Explore an example
An interactive, clickable example of the GGKS research loop and its process map, using entirely invented data. Nothing here is the real Studio.
Private research prototype · bounded demonstration
Governed Genomic Knowledge
Can scientific AI predictions become inspectable, governed knowledge without being mistaken for clinical truth?
The Genesis Governed Genomic Knowledge System (GGKS) is a research demonstrator for governing the passage from external genomic prediction to evidence-bearing, provenance-preserving and explicitly bounded scientific standing. AlphaGenome is the candidate first provider. Its output enters Genesis as a prediction — not as evidence, accepted knowledge or clinical authority.
Open the governed research studio → How GGKS works → Profiles & access →
Take the guided tour
A twelve-screen narrated presentation explaining the GGKS Studio, chapter by chapter, with clear labels for current, synthetic, and planned behavior.
Core proposition
AlphaGenome predicts. Genesis would govern whether that prediction has earned standing.
AlphaGenome predicts possible molecular effects of genomic variation. Genesis AiX would govern whether a prediction has acquired sufficient evidence, corroboration, and bounded standing for an explicitly permitted research use.
1. The scientific-governance problem
A powerful prediction is still only a prediction.
Scientific AI can generate increasingly powerful predictions about genomic variation. But a prediction is not evidence, scientific consensus, Canon admission, or clinical authority — regardless of how the prediction was produced.
Model confidence is not truth
A high-confidence score describes the model's own calibration, not an external, corroborated fact about biology.
Provider prestige is not authority
The reputation of the organization that produced a prediction does not itself establish that the prediction is correct, admitted, or safe to act on.
2. The Genesis distinction
Six invariants a governed system must never blur.
These distinctions are the load-bearing structure of the whole proposal. Each one names a step that a prediction does not get to skip.
Prediction ≠ evidence
A model output is a claim to be examined, not evidence in itself.
Evidence ≠ corroboration
Supporting material must still be weighed against independent, conflicting, and qualifying evidence.
Corroboration ≠ standing
Being well-supported is not the same as having been assigned a governed standing.
Standing ≠ universal authority
A standing is bounded to a scope; it does not become authority for every downstream use.
Scientific standing ≠ clinical authority
Nothing in this system would confer diagnostic, treatment, or clinical decision authority.
A successful call ≠ admitted knowledge
A successful provider call creates a candidate prediction, not admitted knowledge.
Controlling lifecycle
Prediction → Evidence → Inspection → Corroboration → Standing → Authorized Use
This is a lifecycle of gates, not a pipeline that runs on its own. Movement between stages is never automatic — each stage requires its own separately determined outcome, and a prediction can stop, stall, or be rejected at any stage.
1
Prediction
Versioned synthetic provider candidate and receipt; live AlphaGenome remains separately disabled
Synthetic path implementedVersioned synthetic provider candidate and receipt; live AlphaGenome remains separately disabled
2
Evidence
Typed synthetic evidence remains distinct from provider output and interpretation
DemonstratedTyped synthetic evidence remains distinct from provider output and interpretation
3
Inspection
Three-plane pre-call meaning/request inspection and post-call object/provenance inspection
Interactive prototypeThree-plane pre-call meaning/request inspection and post-call object/provenance inspection
4
Corroboration
Competing support/conflict branches are preserved; real independent evidence is not yet connected
Synthetic demonstrationCompeting support/conflict branches are preserved; real independent evidence is not yet connected
5
Standing
Candidate, reviewed and revalidation-required states are enforced in the private prototype
Prototype enforcementCandidate, reviewed and revalidation-required states are enforced in the private prototype
6
Authorized use
Gateway policy, institutional role, purpose attestation, quota and clinical refusal produce an inspectable receipt
Synthetic-only enforcedGateway policy, institutional role, purpose attestation, quota and clinical refusal produce an inspectable receipt
3. Proposed bounded workflow
Every provider call would pass through governed inspection on both sides.
AlphaGenome may be identified as the proposed first provider adapter — not as a partner, endorser, certifier, or installed integration. No such integration exists today.
1
Researcher
Initiates a bounded scientific question
Initiates a bounded scientific question
2
Scientific Workflow Composer
Frames the request inside a governed workflow
Frames the request inside a governed workflow
3
SIW pre-call inspection
Examines the request before any external call is made
Examines the request before any external call is made
4
Genesis Gateway
Bounded, credentialed egress — no proprietary logic exposed in the browser
Bounded, credentialed egress — no proprietary logic exposed in the browser
5
External scientific provider
Candidate first adapter: AlphaGenome — proposed only, not integrated
Candidate first adapter: AlphaGenome — proposed only, not integrated
6
SIW post-call inspection
Examines the returned result before anything downstream can rely on it
Examines the returned result before anything downstream can rely on it
7
Governed Genomic Canon
Where evidence, standing, and lineage would be preserved — not the provider's own corpus
Where evidence, standing, and lineage would be preserved — not the provider's own corpus
8
Reader / Proof Surface
Where a reviewer would inspect the full chain, not just the final answer
Where a reviewer would inspect the full chain, not just the final answer
4. What Genesis would preserve
Governance requires keeping the right record, not the whole corpus.
- Genomic entity and reference context
- Provider, model, and version
- Request purpose and permitted-use scope
- Biological and ontology context
- Provider output and response digest
- Evidence supporting, refuting, or qualifying the claim
- Conflicts and unresolved limitations
- Inspection history
- Corroboration and standing transitions
- Workflow lineage and revalidation events
- Separately determined authorized use
5. What a demonstrator must prove
A reviewer must be able to answer every one of these — from the record, not from trust.
- What exactly was predicted?
- Which provider / model / version produced it?
- What biological context was requested?
- Which independent evidence supports it?
- Which evidence conflicts with or qualifies it?
- What did SIW inspect?
- Why was the current standing assigned?
- Which uses are permitted?
- Which uses are prohibited?
- Can the result be reproduced, challenged, or revalidated?
6. Federation boundary
Genesis would not copy the provider's science. It would preserve the record of governing it.
Genesis AiX would not copy or rehost the provider's large scientific corpus. The provider retains its own data. Genesis would preserve the bounded identities, metadata, provenance, hashes, evidence relationships, restrictions, and standing necessary for governance and reconstruction — a governance record, not a mirror.
7. Current status and explicit nonclaims
What this page does and does not claim.
This page does not claim:
- Any patient-specific data — none is used or proposed for this stage.
- Any clinical use — this is research-only.
- Any diagnostic or treatment capability.
- Any clinical decision support.
- Any autonomous biological experimentation.
- Any affiliation, partnership, endorsement, or certification from Google or DeepMind.
- That any AlphaGenome output has been admitted to the Genesis Canon — it has not. A provider return remains candidate material.