acute conditions to chronic disease management at population scale.

Funding allocation profile

Phase III · Approximate allocation

A · 18%

B · 25%

C · 28%

Safety · 14%

Integ · 15%

Diagnostics

Interpretation

Therapeutics

Safety architecture

Integration

Phase III allocation shifts toward Bucket C as the longest-horizon therapeutic work approaches deployment, and toward integration as full AIHS prototypes require it. Safety architecture remains a substantial fraction — the deployed systems are now powerful enough that safety failures have serious consequences.

Dominant risks during Phase III

Scope creep. Once AIHS systems demonstrate value in narrow indications, institutional pressure to expand scope outpaces validated capability. Phase III discipline about staying within proven envelopes is critical and politically difficult.

Public-trust shocks. Phase III is when AIHS systems become visible to patients and the public. A high-profile failure — adverse event, equity scandal, misuse incident — can substantially set back the field regardless of underlying technical health.

Geopolitical fragmentation. AIHS capability has obvious military, geopolitical, and economic implications. National-level competition could fragment the field, reducing federated learning effectiveness and creating incompatible standards. Mitigated by international governance work but not eliminated.

Equity gap widening. Without deliberate counteraction, AIHS benefits concentrate in wealthy health systems. Phase III is when this gap becomes most visible and most politically consequential.

End-of-Phase III Gate · ~Year 25

Integration value assessment

Has integration of the three buckets produced clinical value beyond what specialised systems alone would produce?

The architectural premise of AIHS is that integration matters — that closed-loop autonomous platforms outperform the alternative of excellent specialised systems coordinated by expert clinicians. Phase III provides the first real evidence on this question at deployment scale.

If integration delivers value

Proceed to Phase IV pursuing universal AIHS capability. Continue investment in the longest-horizon advances (C4 CNS applications, A2 single-cell whole-body mapping).

If integration is marginal

Pivot toward maximising specialised system capability rather than further integration investment. Re-evaluate the AIHS architecture itself.

Research Questions

Open questions from the field

Detail-level questions a researcher or programme planner landing on this page would likely ask. Expand each to see the answer.

Phase III tests whether integration beats specialised systems — how is that comparison actually made?

The strategic question Phase III answers is whether integrating the three buckets produces clinical value beyond what specialised systems alone would deliver. This is a real question with a possible negative answer, and the phase is designed to test it rather than assume it.

The comparison is between two deployment models for the same clinical problem. The specialised model: best-in-class standalone diagnostic AI, plus standalone therapeutic systems, plus a human clinician coordinating between them. The integrated model: an AIHS providing the same functions as a coordinated closed loop with autonomous operation within validated scope. For indications like trauma, sepsis, select cancers, and regenerative medicine, Phase III deploys integrated systems and measures whether they outperform the specialised-plus-human baseline on outcomes, speed, and consistency.

If integration wins clearly, the case for full AIHS is made and Phase IV pursues it. If the specialised-plus-human model matches the integrated one, that is a genuine finding: it would mean the integration premium is not worth its cost and complexity, and the end-of-Phase-III gate would redirect investment toward the specialised systems that demonstrably work. The honest framing is that integration is a hypothesis Phase III tests, not a foregone conclusion — the closed-loop speed and coordination advantages are plausible but must be demonstrated against a strong baseline.

What makes Phase III's autonomy 'within validated scope' different from the universal autonomy of the full vision?

Phase III delivers full AIHS prototypes that are autonomous, but only within validated scope — and that qualifier is the entire difference between a deployable Phase III system and the universal capability that may or may not arrive in Phase IV.

Validated scope means the system operates autonomously only for the indications, patient populations, and intervention types where it has been explicitly validated and where its causal models are trustworthy. A Phase III sepsis system runs autonomously for sepsis in the populations it was validated on; confronted with an unusual presentation, a rare comorbidity, or an out-of-distribution patient, it falls back to the advisory mode and routes the decision to a clinician. The scope is a fence, and the system knows when it is at the fence.

This is what well-calibrated uncertainty (A4) buys and what the safety architecture (C6) enforces — the system's autonomy is bounded by its own knowledge of where it is reliable. The universal-capability vision would extend that fence to encompass most of medicine, which depends on the bottleneck advances (especially B3) generalising far beyond well-characterised conditions. Phase III deliberately does not assume that generalisation; it deploys real autonomous systems inside conservative fences and lets Phase IV determine how far the fence can be pushed.

Coverage in Phase III is explicitly incomplete — how are the priority indications chosen?

Phase III targets trauma, sepsis, select cancers, and regenerative-medicine indications rather than attempting broad coverage, and the selection logic is the same logic that runs through the whole study: go where the integrated approach has the clearest advantage and the science is most ready.

The priority indications share several features. They are high-value — conditions where better, faster, more coordinated intervention saves lives or function at scale. They are time-critical or coordination-intensive — sepsis and trauma in particular reward exactly the fast, integrated, multi-system response that an AIHS provides and that human coordination struggles to match. And they sit in domains where the underlying science is relatively tractable — the causal mechanisms are comparatively well understood, so partial B3 capability is sufficient.

Conditions are deliberately deferred when they depend on the hardest bottlenecks: complex CNS restoration waits on C4, large-volume regeneration on C3 and C7. The phase is structured so that its deployed systems demonstrate integration value in the most favourable, highest-impact contexts first — building the clinical evidence and public trust that any expansion to harder indications in Phase IV would require. Incomplete coverage is a feature of disciplined sequencing, not a shortfall.

Referenced from across the study

Topic Roadmap — → Full Phase III detail with funding allocation, risks, and decision gate Phase IV — Maturity (Years 30+) The fourth phase is conditional.

Continue

Next phase

Phase IV — Maturity

Years 30+. Either AIHS reaches universal capability or accepts a sophisticated partial-system ceiling.

Back

Phase II — Capability Building

The capability work that this phase scales to deployment.

Sideways

Decision gates

All three review moments compared and consolidated.