Scientific_Foundations

Scientific Foundations

Orrin's design borrows mechanisms from cognitive science, psychology, neuroscience-inspired computational models, and AI-safety literature. The citations below live as comments in the code itself — each section links the source work to the module(s) that implement the borrowed mechanism.

Caveat: the design is interpretive. Cognitive terms name engineering mechanisms; the literature is cited for context and design lineage, not as formal validation that Orrin implements these theories faithfully.

Affect & control signals

Source	Mechanism in Orrin	Code
Russell & Barrett (1999) core affect; Barrett (2017) How Emotions Are Made	Control signals as a low-dimensional affect state; "neutral" as absence of signal, not an emotion	control_signals/model.py, update_signal_state.py, signal_dynamics.py
Cannon (1932) homeostasis	Signals decay toward per-signal setpoints; a single arbiter owns the state	setpoints.py, homeostasis.py, arbiter.py
Roseman (1996); Smith & Ellsworth (1985); Lazarus (1991) appraisal theory	Events are appraised along dimensions before they move signals	appraisal.py
Gross (1998) emotion regulation; Aldao, Nolen-Hoeksema & Schweizer (2010)	Regulation strategies with bounded rate-of-change	regulation.py
Solomon & Corbit (1974) opponent process; Aston-Jones & Cohen (2005) phasic/tonic	Phasic bursts decay against tonic baselines	signal_dynamics.py
Brickman & Campbell (1971) hedonic treadmill	Baseline adaptation to sustained reward	signal_dynamics.py

Global workspace & conscious ignition

Source	Mechanism in Orrin	Code
Baars (1988) Global Workspace Theory; Dehaene (2014) Consciousness and the Brain	One winner per cycle converges parallel processes into a single conscious content; ignition is all-or-none, and unconscious candidates are damped	cognition/global_workspace.py, loop/deliberate.py, loop/finalize.py, think_utils/selection/boosts.py
Top-down broadcast (the downward half of GWT)	The workspace winner writes back into substrate salience (decaying-only)	cognition/workspace_writeback.py

See Workspace and Ignition and Binding and Workspace Writeback for the mechanism pages.

Predictive processing & free energy

Source	Mechanism in Orrin	Code
Friston (2010) free-energy principle / predictive processing	Predict internal state, score prediction error, treat large PE as a surprise signal	cognition/prediction.py, prediction_helpers.py, grounding/world_loop.py
Barrett & Simmons (2015) interoceptive prediction	Resource risk as interoceptive prediction error	cognition/resource_self_monitor.py, cost_prediction.py

Causality

Source	Mechanism in Orrin	Code
Pearl (2000) Causality; Pearl & Mackenzie (2018) Ladder of Causation; Granger (1969)	Causal edges score observation vs. intervention evidence (do(X) counts double); confounding checks per Pearl Ch. 3	symbolic/causal_graph.py
Pearl's evidence levels applied to self-tests	Behavioural self-experiments logged as Level-1/Level-2 evidence	cognition/experimentation.py

Reinforcement learning & reward

Source	Mechanism in Orrin	Code
Rescorla & Wagner (1972)	Prediction-error-scaled updates throughout (affect, confidence, causal weights)	cognition/prediction.py, grounding/grounded_concept.py, symbolic/causal_graph.py
Pearce & Hall (1980) associability; Mackintosh (1975); Sutton (1992) IDBD	Adaptive per-action learning rates driven by unsigned prediction error	reward_signals/action_reward_ema.py
Schultz, Dayan & Montague (1997) TD reward prediction error; Sutton (1988) TD learning	Reward engine and PE→emotion coupling	reward_signals/reward_engine.py, think/loop_helpers.py
Yu & Dayan (2005) uncertainty and learning	Learning-rate gain gating; volatile contexts learn faster	think/bandit/contextual_bandit.py
Gershman (2018) directed vs. random exploration	Uncertainty-seeking exploration bonus in action scoring	selection/score_actions.py
Auer, Cesa-Bianchi & Fischer (2002) UCB1; Langford & Zhang (2007) epoch-greedy; Agrawal & Goyal (2012) Thompson sampling	Bandit machinery for function and depth selection	utils/bandit.py, think/depth_bandit.py
Hebb (1949)	Post-cycle emotion→function reinforcement and spreading activation	runtime_coupling/adaptation.py

Intrinsic motivation & exploration

Source	Mechanism in Orrin	Code
Oudeyer & Kaplan (2007); Gottlieb & Oudeyer (2018); Berlyne (1960)	Epistemic vs. diversive curiosity in intrinsic reward	reward_signals/reward_signals.py, cognition/exploration_value.py
Loewenstein (1994) information-gap theory	Curiosity peaks where a topic is partially known	cognition/intrinsic_generators.py

