Source: David Vivancos, LostMindsAI on GitHub. Distilled and processed for Signal; not original content.
TL;DR
An encyclopedia project that reconstructs how historical thinkers actually reasoned about mind, then asks each one how they would build an artificial general intelligence today. Every mind gets two artifacts: an interactive 3D mind map of their worldview and a small, genuinely runnable neural architecture in pure NumPy whose design encodes their cognitive signature. The open-source repo ships 240 minds so far (Gilgamesh to al-Mutanabbī, c. 2700 BCE to 915 CE), out of a planned 1,600+ ending in 1905.
Actionable items
- Use it as a source of unusual design constraints: each architecture turns one historical idea into a mechanism, for example a model that indexes every belief by where it was acquired and keeps a ledger of its own amendments (al-Masʿūdī).
- Borrow the provenance discipline: each mind is labelled by how directly the design follows from the record (187 grounded in stated belief, 33 mediated, 20 extrapolated).
- Copy the “no fabricated metrics” rule for any research code: mandatory finite-difference gradient checks, a real training loop and a self-test suite, with every reported number produced live rather than hard-coded.
- Read the twelve tome questions as a checklist for trustworthy AI design: anchoring truth against rewriting, staying correctable, custody of what a mind holds, warrant for claims.
- Note the fixed chapter skeleton (life, philosophy of mind, how they would confront AGI, how they would build it, the architecture, what they got right and wrong, sources): a good template for comparing many designs side by side.
Key quotes
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?
Every number a program reports is produced live, on the machine that runs it, not hard-coded.