Scientific record
Checked: 2026-10-14.
Hermes does not claim to invent synthetic cognition, cognitive architectures, agent memory, or affective computation. Source S49 ([private source identifier withheld]) corrected the earlier category confidence.
| Primary/first-party source | Supported claim | Limit |
|---|---|---|
| ACT-R — Carnegie Mellon | Established architecture intended to simulate and explain human cognition. | Does not demonstrate reproduction of every human capability. |
| Soar — official overview | Since 1983, a general cognitive architecture combining decision making, memory and learning mechanisms. | Broad program goals are not fully achieved human cognition. |
| Park et al., Generative Agents (2023) | LLM agents with experience records, retrieval, reflection, planning, and reported ablations of believable behavior. | Believability does not establish human-equivalent mechanisms or experience. |
| Packer et al., MemGPT (2023; revised 2024) | Hierarchical memory management and control flow extend usable LLM context; tested in document analysis and multi-session conversation. | Memory engineering is not biological-memory reproduction. |
| USC ICT — computational emotion research (2010) | Computational appraisal and virtual humans modeling emotion's influence on thought and behavior substantially predate Hermes. | First-party account, not the full EMA specification. Two older EMA PDFs were unavailable during verification. |
| Croissant et al., Appraisal-Based Chain-of-Emotion (2023) | Affective LLM agents have been constructed and tested on emotional-intelligence tasks and game-agent emotion simulation. | Reported behavioral gains do not establish all regulatory functions targeted by Affect. |
Hermes interpretation
Proposed emphasis: capability decomposition → empirical replication → perturbation → modular recombination → cognitive evolution. Begin with a capability boundary, test its empirical distinguishability, instantiate and vary it, then test combinations.
The sources establish adjacent traditions, not that this combination is unique or unoccupied. A stronger claim requires deeper capability-specific comparisons. Synthetic Cognition Infrastructure is broad/adjacent language. At the October 14 review, Experimental Cognitive Engineering and Cognitive Evolution Engineering were exploratory names. On October 24, the user adopted Experimental Cognitive Engineering as Hermes’ descriptive methodological orientation; this does not change the prior-art finding or establish disciplinary novelty. Cognitive Evolution Engineering remains exploratory. See current method.
The discussion also named CORTEX/ATERNA-AI, Synthetic Cognition Labs, Archotec, Project Deca, Kaine, and NTNU work. Their descriptions remain historical search claims in the source extract; this targeted verification did not establish every description, current status, or performance.
For each comparison, inspect implementation, operational tasks, interventions, reproducibility, and limits. For Affect, distinguish expression, appraisal, and regulation of attention, salience, valuation, memory, motivation, risk perception, and action selection. For Fiction, compare useful transformation while retaining retrieval as substrate.
See thesis evolution (not included in this public selection: ../../lab/thesis-evolution.md) for the reasoning behind the correction.