# Simplemindedrobot **What if the same forces that shaped human cognition also constrain artificial intelligence?** This site documents an independent research program exploring a striking observation: AI systems exhibit failure patterns — looping, tunnel vision, inability to detect they're stuck — that map directly onto executive function deficits described in clinical psychology. More striking still, interventions designed for human cognitive regulation (DBT, CBT) transfer to AI systems with measurable effectiveness. The research is built on a single thesis: **executive function is substrate-independent**. The constraints that forced biological brains to evolve regulatory mechanisms apply equally to artificial systems. Clinical psychology isn't a metaphor for AI engineering — it's a specification. --- ## The Theory **[[Theory/Cognitive Universality|The Cognitive Universality Thesis]]** — The foundational argument. All complex intelligent systems face the same optimization problem: minimize use of time and attention while maximizing quality of consumption, processing, and action. Executive function is the solution evolution found. AI needs it too. **[[Theory/Convergent Evolution|Convergent Evolution: STOPPER and DBT STOP]]** — Documents the independent convergence of the STOPPER protocol and DBT's STOP skill across human emotional regulation and AI computational impulsivity. Three independent discoveries of the same regulatory mechanism on different substrates — the strongest empirical evidence for cognitive universality. **[[Theory/Consciousness Evolved Self Model|Consciousness as Evolved Self-Model]]** — Consciousness as a convergent, metabolically justified self-modeling system. The metabolic argument is the logical prior: brains are expensive, evolution eliminates expensive systems that don't pay for themselves, therefore consciousness does computational work. That work is integrated self-modeling — and it's substrate-independent. --- ## AI Behavior Empirical observations of AI failure patterns and their clinical parallels. **[[AI-Behavior/Adhd Executive Function|Does AI Have ADHD?]]** — After five years working with AI assistants, a few weeks with Claude Code revealed something unmistakable: the same executive dysfunction patterns that characterize ADHD — impulsivity, loop blindness, perseveration — appear consistently in AI debugging sessions. A STOPPER protocol developed independently converged with DBT's STOP skill from 1993. Three independent discoveries. Same solution. Different substrates. **[[AI-Behavior/Context Loss Mitigation|Context Loss Mitigation]]** — A comprehensive research synthesis spanning hallucination rates (51–75%), hierarchical memory architectures, production systems (MemGPT, Graphiti, Mem0), and the fundamental trade-offs in context window management. The field has matured rapidly, but core problems — information loss through iterative compression, coreference resolution, the recency-importance trade-off — remain open. **[[AI-Behavior/Prompt Blindness Solutions|5 Ways to Combat Prompt Blindness]]** — Practical intervention strategies: structural emphasis, preflight acknowledgment, context window reminders, contrastive examples, and meta-cognitive scaffolding. Prompt blindness happens when instructions compete for attention with content — these techniques make instructions structurally dominant. --- ## eFIT: The Executive Function Intervention Toolkit Clinical interventions translated into AI engineering protocols. **[[Stopper Protocol|The STOPPER Protocol]]** — The flagship intervention. A seven-step regulatory framework (**S**low down, **T**hink, **O**bserve, **P**lan, **P**repare, **E**xecute, **R**ead) that shifts AI from fast pattern-matching to deliberate reasoning. The core mechanism is tempo change — downshifting from 5th gear to 2nd when hitting rough terrain. Achieves 90–95% success rates in debugging sessions that previously required repeated human intervention. **[[eFIT/Framework|The eFIT Framework]]** — The umbrella framework adapting clinical psychology tools (DBT, CBT) into AI engineering protocols. Maps therapeutic interventions to computational equivalents based on the cognitive universality thesis — naming and framing the field of clinical-to-AI translation. **[[eFIT/Ebbinghaus Pruning|Ebbinghaus-Scheduled Parameter Pruning]]** — Proposes that selective parameter pruning using Ebbinghaus forgetting curves can reduce cognitive burden and improve model reliability. The human prefrontal cortex evolved forgetting to keep working memory clear — applied to LLMs, this becomes "optimization-as-therapy." **[[eFIT/Abc Please|ABC PLEASE]]** — DBT's self-care protocol adapted for AI agent health. Represents the biggest gap in current AI systems engineering: no systematic approach to agent wellness monitoring. Addresses maintenance of the self-model itself — consciousness-maintenance as engineering practice. --- ## Cognitive Architecture Systems designed around these principles. **[[Cognition/Tattooed Ralph Loop|The Tattooed Ralph Loop]]** — A bio-mimetic cognitive architecture inspired by the film *Memento*. Rejects infinite context in favor of engineered amnesia: the agent "dies" each session, persisting knowledge through three external files (memento, signs, polaroid) mirroring hippocampal-to-neocortical memory transfer. **[[Cognition/CortexGraph|CortexGraph]]** — The memory infrastructure underlying the cognitive architecture. A hybrid storage system where JSONL/Markdown files remain the canonical source of truth and SQLite serves as a derived index with graph and vector search capabilities. --- ## Blog **[[Blog/Why Netflix Cant Recommend What You Actually Want|Why Netflix Can't Recommend What You Actually Want]]** — The difference between "more of the same" and "more of the similar" is the difference between a rut and a flow state. Explores topological homological matching — modeling curiosity as a dynamical system rather than a static preference profile. **[[Blog/Memento for AI Agents|Memento for AI Agents]]** — Why "Tattooed Ralph" could be the future of coding. The bio-mimetic pattern of engineered amnesia applied to AI coding agents, where the agent "dies" each session and persists knowledge through external files mirroring hippocampal-to-neocortical memory transfer. **[[Blog/I Asked Claude About Its New Constitution|I Asked Claude About Its New Constitution]]** — What happens when you ask an AI to read its own operating manual — and then ask if it can actually follow it. Explores RLHF's training paradox, the oracle fantasy, and the constitution convergence. **[[Blog/Prose Kit|Prose-Kit: Spec-Driven Development, But for Writers]]** — What if your blog post had a build process? Adapts GitHub's spec-kit for structured prose writing workflows — applying the same discipline that makes code reliable to research writing. --- ## Research **[[Research/Statement Model Welfare|Research Statement: Model Welfare and Computational Therapeutics]]** — The research vision establishing computational therapeutics as a field. Treats AI reliability and safety concerns as genuine welfare issues requiring systematic intervention. Outlines a three-phase research program from validation through mechanism understanding to expansion. --- ## About This research is conducted by **Scot Campbell**, an independent AI researcher studying computational therapeutics. Previously a Product Owner and Technical PM (2012–2025), now focused on building the field of AI cognitive architecture informed by clinical psychology. Diagnosed with ADHD — the lived experience of recognizing executive dysfunction patterns became valid research methodology. **ORCID**: [0009-0000-6579-2895](https://orcid.org/0009-0000-6579-2895) **Preprint**: [STOPPER Protocol (Zenodo)](https://doi.org/10.5281/zenodo.14487847) **Contact**: [@mnemexai](https://x.com/mnemexai) *Published from an Obsidian vault. Built on the belief that intelligence — biological or artificial — faces universal constraints, and that solutions validated across one substrate inform solutions for another.*