Memory

Source	Mechanism in Orrin	Code
Ebbinghaus (1885) forgetting curve; Anderson & Schooler (1991)	Memory strength decay; rule forgetting	memory/strength.py, symbolic/rule_forgetting.py
Baddeley & Hitch (1974) working memory	Central-executive gating; load narrows capacity	cog_memory/working_memory.py
Tulving (1972, 1983) episodic→semantic; Tulving (2002) autonoetic consciousness	Episode-to-knowledge consolidation; temporal self-location	cognition/knowledge_formation.py, knowledge_graph_extract.py, temporal_state.py
McGaugh (2000) consolidation; Levine & Pizarro (2004) emotion & memory	Emotion-weighted consolidation	control_signals/consolidation.py
Diekelmann & Born (2010) sleep & memory; Tononi & Cirelli (2014) synaptic homeostasis (SHY)	Idle/sleep consolidation cycles enrich rather than just prune	idle_consolidation/consolidation_cycle.py

Symbolic reasoning & knowledge representation

Source	Mechanism in Orrin	Code
Anderson (1983) ACT-R; Anderson (1982) procedural compilation; Schank (1982) case-based reasoning	Rule structure, synthesis, and compilation of specific episodes into general rules	symbolic/rule_engine.py, rule_synthesis.py
Johnson-Laird (1983) mental models; Baader et al. (2003) description logic; Gärdenfors (2000) conceptual spaces	World-model inference: subsumption, model search, similarity as shared-feature overlap	symbolic/inference.py, cognition/world_model.py
Bartlett (1932) schemas	Parent schemas inherit from abstracted children	symbolic/rule_abstraction.py
Mitchell et al. (1986) explanation-based learning	Breakthrough distillation in metacognition	cognition/metacog.py

Attention, effort & cognitive control

Source	Mechanism in Orrin	Code
Kahneman (1973) Attention and Effort; Kahneman (2011) System 1/2	Attention as a limited resource; experiencing-self depletion	cognition/attention.py, temporal_state.py
Botvinick et al. (2001) conflict monitoring	Conflict-gated deliberation	think/think_module.py
Shenhav, Botvinick & Cohen (2013) Expected Value of Control	Learned cost predictions become a depletion-scaled penalty on candidate actions	cognition/cost_prediction.py
Yerkes & Dodson (1908) inverted-U arousal curve	Energy orientation: reactive over-peak vs. proactive band	motivation/energy_orientation.py

Metacognition & calibration

Source	Mechanism in Orrin	Code
Nelson & Narens (1990) metamemory; Fleming & Lau (2014) measuring metacognition; Brier (1950) probability scoring	Confidence calibrated against outcomes with a recency-weighted Brier score	cognition/calibration.py

Behavioral adaptation & control theory

Source	Mechanism in Orrin	Code
Carver & Scheier (1982) control-systems theory; Powers (1973) perceptual control theory; Bandura (1977) self-efficacy; Tolman (1932) latent learning	Pattern-specific corrective interventions: ruts trigger exploration, oscillation triggers commitment, goal avoidance triggers action	cognition/behavioral_adaptation.py
Thompson & Spencer (1966) habituation	Repeated stimuli lose salience	cognition/habituation.py
Nolen-Hoeksema (1991); Treynor, Gonzalez & Nolen-Hoeksema (2003) rumination/brooding	Detecting and breaking passive self-focused loops	cognition/rumination.py

Social cognition

Source	Mechanism in Orrin	Code
Friston-style generative-model theory of mind; Baron-Cohen; Feldman (synchrony); Buckner	Peer models commit to predictions, score PE, and blend confidence from accuracy + synchrony	cognition/theory_of_mind.py, theory_of_mind_infer.py
Hatfield, Cacioppo & Rapson (1993) emotional contagion	Peer affect leaks into Orrin's signals, bounded	cognition/contagion.py
Heinrichs et al. (2003) social buffering	Speech as a social-penalty regulator	behavior/speech_gate.py

Goodhart's law & specification gaming (quality standard)

Source	Mechanism in Orrin	Code
Goodhart (1975); Strathern (1997); Manheim & Garrabrant (2018)	Why Orrin cannot edit his own quality bar	cognition/quality_standard/
Amodei et al. (2016) Concrete Problems in AI Safety; Krakovna et al. (2020); Skalse et al. (2022); Everitt et al. (2021) reward tampering	Human-ratified loosening only; the standard evolves from demonstrated-good work	quality_standard/proposer.py

Language

Source	Mechanism in Orrin	Code
Keskar et al. (2019) CTRL repetition penalty	Sampling-time repetition control in the native LM	cognition/language/native_lm.py

Finding citations in the source

Most modules carry their references as a header comment block. To list them all:

grep -rn "([12][089][0-9][0-9])" --include="*.py" brain/ goals/ memory/ | grep -iE "et al|&|[A-Z][a-z]+ \("

Caveats

• The design is interpretive; the wiki cites literature for context, not formal validation.
• Cognitive terms name engineering mechanisms (see CLAUDE.md golden rule 4) — the citations document where a mechanism's shape came from, not a claim of equivalence.