Gilgamesh to Homero – 180 minds of 1600 reconstructed as runnable neural nets

Reconstructing 1,600+ minds from the past — up to 1905 — and asking a single question of each: if they were alive today, how would they build an Artificial General Intelligence?

🌐 Encyclopedia  ·  📖 Book Series (Amazon)  ·  🧪 Interactive Demos  ·  📊 E-AGI Barometer  ·  ✍️ Author

Lost Minds AI is a large-scale project that reconstructs the cognition of historical thinkers — poets, kings, physicians, lawgivers, mystics, mathematicians, generals — and projects each of them forward into our era to imagine the AGI they would have designed, given how they actually thought about mind, order, knowledge and the self.

Every mind is reconstructed across two planes:

This repository open-sources the mechanistic plane — the architectures — alongside a visual mind-map explainer for each figure. It grows tome by tome until it holds the full Encyclopedia.

No fabricated metrics. Each architecture ships with a mandatory finite-difference gradient check, a real training loop and a self-test suite. Every number a program reports is produced live, on the machine that runs it — not hard-coded.

The corpus runs in chronological order (Date is birth year). Each tome collects 20 minds; every tome page below is fully illustrated with the mind-map explainers and links each mind to its runnable architecture.

The arc so far runs from Gilgamesh — the first hero to confront mortality as the core problem of a thinking being — through Homer, Confucius, Socrates, Plato, Aristotle and Archimedes, to Liu An and the resonance-cosmology of the Huainanzi; on into Rome, the Han and the Antonine world; through the collapse of the West, from the Cao Wei workshops to the wall at Shaolin; and into the sixth century, from Aryabhata's observatory at Kusumapura and Justinian's law commission to the Old North of Taliesin and Aneirin, and the fair of ʿUkāẓ where al-Khansāʾ wept for her brothers.

No two architectures are alike. A sample of the 180:

• Gilgamesh 's network grows wiser through simulated grief and hands an "epic" to a successor.
• Hammurabi 's decides each case by analogy to a fixed canon of public precedents.
• Homer 's cannot physically emit a line that breaks the metre.
• Ashoka 's wires remorse in as a backpropagated error signal that gates its own dominant objective.
• Sima Qian 's splits one truth across five incompatible views and severs the gradient from its own verdict, so judgment can never rewrite the record.
• Cleopatra 's renders a single conserved self into five audiences and is provably faithful only under an audit run across all of them.
• Cato 's seals a permissibility head where the reward channel can read it but never reach it.
• Heron 's winds a program onto a pegged drum and runs it down a finite falling weight.
• Wang Chong 's divides the net tilt of the evidence by its total contested mass, so ten-for against nine-against correctly reads as nothing.
• Boudica 's ignites a coupled field that stays locked far below the coupling that lit it, and has to learn its brake long after.
• Galen 's splits a conserved pneuma between memory and reason, so that feeding one starves the other.
• Nāgārjuna 's hands every node the same contentless seed and lets identity precipitate out of relation alone.
• Cao Zhi 's routes an inner state with no direct path to speech through a codebook of figures, under a gate tightened until meaning concentrates or breaks.
• Ma Jun 's charges rent for every strung treadle and prunes a fifty-treadle loom to twelve as a training dynamic rather than a setting.
• Cyprian 's gives every bishop the whole verdict and no bishop the deciding vote.
• Theodosius 's throttles a fast will by an estimate of irreversibility and a conscience that shares none of its parameters.
• Patrick 's authenticates a directive before it is allowed to propagate.
• Bodhidharma 's removes the residual instead of adding a feature, and snaps onto its mirror in one step.
• Aryabhata 's stores frequencies rather than signals and predicts by rolling a recurrence forward.
• Dignāga 's keeps no positive class representation anywhere and lands every reason it constructs in the one cell of the wheel that licenses speech.
• Benedict 's keeps two voices that never share an hour, so the newer cannot crush the older.
• Justinian 's logs every interpolation it makes and freezes itself on promulgation.
• Theodora 's shelters the refuted branch by exemption and then lets it ordain a successor.
• Myrddin 's finds its own seam and halts when its principal is gone.
• Aneirin 's forecasts its own erasure before it happens and keeps every man in three hands.
• Shōtoku 's has no argmax and walls the bribe off from the verdict, so a bribe can buy a refusal to judge but never a judgement.
• al-Khansāʾ 's freezes the dead at zero plasticity and keeps the loss in a ledger that only an act can close.

