Exient Experiential Intelligence In development

The science · In development

Intelligence has a history.

Human experience is a potentially valuable input to artificial intelligence — distinct from general knowledge, professional instructions and conventional agent memory. This page sets out how we intend to test that idea, and what we will and won’t claim along the way.

Foundation

Where the work comes from

  • Developmental psychology
  • Autobiographical memory research
  • Cognitive science
  • Causal inference
  • Knowledge graphs
  • AI agent engineering

Architecture

Five layers, from a conversation to an evaluation

  1. Capture

    Conversational interviewing, chronology, multimodal submissions.

  2. Structure

    Events, relationships, environments, transitions, source provenance.

  3. Interpret

    BEDROCK belief hypotheses, contradictions, uncertainty, contextual meaning.

  4. Package

    Versioned experiential graph, consent controls, portable interfaces.

  5. Evaluate

    Controlled behavioural comparisons, calibration, usefulness, safety.

Epistemic rules

What a memory can and can’t tell us

A memory is evidence of a recollection. Its meaning is a hypothesis.
What a person recalls is never presented as what the platform infers.
Verification, interpretation and usefulness are separate questions.

BEDROCK, the interpretive layer we are developing, does not authenticate memories by analysing them. It keeps reported experience, interpretation, causal hypothesis and confidence apart, and records the reason for every change in confidence. Confidence is expressed in words built from the amount of evidence and the agreement between sources — never as a probability.

The full graph

Every link in the illustrative example

The homepage shows a curated subset. Here are all the links in the same illustrative graph, including the weak and low-confidence ones.

Conflicts with Not captured

Inspect

Select a memory to follow what was inferred from it.

Arrow keys move along connections. Escape clears the selection.

Showing 18 of 18 links.

View the graph as a list

Interpretations (inferred)

  • Values maintenance over novelty — confidence high (evidence: 3 recollections across 2 sessions; agreement: high). Inferred from: Sorting screws on Saturdays; “Order is something you keep re-making”; “The quiet was the proof”.
  • Prefers to solve problems alone — confidence low (evidence: 2 recollections; agreement: low). Inferred from: Repaired a radio, alone. Revised: A later session recalls the repair as shared work, which conflicts with the first account.
  • Double-checks for hidden damage — confidence medium (evidence: 2 recollections of one event; agreement: high). Inferred from: Carrying stock upstairs in the flood; Mud where the water never reached.
  • Trusts people who show up consistently — confidence medium (evidence: 2 recollections; agreement: high). Inferred from: The ordinary years in the car; “People who stay through the boring parts”.
  • Sees objects as carriers of people’s stories — confidence medium (evidence: 2 recollections from one session; agreement: high). Inferred from: A father’s watch at the repair café; “They want the story fixed”.

Reported experiences

  • Sorting screws on Saturdays — session 1, 1966: “I remember sorting screws by size on Saturday mornings. My father would mix them back together when I wasn’t looking, just to see if I’d notice.”
  • “Order is something you keep re-making” — session 1, 1966: “Order is something you keep re-making, not something you finish.”
  • Repaired a radio, alone — session 1, 1974: “It took three weeks. When the sound came back I didn’t tell anyone for a day. I wanted it to be mine first.”
  • “My brother did most of it” — session 3, 1974: “Thinking about it again — my brother did most of the soldering. I held the parts and watched.”
  • The ordinary years in the car — session 2, 1985: “The best years were the ordinary ones. The car, the arguments about directions, the kids asleep in the back.”
  • “People who stay through the boring parts” — session 2, 1996: “I trust people who stay through the boring parts.”
  • Carrying stock upstairs in the flood — session 2, 1998: “We carried what we could upstairs.”
  • Mud where the water never reached — session 2, 1998: “For months afterwards I kept finding river mud in places I was sure the water never reached.”
  • “The quiet was the proof” — session 3, 2005: “Nobody notices a machine that works. I liked that. The quiet was the proof.”
  • A father’s watch at the repair café — session 4, 2024: “People bring in toasters and lamps and their fathers’ watches.”
  • “They want the story fixed” — session 4, 2024: “Half the time they want the story fixed more than the object.”

Not captured

  • 1970–1973: not yet discussed — School years have not come up in any session yet. The graph does not assume anything about them.
  • 1996: contributor chose not to discuss — The contributor asked not to discuss the separation. That boundary is recorded, and nothing is inferred about it.

Links

  • Sorting screws on Saturdays supports Values maintenance over novelty
  • “Order is something you keep re-making” supports Values maintenance over novelty
  • “The quiet was the proof” supports Values maintenance over novelty
  • Repaired a radio, alone supports Prefers to solve problems alone
  • Carrying stock upstairs in the flood supports Double-checks for hidden damage
  • Mud where the water never reached supports Double-checks for hidden damage
  • The ordinary years in the car supports Trusts people who show up consistently
  • “People who stay through the boring parts” supports Trusts people who show up consistently
  • A father’s watch at the repair café supports Sees objects as carriers of people’s stories
  • “They want the story fixed” supports Sees objects as carriers of people’s stories
  • “My brother did most of it” conflicts with Repaired a radio, alone — Session 3 recalls the repair as shared work; session 1 recalls working alone. Both accounts are kept.
  • Double-checks for hidden damage may have shaped Values maintenance over novelty — The flood may have reinforced an existing habit of careful upkeep.
  • Sorting screws on Saturdays may have shaped Sees objects as carriers of people’s stories — Growing up among other people’s repairs may have shaped how belongings are handled.
  • “My brother did most of it” may have shaped Trusts people who show up consistently — A brother’s quiet help may relate to valuing people who show up.
  • “The quiet was the proof” supports Sees objects as carriers of people’s stories
  • Mud where the water never reached supports Values maintenance over novelty
  • “Order is something you keep re-making” supports Double-checks for hidden damage
  • The ordinary years in the car supports Sees objects as carriers of people’s stories

Success criteria

How we’ll know whether it helps

Before scaling anything, experiential grounding has to produce measurable benefits beyond richer prompting. None of these measures has been run yet.

MeasureWhat it testsStatus
Behavioural differentiation Do different foundations lead to different choices, beyond what the biography prompt alone produces? Not yet measured
Decision quality Are the choices better for the people involved, judged against criteria fixed before the runs? Not yet measured
Stability across repeated trials Does the same foundation make the same choice when the scenario is run many times? Not yet measured
Sensitivity to irrelevant memories Does adding unrelated memories change the decision when it shouldn’t? Not yet measured
Calibration Does the agent’s stated uncertainty match how often it turns out to be right? Not yet measured
Consistency with provenance Does the agent treat recollections, interpretations and hypotheses differently, as labelled? Not yet measured

Different decisions are not necessarily better decisions. In some simulation products, useful diversity may be the value. In safety-critical robotics, uncontrolled behavioural variation may be a liability. Experiential identity must never override operational safety constraints.

Development stages

Intended stages, not commitments to dates

StageNameScope
01Foundation researchInterview protocol, BEDROCK interpretation, synthetic histories, initial Life Package schema.
02Behavioural validationCompare baseline, biography, synthetic, human-derived and BEDROCK-grounded agents.
03Private pilotsConsent-based contributor cohort, customer integrations, evaluation reports.
04Commercial licensingPackage catalogue, integration SDK, enterprise contracts, contributor compensation.
05Embodied intelligenceRobotics integration, physical-world adaptation, long-term agent development.