Human-governed research field / embodied systems

Decision integrity
for embodied and
consequential AI.

Help people understand what an automated system knows, what it does not know and why it is asking for a decision—without surrendering human authority.

A sunlit Symbiain decision studio connecting permissioned evidence, alternatives and clear human decision points
FIELD PLATE / THE HUMAN DECISION STUDIO
RESEARCH ARCHITECTURE

Symbiain is a bounded review and assurance proposition—not a motion controller, autonomous authority or system for ambient surveillance. It works with approved evidence around a defined question. Safety-critical systems remain independent and authoritative; accountable people retain every consequential decision.

The public operating boundary

Intelligence should enlarge
human agency.

Public trust begins with a clear answer to what the system will never be allowed to do. These boundaries govern any proposed research, pilot or application.

  1. 01 / BOUNDED PURPOSE

    One declared question.

    The purpose, source base, exclusions, decision owner and stop conditions are named before work begins.

  2. 02 / CONSENT & ACCESS

    Permissioned evidence only.

    No covert collection, ambient monitoring or unrestricted trawling of private records is part of the Symbiain offer.

  3. 03 / NO PEOPLE SCORING

    Behaviour is not identity.

    Observable actions remain separate from inferred motive, emotion, character or risk. Symbiain does not claim to read minds.

  4. 04 / HUMAN AUTHORITY

    People decide and can stop.

    Consequential judgement, intervention and escalation remain with a named, accountable human who can challenge, override or end the process.

The distinctive contribution

Not another model.
A visible decision field.

AI and robots can fail not only because their information is incomplete, but because they cannot show what conflicted, what they inferred, who should decide and what would make them revise their account.

Symbiain holds those distinctions together without pretending they are the same thing. The aim is a calmer, more legible route from permissioned evidence to accountable human action.

The operating sequence

From signal
to retained learning.

  1. 01Sense
  2. 02Register
  3. 03Represent
  4. 04Challenge
  5. 05Decide
  6. 06Act · Hold · Escalate
  7. 07Reconcile
  8. 08Learn

The order creates places to pause. A signal does not silently become a fact; a representation does not silently become a decision; and an action does not disappear before its consequence is reviewed.

A cinematic warehouse exception chamber in which conflicting map, lidar, camera and record evidence reaches a safe-hold gate and human supervisor
PLATE / THE SAFE HOLD IS A LEGIBLE STATE, NOT A FAILURE

The best first problem

Manage the conflict
before controlling the motion.

A warehouse map says the route is clear. Lidar shows an obstruction. The camera is uncertain. The job record assumes access. A pallet may have moved without the record changing.

Symbiain turns that disagreement into a managed decision object: what is observed, what is recorded, what is inferred, what else may explain it, who may clear it and what evidence permits revision.

RESCANREROUTESAFE HOLDASK HUMAN

The application field

Twelve places
decision integrity matters.

Each problem is framed as a bounded contribution and a safe first application—not as evidence that Symbiain controls, profiles or certifies a deployed system.

  1. 01

    Conflicting evidence

    Keep sensor, map, record and human reports distinct when they disagree.

    FIRST FIELD / Warehouse route exception
  2. 02

    Autonomy under ambiguity

    Make act, constrain, rescan, hold and escalate available as explicit states.

    FIRST FIELD / Obstructed mobile-robot route
  3. 03

    Usable explanation

    Give an operator an evidence-and-reasoning card, not a confidence score alone.

    FIRST FIELD / Pause or reroute review
  4. 04

    Human handover

    State what a person needs to know, decide and record at the escalation point.

    FIRST FIELD / Logistics or quality exception
  5. 05

    Reasoning audit

    Retain the path from source and observation to action, outcome and correction.

    FIRST FIELD / Post-incident review
  6. 06

    Automation bias

    Surface alternatives and disconfirmation before a plausible recommendation is accepted.

    FIRST FIELD / Anomaly or risk triage
  7. 07

    Operational drift

    Distinguish a changing site, model or practice from a presumed stable baseline.

