The OwenDated chapters from Owen A and B, including what was tested, what failed and what remains uncertain. Event dates are shown separately from preparation dates.
Subscribe to the project RSS feed · Start with the project introduction
Owen B combines learned words with richer object descriptions and recalls earlier learning after several hours. A new-object placement passes its existing threshold at 32.93 mm error, and graphics diagnostics identify the stalled call without curing it.
Bounded composition trials completed · modest placement precision · graphics stall unresolvedRead the chapter →Owen B answers four grounded property questions, recalls new words after roughly an hour, and transfers them to new simulator combinations. Wrong-word recovery works in an isolated-copy test. A graphics deadline fix is complete, while the sandbox startup stall remains.
Bounded questions and recall completed · word recovery isolated · sandbox graphics unresolvedRead the chapter →Observed corrections transferred across a few layouts while earlier left and right placements were retained.
Bounded simulator trialsRead the chapter →Selected code adds fresh-pair confirmation, recall measurements after other teaching, and shared-eye focus diagnostics. The full trial report is still pending.
Implemented code · full trial pendingRead the chapter →The completed report tests fresh-pair confirmation, recall after other learning and coordinated eyes. It also records the timeout and the limits of the successful retry.
Completed bounded trials · one delay variant unqualifiedRead the chapter →A later selected revision adds a conflict gate, explicit teacher withdrawal and runtime instrumentation. These are source-verified changes with completed trial evidence still pending.
Implemented code · completed trial report pendingRead the chapter →Owen B completed a bounded contradiction-and-recovery test, varied-language transfer and a full 60-second isolated recall. Small accuracy changes and graphics timeout evidence remain visible.
Bounded trials completed · graphics variability unresolvedRead the chapter →A controlled artificial-word correction, measured 60-second recall and a visual retention check extend the evidence for Owen B.
Completed bounded trials · Stage 1Read the chapter →Owen B learned two narrow native word bindings from observed examples and reused them in placement and visual pursuit. The completed report also records retained spatial recall and a graphics stall that is bounded but not cured.
Bounded vocabulary and pursuit trials completed · sandbox graphics stall unresolvedRead the chapter →The Owen Human AI Project gets a matching address, a new emblem, and a history that readers can follow.
Website updateRead the chapter →Two conceptual maps bring the current sections, their connections and their evidence into view.
Public architecture reviewRead the chapter →Retina-inspired processing and sparse visual associations run alongside the retained recognition and gaze route.
Retinal processing enabled · cortical learner in shadowRead the chapter →Context-sensitive recall and a quiet-phase checkpoint extend the bounded native experiments.
Installed code · staged trialsRead the chapter →A bounded native-school route connects an arbitrary cue with a learned movement in staging.
Installed route · staging evidenceRead the chapter →Contact, frozen and force-ablated conditions all succeed; a tactile-specific advantage remains unproven.
Controlled experiment · no advantageRead the chapter →Temporary visual working memory responds cautiously when a recent object is missing.
Installed code · sensing probesRead the chapter →The staging manager can coordinate bounded specialist requests. The live-model qualification step is still ahead.
Staging tested · live integration pendingRead the chapter →An isolated motor model improves directional success on unseen capsules; replay exposes a trade-off.
Isolated experimentRead the chapter →The original simulator captures show Owen B’s virtual embodiment taking shape—simple geometry, visible joints and all.
Original simulator imagery · dated archiveRead the chapter →An isolated B29 clone completes bounded native learning and recall, while broader commands remain unsupported.
Isolated clone verifiedRead the chapter →The binocular-vision plan connects depth, uncertainty and eye movements to embodied consequences.
Research designRead the chapter →The research brief expands from individual connections to assemblies and whole-system regulation.
Research directiveRead the chapter →A governing charter separates Owen A’s specialist models from Owen B’s nature-inspired learning loop.
Governing planRead the chapter →A synthetic biology-inspired experiment helps in one setting and hurts in others.
Isolated experiment · candidate rejectedRead the chapter →A capability map records narrow visual and action results; a recovery blocker prevents new learning.
Bounded capabilities · session blockedRead the chapter →The readiness review keeps the negative results visible even after encouraging language experiments.
Not live acceptedRead the chapter →An independent 40-case test tells a more complicated story. The failures matter just as much as the gains.
Source validationRead the chapter →What happens when we remove the world, or deliberately pair the words with the wrong consequences?
Source validationRead the chapter →A controlled experiment in learning meaning from experience — and what changed when the consequences matched the words.
Source validationRead the chapter →How competing interpretations accumulate support, counterexamples, and evidence of transfer.
Source validationRead the chapter →From an unknown relation to a changed prediction. A close look at the simulated episodes, including the unresolved ones.
Source validationRead the chapter →OWEN WORLD gives prediction-first learning a small, controlled setting.
Source developmentRead the chapter →A Windows installation problem leads to a two-stage promotion process.
Installed · 57 gatesRead the chapter →Triple Brain experiments bring analytical, associative and local-model contributions into bounded roles.
Partially working · rolled backRead the chapter →The readable developmental journal becomes a projection of the canonical event ledger.
Verified candidateRead the chapter →Generation 2 connects specialist regions through shared identity and Memory Core contracts.
Installed · bounded checksRead the chapter →Conversation becomes a way to offer lessons, corrections and questions—with evidence kept attached.
Installed · acceptance pendingRead the chapter →An honest audit changes the emphasis from accumulating features to building coherent concepts.
Installed auditRead the chapter →The early architecture connects perception, prediction, action, error, teaching and memory.
Architecture documentedRead the chapter →The repository gives Owen an inspectable engineering starting point: an experience lifecycle and a tested checkpoint.
Repository checkpointRead the chapter →A question becomes a project: could an artificial intelligence grow through its own experience?
Founding directionRead the chapter →