Historical Lineage

help edit space_dashboard

Dörner's Psi Model

The motivational lineage begins with Dietrich Dörner's "Psi" cognitive model. Psi centers on demands — physiological needs (food, water) and abstract needs (competence, certainty, affiliation) — each with a target range. When a demand deviates from its target, an urge develops to restore equilibrium. Behavior is modulated by four parameters: Activation, Resolution level, Certainty, and Selection threshold. Joscha Bach's MicroPsi embodied this in a serious software architecture.

OpenPsi (2011)

Cai, Goertzel, and Geisweiller realized the Psi model within the OpenCog framework, mapping Psi's demands to goal atoms in the AtomSpace, modulators to continuous parameters, and pleasure/pain signals to reinforcement learning over Hebbian links. This proved its worth in early agents (virtual pets, robotics), but was mostly hand-tuned and essentially single-objective at decision time.

OpenPsi Implementation Lessons (mailing-list-backed)

  • Context/action conflation: OpenPsi's rule format packed all context predicates and actions into a single unordered AndLink, making it impossible to programmatically distinguish action from precondition. This is exactly what MetaMo's "Modular Appraisal-Decision Interface" resolves. (OpenPsi-review, 2016)
  • Events→modulators→emotions causal flow: In the robotics implementation, "events don't influence emotions directly but rather through the lower level modulators." This causal architecture is precisely what MetaMo formalizes as the appraisal comonad updating modulator vectors. StateLink-based introspective logging prefigured MetaMo's emphasis on auditable decision-making. (Explaining-and-logging-reasons-for-emotional-changes, 2017)
  • Temporal reasoning gap: OpenPsi lacked native temporal reasoning — expressing "if you haven't done X for roughly 10 seconds" required "nasty tortured constructions." MetaMo inherits this aspiration but has not yet implemented temporal goal reasoning. (OpenPsi-review, 2016)
  • ECAN-mediated goal pursuit: Goertzel outlined an OpenPsi+ECAN integration pattern: repeatedly stimulate the query atom with STI, create a "learn new things" Goal, and use a PredictiveImplication rule connecting finding answers to fulfilling the goal. (Setting-Goals-and-Actions-as-OpenPsi-Rules, 2017)

MAGUS — Decision-Monad Component (External Lineage)

The decision side of MetaMo's pseudo-bimonad (\(D\) in \(F = D \circ \Psi\)) is not a Goertzel/Lian invention. The AGI 2025 second paper (Embodying Abstract Motivational Principles, §2 + conclusion, reference [8]) attributes MAGUS — the hierarchical decision/goal system — to Mikeda 2024, a separate Magi-side conceptual framework. MetaMo uses MAGUS as the decision monad in the bimonad coupling; it does not implement, replace, or extend MAGUS itself.

In MetaMo-Python,magus/decision.py the file contains class MagusDecision(DecisionMonad), but this is a compact reference class (scoring + goal-update logic) sufficient for demonstrating the bimonad coupling — not the full Magi Archive MAGUS framework. Wiki text describing MetaMo should say it uses or instantiates MAGUS as \(D\); it should not imply MetaMo authors created MAGUS, nor that the Python reference implements the complete Mikeda 2024 specification.

MetaMo (2025)

MetaMo evolves OpenPsi into a formally grounded, multi-objective framework, with MAGUS slotted in as the decision-monad component. The AGI 2025 paper positions MetaMo as sitting "one level higher as an abstract design language." Where OpenPsi had four hand-tuned modulators, MetaMo generalizes to a \(K\)-dimensional modulator vector governed by the appraisal comonad. Where OpenPsi selected actions by comparing urge magnitudes, MetaMo composes appraisal and decision via the pseudo-bimonad with checkable stability certificates.

The relationship is best read as MetaMo subsumes/extends OpenPsi at the formal layer — not replacement. OpenPsi's executable substrate (in hyperon-openpsiat3b356c5 HEAD ) operates as the operational predecessor; targeted searches for "MetaMo" / "MAGUS" / "individuation" / "transcendence" in that repo return zero matches, confirming hyperon-openpsi has not been re-authored as a MetaMo strict-literal implementation. The evolution preserves Psi's core insight — that motivation drives all goal-directed cognition — while providing the formal apparatus for safe self-modification, multi-agent coordination, and auditable decision-making.

Cluster-pilot extraction archive: scripts/archive/non_clustered_haa_pilot/source2_metamo_paper/.