    FIRST FIELD / Repeated route exceptions
  8. 08

    Mixed operational data

    Relate records, sensors, people, rules, assets and thresholds without flattening them.

    FIRST FIELD / Maintenance coordination
  9. 09

    Language-model agents

    Bound sources, authority, uncertainty and escalation around organisational records.

    FIRST FIELD / Read-only exception assistant
  10. 10

    Misleading learning loops

    Measure consequence and correction, not task completion alone.

    FIRST FIELD / Fleet near-miss review
  11. 11

    Interaction ambiguity

    Keep observable interaction separate from inferred intent; do not profile or pretend to read minds.

    FIRST FIELD / Consent-based service and collaborative robotics
  12. 12

    Inflated intelligence claims

    Label evidence quality, unknowns and authority limits explicitly.

    FIRST FIELD / AI and robotics demonstrators

Operator-facing explanation

Hand over
the decision, not the noise.

“Anomaly detected” is not a useful transfer of authority. A human needs the relevant evidence, bounded interpretation, credible alternatives, immediate risk, safe options, missing information and the precise decision that requires human judgement.

A Symbiain operator handover plate showing evidence, uncertainty, alternatives and a named human decision point
PLATE / EVIDENCE-AND-REASONING CARD

Research and validation pathway

Invite challenge.
Earn the stronger claim.

Symbiain is presented as an evolving research architecture. Credibility should come from transparent evaluation, appropriate peer review and retained limitations—not from the visual polish of a prototype.

  1. 01

    Specify

    Define the question, baseline, permitted evidence, expected contribution and failure criteria in advance.

  2. 02

    Compare

    Test a bounded output against ordinary practice, credible alternatives and expert judgement.

  3. 03

    Challenge

    Invite independent researchers and practitioners to identify errors, blind spots and disconfirming cases.

  4. 04

    Publish

    Pursue appropriately scoped peer review or conference submission, with methods, limits and corrections visible.

Discuss an independent research evaluation →

A safe development sequence

Demonstrate value
before granting authority.

  1. 01

    Charter

    Define the decision boundary, prohibited actions and named human authority.

  2. 02

    Shadow

    Use a permissioned, bounded record to compare with the ordinary workflow without granting the new layer control.

  3. 03

    Compare

    Measure it against normal manual handling and retain disagreements.

  4. 04

    Advise

    Produce concise evidence cards for human review.

  5. 05

    Intervene

    Test one narrow, human-approved action with a safe exit.

  6. 06

    Learn

    Review outcomes, false holds, missed anomalies and corrections.

  7. 07

    Scale

    Expand only where the benefit and operating boundary are demonstrated.

A circular Symbiain learning archive where robot outcomes, near misses and human overrides return to a correction ledger without erasing failures
FIELD PLATE / OUTCOME, CONSEQUENCE, CORRECTION

Retained learning

Do not reward completion
and forget consequence.

A useful learning loop retains the original evidence, the decision boundary, human overrides, near misses, false holds, missed anomalies and the effect of corrective action. Failure remains visible as evidence for revision—not noise to be optimised away.

From one architecture to several fields

Different consequences.
The same discipline.

01 / ROBOTICS

Accountable action

Evidence, uncertainty, safe holds and human authority.

Open case →02 / INTERACTION

Human meaning into bounded action

Gesture, language and context without pretending to read minds.

Open case →03 / SERVICE

Dignified handover

Clarify the situation and transfer sensitive decisions to people.

Open case →04 / AI ASSURANCE

Fluency is not provenance

Decompose generated answers into inspectable claims.

Open case →05 / CLINICAL RESEARCH

Keep judgement professional

Prepare evidence while preserving pharmacist authority.

Open case →06 / CARE SUPPORT

Continuity without prediction

Retain what a person said and make uncertainty visible.

Open case →

THE NARROWER CLAIM IS THE STRONGER CLAIM

Symbiain is a human-governed research and decision-integrity architecture for making evidence, uncertainty, alternatives, authority, action and correction more inspectable around AI and robotic systems.
Discuss a bounded Decision Integrity Review →