Kayser Medical · Independent Research and Advisory
Building a Governance Layer Between Humans and AI.
Kayser Medical researches how to separate an AI proposal from permission to act: who approves it, what evidence they need, and how to pause or reverse the action. Independent advisory work draws on clinical genetics and laboratory experience.
Sense the state. Govern the transition. Preserve the brake.
One workflow. A human-reviewed report. A 30-minute findings call. See a sample review Support the research
Michael A. Kayser, DO, FACMG · Clinical Geneticist · Physician-Researcher · Systems Investigator
One evidence-labeled research record across Kayser Medical and Resona.
What Is Kayser Medical?
A Clinical Lens on Machine Governance.
Clinical work requires decisions under uncertainty. Findings must remain reviewable, and new evidence must be able to change an earlier conclusion.
Kayser Medical applies that discipline to human-AI systems: observe carefully, separate signal from interpretation, preserve provenance, distribute authority, and make correction possible in both directions.
Telemetry
What is happening in the human-machine system?
Governance
Who has authority to interpret the signal, change system state, or stop the process?
Interlock
What must the system do when evidence, human capacity, or operating conditions cross a safety boundary?
Govern movement, not truth.
Observe
Measure state without pretending measurement is judgment.
Speak
Convert observations into explicit, reviewable claims.
Adjudicate
Require independent authority for consequential transitions.
Actuate
Execute only an authorized action, with a way to pause or reverse it.
Evidence supports or revises a claim. Consent, policy, and designated authority determine whether an action is permitted. A persuasive answer does not authorize its own execution.
Resona
Human-State-Aware AI Governance.
Resona is a developing research architecture for AI systems that respond not only to the user's words, but also to the conditions under which the interaction is occurring. It investigates whether physiological, behavioral, temporal, and conversational signals can help an AI system adjust pace, complexity, escalation, and stopping behavior.
System Flow
- Human state and context
- Telemetry
- Explicit state estimate
- Independent policy gate
- Adaptive AI response
- Reviewable transition record
Telemetry
Potential inputs may include heart-rate variability, breathing, interaction tempo, behavioral signals, and contextual evidence. These inputs are signals, not diagnoses.
Governance
Policy determines what transitions are allowed and who has authority to authorize them. Observation, interpretation, adjudication, and execution should not silently collapse into one actor.
Interlock
The AI can slow, simplify, request confirmation, preserve an unresolved question, decline escalation, or stop. The brake must remain available even when continuation is narratively attractive.
Exploratory architecture and active single-subject development. The chat demo does not enforce the proposed independent action gate. Not a validated diagnostic system, medical device, or substitute for clinical judgment. Read implementation and validation status.
AI Governance
Correction Must Work in Both Directions.
Residual Preservation
Preserve what remains unresolved; retire what has been resolved.
Clean Exit Rule
A system must correct false negatives and stale positives. Suspicion cannot become permanent merely because it once had support.
Provenance Rule
A resolved claim followed by a new concern becomes a new claim. Do not silently mutate the old proposition.
Independent Warrant
Failure of one hypothesis may identify a candidate boundary, but the failure does not prove the boundary. The boundary requires independent evidence.
Founder-Stop Constraint
The founder, operator, or most invested participant must not possess unilateral authority to override the evidence or disable the brake.
Reversible Depth
As system consequences deepen, transition requirements should become more explicit, independent, and reversible.
The purpose is not to make AI omniscient. It is to make consequential movement inspectable, reversible, and stoppable.
QOTE
An Exploratory Pattern Language for Complex Systems.
QOTE examines whether recurring structures of prediction, error detection, feedback, oscillation, constraint, and correction can provide a useful comparative language across biological, computational, organizational, and economic systems.
QOTE does not claim that these systems are physically identical. Shared vocabulary is not proof of a shared mechanism. Analogies must produce measurable variables, discriminating experiments, or useful governance constraints to earn their place.
Cerebellar learning
Prediction-error updating
Established scientific framework
Machine learning
Loss-based parameter updating
Established computational method
Human-AI interaction
State-aware pacing and gating
Resona research hypothesis
Organizations
Distributed authority and stopping rules
Governance application
Cross-domain unification
Common causal mechanism
Open question
Pattern is a hypothesis generator, not proof.
Read the historical QOTE research archiveThe Clinical Foundation
Discipline Under Incomplete Evidence.
Michael A. Kayser, DO, FACMG is a clinical geneticist with experience spanning rare disease, biochemical genetics, neurodevelopmental disorders, hereditary cancer, laboratory medicine, and the care of medically complex patients.
Clinical genetics requires decisions under incomplete evidence. It also requires disciplined revision: a variant can be reclassified, a differential diagnosis can narrow, and an earlier concern can close. That clinical habit—preserve uncertainty without preserving error—is foundational to the Kayser Medical approach to AI governance.
Clinical availability: Kayser Medical is not currently accepting patient-care appointments through this website.
Training & Credentials
- Doctor of Osteopathic Medicine
- Residency training in Pediatrics
- Residency training in Clinical Genetics
- Fellowship training in Clinical Biochemical Genetics at the National Human Genome Research Institute
- Fellow of the American College of Medical Genetics and Genomics
Professional background is presented separately from research and software status. Current licensure and specialty certification have not been independently reverified for this website update (October 11, 2026).