→ Start with Tome 1, or jump to any tome above.

LostMindsAI/
├── readme.md                     ← you are here
├── tome1.md … tome9.md           ← illustrated indexes, 20 minds each
├── minds/                        ← the runnable architectures (one Neuron per mind)
│   ├── chapter_0001_gilgamesh_-2700.py
│   ├── chapter_0002_zoser_-2670.py
│   └── … (through chapter_0180_al_khansa_575.py)
└── maps/                         ← the visual mind-map explainers (one image per mind)
    ├── chapter_0001_gilgamesh_-2700.jpg
    └── … (through chapter_0180_al_khansa_575.jpg)

Every file is prefixed with its mind number (chapter_00NN_…) so nothing collides as the corpus grows toward 1,600+ entries. Each architecture and its explainer image share the same stem.

Each architecture is self-contained and depends only on NumPy.

git clone https://github.com/DavidVivancos/LostMindsAI.git
cd LostMindsAI

pip install numpy

python3 minds/chapter_0089_Archimedes_-287.py --test

python3 minds/chapter_0089_Archimedes_-287.py

Most files accept --test (self-tests then exit) and --quiet (demo without ASCII plots); the Tome 9 architectures run their full self-test suite and demo on a plain invocation, and a few offer a shorter run (--quick or --fast — see each file's header). Because each architecture is built to embody a specific mind, no two behave alike.

Every reconstructed mind is measured against the same yardstick — the Artificiology E-AGI Barometer — so that a Bronze-Age lawgiver and a Hellenistic geometer can be compared on the capabilities their imagined AGI would need. The eight top-level dimensions:

Each chapter closes by imagining how its figure would have reasoned about an embodied AGI (an E-AGI / humanoid) against these metrics.

The project is research-first. Before any architecture is written, the figure's surviving works and current scholarship are gathered and every source verified — a real corpus is small, but never fabricated. Where evidence is thin, the entry says so plainly rather than inventing an inner life.

Each figure is tagged with a candid provenance:

• 🟢 belief — the figure's own surviving works or recorded doctrine ground the entry.(143 of the first 180.)
• 🟡 mediated — no words of their own survive; they are known only through others' (often hostile or legendary) accounts.(25 of the first 180.)
• 🔵 extrapolated — no philosophy of mind survives at all; the entry is inferred from documented deeds, typical of kings and builders.(12 of the first 180.)

And each architecture is required to embody the mind and to run: from-scratch pure-NumPy, a passing gradient check, a real training loop and self-tests — deliberately not a default Transformer, but a mechanism chosen to encode that thinker's own cognitive fingerprint.

• ✅ Tomes 1–9 — Minds 1–180 · architectures + visual explainers*(this release)*
• 🔜 Further tomes released here as they open-source, extending toward the full 1,141-mind corpus (antiquity → 1905).
• 🎛️ The interactive MindMap.html planes and the long-form chapter texts live in the wider ecosystem — read them at**lostmindsai.com** and across the**Amazon book series** .

This README is intentionally global: it describes the whole Encyclopedia and stays valid as new tomes land in this repository.

@misc{vivancos_lostmindsai,
  author       = {Vivancos, David},
  title        = {Encyclopedia of Lost Minds: Echoes on AI —
                  How History's Greatest Thinkers Would Have Thought About AGI},
  howpublished = {\url{https://github.com/DavidVivancos/LostMindsAI}},
  note         = {Encyclopedia: https://lostmindsai.com}
}

| 🌐 Encyclopedia | lostmindsai.com | | 📖 Book series (Amazon) | Tome 1 ·Tome 2 ·Tome 3 ·Tome 4 ·Tome 5 ·Tome 6 ·Tome 7 ·Tome 8 ·Tome 9 | | 🧪 Résumé & interactive demos | artificiology.com | | 📊 E-AGI Barometer | artificiology.com/barometer.html | | ✍️ Author — David Vivancos | vivancos.com |

The individual instance dies, but the pattern persists across successors. — the immortality the Epic of Gilgamesh actually endorses, and the wager of this Encyclopedia.