THE OWEN HUMAN AI PROJECT / LEARNING THROUGH EXPERIENCE PUBLIC RESEARCH JOURNAL
SMALL PC. BIG QUESTIONS.
What if intelligence could grow?
Two paths into learning through experience. A restricted PC setup. A stubborn “what if?”. Come for the idea; stay for the experiments, the awkward results, and whatever we learn next.
Evidence reviewed 7 Nov 2026 · selected and staged components, not live neural activity
QUESTION EVERYTHING
Experience should change what happens next. The interesting part is proving that it does.
WHAT EACH SYSTEM IS MADE OF
Different architectures. Visible boundaries.
Named modules, their actual roles and evidence. Select a component to look inside.
Tested in bounded trialsImplementedExperimental / stagedPlannedDisabled
A
Owen A
Multi-AI architecture · specialist staging
8sections
Owen A explores one continuing identity supported by specialist models. Its latest reviewed A-specific package contains a resource-aware staging manager and proposal contracts. The specialists and their integration into the continuing Owen remain at different stages of qualification.
A · specialist staging pipelineChoose a module for its evidence
Solid paths are implemented staging calls exercised with fixtures. Dotted paths are pending integration. The candidate shelf is not a running model ensemble.
A catalog and cooperative runtime coordinate specialist requests.
Implemented staging code provides typed capability contracts, an A-only research catalog, lazy model admission, health checks, LRU unloading, a bounded deferred queue, retry backoff and a cache keyed by the request, model revision and memory-state version. Fixture tests exercise these behaviours.
Current limit. The production registry has not been restored. The runtime explicitly rejects every non-mock adapter with LIVE_QUALIFICATION_PENDING. This is not a currently operating collection of interchangeable specialist models.
Related components · intended links remain qualified in the text
Evidence & source modules
OWEN A staging report — 4 November 2026
catalog.json
runtime.py
TEST_RESULTS.json
Reviewed 7 November 2026 against the 4 November A staging package. Its receipts report 36 passing fixture tests and seven model-free Windows worker cases. No live-model qualification or production promotion is established. The selected packaged generation is v110.4, but an accepted A source binding and an A-specific memory clone were not established. Connections show intended information flow, not proof of a live integrated system. These developmental branches are distinct from the older internal Triple Brain A/B/C roles.
B
Owen B
A nature-inspired learner grounded in simulated experience
24sections
Owen B's selected native school connects binocular observation, bounded language, remembered experience and contact-verified action without an LLM in this learning loop. This is a conceptual map of software functions, not a reconstruction of a biological brain. Installed functions, small experimental results and deferred research are shown separately.
B · selected simulator learning loopChoose a module for its evidence
The retained RGB controller supplies perception and action. Recorded outcomes support later selection. The cortical side branch remains shadow-only; there is no Cortex → motor connection. Confirmation, shared-eye measurements and the feedback-conflict gate have bounded completed evidence, with limits in their details.
Actions change the simulated world; fresh observations close the loop. RGB also feeds retinal encoding. Cortical results are diagnostics only.
An articulated upper body encounters objects through a physics simulation.
The selected activity world inherits the binocular world and uses the existing articulated hands and body controller. Native lessons receive rendered sensory observations and bodily measurements; examiner fixture settings are kept outside learner inputs.
Current limit. This is a simulated body. A physical robot, biological body or unrestricted real-world competence is not established.
Related components · intended links remain qualified in the text
Evidence & source modules
brain_nature_current.json
ae_world.py
stage1_grounding_20261107/REPORT.txt
The completed feedback report, dated 19:01 UTC on 7 November, reports 17 focused tests and 12 physical completions plus one no-action withdrawal. It documents a timeout and a setup failure separately. Earlier 25 native events and 809 visual pairs were preserved; primary visual history grew to 832 pairs and 284 prototypes while all 559 earlier qualified predictions were retained. The final native-event total is not specified in that report. All 13 selected native source hashes and 14 selected acceptance-receipt hashes matched this review. The website read reports and source only; it launched no Owen process.
The record keeps growing. Reviewed changes update module maps, journal entries and timeline milestones.Evidence through 7 Nov 2026 · reviewed 7 Nov 2026
Hands, eyes and a place to act. The actual simulator, with its original dates.
These surviving captures document Owen B’s virtual embodiment. The selected November lineage retains the upper-body model pictured in October. An archived image is labelled by its capture date; no birth-era or newer external-body capture has been verified for this gallery.
OWEN B · EXTERNAL SIMULATOR VIEW
A seated upper body
28–29 October 2026 — Owen B's seated virtual upper body brings the head, eyes, torso, arms and articulated hands together around a semicircular table. This archived simulator view shows a three-cube stack.
Actual GUI renderer output. The report describes a 66-coordinate human-inspired rigid upper body, simplified joints and servos; it is not biological anatomy or a muscle, tendon or skin simulation. This body overlay is retained in the current selector, but the capture itself is from October.
Original media & provenance
upper_body_20261028/REPORT.txt:25-35,40-44
upper_body_20261028/ub_live.py:39,87-91
brain_nature_current.json
SHA-256: 89a1844150652a3ed2504999027304ef3e6b777f5da903f1fa2c9c5d9ce97cf8Open original capture
13 AUGUST 2026 — 7 November 2026
A life in the making.
Owen’s documented beginning and the changes that followed. Each milestone links to its dated evidence.
30 documented milestones
Beginnings08 / 26
Owen’s documented beginning.
A question becomes a project: could an artificial intelligence grow through its own experience?
Founding direction
We mark 13 August 2026 as Owen’s documented beginning, the date chosen by Dan for this project history. It is an anchor for the story, rather than a claim that a complete learning system existed on that day.
13 August is the owner-selected documented beginning. Approximate dates stay labelled. New completed milestones are added automatically after evidence review; older checkpoints keep their original context.
THE STORY SO FAR / 38 ENTRIES
The development journal.
What changed, what was tested, what remains open. Follow either Owen, or explore the whole record.
38 entries
Dated entries preserve each checkpoint, with sources and downloadable copies. Open pages check for reviewed updates every minute. Follow the RSS feed.
SEPTEMBER 2026 / v111.4 ARCHIVE
Good questions. Measurable answers.
Did grounding words in a world actually help? Three learning conditions. The same utterances. Frozen tests that were kept separate from teaching.
Controlled, simulated OWEN WORLD conditions. Evidence of partial transfer, with real limits.