Professional and clinical authority record →Research Evidence
From Observation to Testable Work.
Early single-subject observations helped generate the Resona research questions. They do not establish clinical efficacy. Future work must define the baseline, measurement interval, outcome criteria, adverse events, confounders, and preregistered decision rules before broader claims are made.
View source status and missing methods record →
Preliminary single-subject observation
A reported reduction in defined episodes during an HRV-biofeedback protocol.
A quantified effect size is intentionally not displayed as a headline figure. Any specific percentage will be published only alongside a methods record documenting raw counts, baseline and intervention periods, episode definition, collection method, missing observations, concurrent interventions, analysis method, and limitations.
The creative and development studio
Ideas you can open.
Explore the dated collection of v0-built apps and prototypes alongside Michael Kayser’s Suno music and shared NotebookLM conversations.
Publications & Writing
Research & Writing.
Published writing appears first. Working drafts are labeled In Development; essays and exploratory research are distinct from peer-reviewed clinical publications.
- Substack EssayPublished
The Sabbatical Is Over. The Work Is…
A personal update on the return to work and the direction of independent research.
- Peer-Reviewed PublicationPublished
Increasing Primary Care Provider-Initiated Geneticist Referrals in 22q11.2 Deletion Syndrome
Peer-reviewed clinical-genetics research examining whether targeted education improves recognition and referral consideration for 22q11.2 deletion syndrome.
- Working PaperIn Development
Resona: Building AI That Reads the Body
The case for a physiological regulatory layer between humans and AI systems. Frames the research questions Resona is being built to test.
- Field NotesIn Development
The Coherence State Index
An exploratory approach to describing and classifying human regulatory states in real time. Signals are treated as estimates, not diagnoses.
- EssayIn Development
AI Safety Is a Physiology Problem
Argues that the human at the interface is a load-bearing part of any AI safety system. Considers what changes when the receiver can be dysregulated.
- Working PaperIn Development
The Independent Policy Gate: Governing Consequential Transitions
Describes why observation, interpretation, adjudication, and execution should not collapse into one actor. Explores authority and reversibility for state changes.
- Working PaperIn Development
QOTE: An Exploratory Pattern Language for Complex Systems
Examines whether shared structures of prediction, feedback, and correction offer a useful comparative language across domains. Treats analogy as a hypothesis generator, not proof.
- Single-Subject ProtocolIn Development
HRV Biofeedback for Dissociative Episode Reduction
A single-subject (n=1) protocol and observation log. Hypothesis-generating only; not evidence of general clinical effectiveness.
About Michael Kayser
The Physician Behind the System.
Michael A. Kayser, DO, FACMG is a clinical geneticist, physician-researcher, and systems investigator. His work has moved through academic biology, pediatrics, clinical and biochemical genetics, laboratory medicine, healthcare delivery, regulation, and AI-assisted research.
Across those settings, the recurring problem has been the same: complex systems accumulate signals faster than institutions can interpret them. Kayser Medical studies how those systems can preserve uncertainty, distribute authority, correct themselves, and stop safely.
Collaboration
Independent Work Requires Independent Review.
Kayser Medical welcomes serious collaboration with clinical researchers, physiologists, AI-safety investigators, governance scholars, engineers, evaluators, and institutions interested in testable human-AI safety systems.
Independent advisory
Resona AI Workflow Review
A focused review of one AI workflow: who can act, what evidence supports the action, and where a human should be able to pause or reverse it.
- Structured intake covering one workflow and up to three supporting documents.
- A human-reviewed report with an authority map, findings, uncertainties, and three prioritized improvements.
- A 30-minute findings call, with delivery within five business days after complete intake.
$495 USD, one-time. Checkout leads to the intake form. Implementation is quoted separately. This review is not certification or a clinical consultation.
Illustrative sample · hypothetical workflow
When an AI assistant proposes a refund
A support assistant drafts a refund and can call a payment tool. The review asks whether a plausible recommendation can become a financial action without a clear approval boundary. This is a sample of the report format, not a client case or a validated outcome.
Authority map
- Assistant · proposes
- Cites the order and refund policy. Cannot approve its own recommendation.
- Account owner · authorizes
- Approves the exact order, amount, and destination within the applicable policy.
- Payment service · executes
- Checks approval and permissions before acting; records the result for reconciliation.
Three prioritized improvements
- Require approval before execution. Separate drafting from the payment permission; bind approval to the exact transaction.
- Check again at the tool boundary. Confirm the current policy, amount limit, destination, and approval expiry; prevent duplicate refunds with a request identifier.
- Make correction practical. Log proposal, approval, and receipt; provide a pause path and a named owner for reconciliation when reversal is unavailable.
Unresolved in this example: actual permissions, refund reversibility, policy exceptions, and approval records. A real review identifies these from the supplied workflow and documents.
Research
Protocol development, measurement, falsification, replication, and publication.
Engineering
Telemetry integration, policy-gate architecture, provenance, audit trails, and human-state-aware interfaces.
Governance & Evaluation
Independent review, transition authority, stopping conditions, adversarial testing, and correction mechanisms.