Why gates matter
AIHS is a multi-decade programme. Over that span, the field's assumptions, capabilities, and constraints will all change. A roadmap that does not have explicit reassessment points becomes either inflexible (locked in to early-phase assumptions) or directionless (continuously drifting without anyone responsible for confirming direction).
The three gates below are structured around the three questions on which AIHS strategic direction most depends. Each is a question that cannot be answered when funding starts but can be answered when the gate arrives. The work of each phase produces the evidence the next gate requires.
Gates are not pass-fail. The decisions are recalibrations — adjusting ambition to evidence — rather than continue-or-stop verdicts. The conservative branch at each gate is still a programme. A field that hits the conservative branch at all three gates produces transformative medical capability, just not the universal capability the original AIHS vision implies.
Gate 1 · End of Phase I · ~Year 5
Architecture viability assessment
Are the three buckets developing in genuinely coupled fashion, with integration patterns visible early — or in parallel silos with weak coupling?
This is the operational gate. It asks whether the field is structurally configured to integrate. Most of Phase I capability work is low-risk — the high-risk question is whether the institutional structure of the field will support what Phase II requires.
What evidence informs this gate: publication patterns (are cross-bucket collaborations producing joint papers?), funding patterns (are major grants going to integrated teams or single-bucket teams?), standards work (are shared standards emerging or are buckets establishing incompatible conventions?), and clinical pilot outcomes (do early closed-loop demonstrations work?).
Who decides: the funding agency or coalition of funders. This is fundamentally a programme-management decision, not a scientific one. The relevant scientific community can advise but the decision belongs to whoever is paying for the work.
What this gate is not: it is not a check on whether scientific breakthroughs have occurred. Phase I produces infrastructure, not breakthroughs. Confusing this gate with a science-breakthrough check would lead to premature pessimism.
Gate 2 · End of Phase II · ~Year 15
B3 viability assessment
Has causal disease modelling reached the level needed for meaningful counterfactual reasoning in patient care?
This is the most consequential gate in the programme. It determines whether full AIHS is plausible at all. The answer hinges on whether the central scientific challenge — causal modelling of complex biological systems — yields to the approaches being scaled in Phase II.
What evidence informs this gate: direct outcomes of causal-modelling research programmes — do the models produce reliable counterfactual predictions when tested against held-out interventional data? Performance on standardised benchmarks. Clinical pilot outcomes when causal predictions are used in patient care. Failure mode characterisation — when the models are wrong, are they wrong in predictable ways?
Who decides: the scientific community via consensus. Unlike Gate 1, this is fundamentally a scientific assessment. Funding agencies should defer to scientific consensus while ensuring the assessment is conducted transparently and against pre-registered criteria.
Why this gate is hard: "Meaningful counterfactual reasoning" is itself a judgement call. The honest threshold is whether the causal models, used in patient care, demonstrably improve outcomes compared to non-causal alternatives. This requires expensive clinical evaluation. Pre-registering the evaluation methodology before Phase II ends is essential.
The realistic outcome: Most likely a partial yes. B3 yields for well-characterised disease classes and remains limited for the long tail. This positions the field for productive Phase III work within realised scope, while continuing B3 research as a longer-horizon effort.
Gate 3 · End of Phase III · ~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 excellent specialised systems coordinated by expert clinicians. Phase III provides the first real evidence on this question at deployment scale.
What evidence informs this gate: head-to-head clinical outcomes comparing AIHS systems with specialised-system-plus-expert alternatives. Population-level health outcomes in regions that deployed AIHS first. Cost-effectiveness analysis. Equity outcomes — does integration help close access gaps or widen them?
Who decides: a combination of scientific community, clinical leadership, regulators, and funders. The question is partly scientific (does the integrated system perform better?), partly clinical (does it serve patients better?), and partly strategic (is the additional cost and complexity worth it?). No single constituency can answer it alone.
Why this gate is the hardest to set up: head-to-head clinical comparison requires that both approaches exist at scale and be deployed in comparable settings. Achieving this requires Phase III to deliberately maintain both integrated and specialised system development — which costs more and creates institutional tension. Without this discipline, Gate 3 cannot be evaluated honestly.
The honest gate strategy
A few principles for using gates well:
Pre-register the criteria. Decide what evidence will satisfy each gate before the work that produces that evidence is done. Post-hoc criterion adjustment is the most common failure mode of decision gates in research programmes.
Communicate uncertainty publicly. The conservative branch at each gate is still a successful outcome. Public communication should normalise this — treating the conservative branch as a programme success rather than a failure. The alternative is institutional pressure to declare optimistic outcomes regardless of evidence, which corrupts gate decisions.
Separate the decision from the decision-maker. The funding agency makes the resource allocation decision. The scientific community provides the evidence assessment. Mixing these (funder also assesses science, or scientists also make funding decisions) reduces gate quality.
Plan for partial outcomes. Each gate has explicit binary framing for clarity but reality will be a spectrum. Plan operational responses for partial-pass outcomes at each gate, not just full-pass or full-fail.
What the gates do not cover
The three named gates address the strategic direction of the programme. Several other review activities are necessary but operate on different timescales.
Annual portfolio review. Within-phase allocation decisions, individual grant performance, and operational adjustments happen on an annual cycle. Gates are for strategic direction, not operational management.
Safety incident review. Any significant safety event triggers immediate review regardless of phase. The gate structure is for planned milestones; safety review is event-driven.
External shock response. Major external events — a transformative new technology emerging outside the programme, a public-trust event affecting the field, a geopolitical shift — may require gate timing or framing to change. The gate dates are nominal targets, not delivery commitments.
Equity gap monitoring. Equity outcomes should be monitored continuously, not only at gates. By the time a gate arrives, equity gaps have had years to widen.