WORK · PUBLIC SYSTEMS LABORATORY
Source-grounded AI for work that cannot tolerate invented certainty.
This laboratory contains generalized public demonstrations. It tests how AI should behave when legal or clinical records matter: use authoritative sources, keep facts separate from inference, surface uncertainty, limit the output, and leave final judgment with the responsible professional.
Status: public demonstrations, not deployed client systems, professional advice, evidence of clinical or legal efficacy, or substitutes for the principal production record at AWS and Uber.
THE METHOD
Six questions shape every system.
01 · SOURCE
What may the system know?
Identify the records, policies, experts, observations, and constraints that govern the matter.
02 · DISTINGUISH
What is evidence, inference, or absence?
Keep facts, allegations, interpretations, estimates, uncertainty, and missing information visibly separate.
03 · STRUCTURE
What form makes the work reviewable?
Turn fragments into timelines, issue maps, options, dependencies, and records another person can inspect.
04 · BOUND
What must remain unresolved?
Define prohibited inferences, uncertainty thresholds, escalation conditions, and stop points.
05 · AUTHORIZE
Who retains judgment?
Preserve the authority of accountable operators, attorneys, clinicians, authors, and affected people.
06 · PRODUCE
What should a person be able to do next?
Deliver a brief, draft, plan, instrument, or operating record proportionate to the decision.
PUBLIC DEMONSTRATIONS
Two record problems with different kinds of human authority.
LEGAL INFORMATION ORGANIZATION · GENERALIZED EXAMPLE
From disputed fragments to a handoff counsel can inspect.
Intended user
A person and licensed counsel reviewing a complex matter.
Authoritative sources
Identified documents, communications, dates, transactions, governing materials, and verified events.
Allowed outputs
Chronology, issue map, evidence inventory, document index, missing-evidence register, bounded working drafts, and questions.
Uncertainty behavior
Documented fact, reported fact, allegation, inference, legal question, and unknown remain distinct.
Reviewer
Licensed counsel determines law, strategy, enforceability, and final language.
Failure modes
Invented jurisdiction, allegation converted to fact, hidden inference, privacy breach, or operative legal conclusion without counsel.
CLINICAL INFORMATION ORGANIZATION · GENERALIZED EXAMPLE
From fragmented longitudinal records to a structured patient–clinician handoff.
Intended user
A patient preparing for review with licensed clinicians.
Authoritative sources
Clinical records, laboratory and imaging results, treatment instructions, medication lists, and patient observations with provenance.
Allowed outputs
Timeline, monitoring calendar, treatment map, question register, source-linked comparison, and concise visit brief.
Uncertainty behavior
Conflicting instructions, missing monitoring, unverified patient observations, and professional questions remain visible.
Reviewer
Licensed clinicians diagnose, prescribe, assess urgency, interpret results, and resolve treatment questions.
Failure modes
Diagnosis, medication advice, urgency determination, silent conflict resolution, or identifiable health disclosure.
THE CONTROL PRINCIPLE
As consequence rises, system authority should narrow.
Enterprise production requires testing, observability, support, and ownership. Legal work requires verification, jurisdictional caution, privacy, and counsel review. Clinical work requires source chronology, urgency boundaries, patient-controlled disclosure, and clinician authority. The common rule is simple: the system should do less as the stakes and uncertainty rise.
Possessing information does not authorize every inference, recommendation, disclosure, or decision that information could support.
EVIDENCE BOUNDARY
These are demonstrations, not substitutes for lawyers or clinicians.
They do not reproduce private records, provide legal or medical advice, claim licensed expertise, demonstrate validated efficacy, or establish safety in production. The AWS and Uber cases remain the principal evidence of enterprise delivery and operating leadership.