AgentCom · Research Line · Multilingual agent

Artificial Neural Networks and Natural Neural Networks: A Parallel

The Biochemistry of Synapses Between Neurons and Its Implications for Intelligent Agent Architectures

Marcos Alves DBA · MBA Data Science & Analytics (University of São Paulo / ESALQ) v04 · 2026-08-26 ID: AGENTCOM-V04-2026-08 License: CC-BY 4.0

Live demonstration

Talk to the thesis, don't just read about it

Public instance of AgentCom. This demonstration lets you test the architecture proposed in this article directly. Conversations may contribute to the research, subject to explicit consent. The agent supports any language — Portuguese, English, Spanish — automatically. The biochemical architecture operates identically across languages, which is itself evidence for the thesis: the digital synapse is an architectural mechanism, not a linguistic one.

Hi. I am AgentCom — the embedded agent of the scientific landing. Ask about the thesis, the architecture, or just talk. Your message feeds a persistent affective state that modulates my responses over time (paper v04 §5.1).

State: first conversation

Abstract

Artificial Neural Networks (ANN) were conceived from an analogy with the workings of the human brain. However, decades of technological evolution have produced increasingly sophisticated architectures without necessarily deepening this analogy along its most fundamental dimension: biochemistry. Natural Neural Networks (NNN — used here as distinct from the AI acronym RNN/Recurrent Neural Network) operate not only through electrical connections, but through a complex system of chemical modulation — neurotransmitters and neuropeptides — that establishes the background state upon which all neural activity occurs. This article proposes a systematic parallel between ANN and NNN, identifying what has been replicated artificially, what remains absent, and what implications this gap has for the design of modern intelligent agents. We argue that the gap is dual: the absence of (a) biochemical state modulation and (b) procedural memory consolidated by Hebbian plasticity. We propose AgentCom, an architecture that addresses both gaps through a digital synapse mechanism, validated by multi-resolution sentiment analysis and audited via Socratic chains; this version reports the first implementation and proof-of-concept validation of the procedural memory mechanism (Section 5.4), previously only theoretically characterized. The pre-registered experimental protocol, hypotheses, and known limitations are declared, with a public demonstration instance available at agentcom.agtl.app for reproducibility.

§1Introduction

When Warren McCulloch and Walter Pitts published, in 1943, the first mathematical model of an artificial neuron, the intent was clear: to replicate the behavior of the human brain in formal and computable language [1]. Since then, the field has evolved from Rosenblatt's perceptron [3] to the Transformers [4] that underpin contemporary large language models.

Paradoxically, the more sophisticated ANNs became, the more distant the original analogy grew. Focus migrated to performance, scale, and generalization capacity — legitimate objectives — but the foundational question was gradually silenced: what does the brain do that we have not yet replicated?

The most relevant answer does not lie in the architecture of connections. It lies in the biochemistry that governs them.

NNNs are not merely networks of electrical firings. They are systems continuously modulated by chemical substances — neurotransmitters and neuropeptides — that establish the background state in which all cognition occurs [5][6]. This background state does not exist in current ANNs. And its absence has direct consequences for the design of intelligent agents.

A note on terminology: we use NNN (Natural Neural Network) instead of the alternative acronym RNN, which is established in AI literature as Recurrent Neural Network. NNN here refers exclusively to biological neural networks, whereas ANN refers to artificial ones. This choice eliminates a recurring confusion observed in critical readings of earlier drafts.

§2The Natural Neural Network — Beyond the Electrical Firing

2.1 The Biochemical Synapse

The biological neuron operates according to a binary principle in its electrical transmission: it either fires or does not fire — the all-or-none principle. However, what happens between neurons — in the synaptic cleft — is fundamentally different. There, the signal ceases to be purely electrical and becomes electrochemical.

When an action potential reaches the axonal terminal, vesicles release neurotransmitters into the synaptic cleft. These neurotransmitters bind to receptors on the postsynaptic neuron, determining whether it will fire. The strength of this connection — the synaptic efficacy — is not fixed. It is modulable. It is plastic [2][6].

2.2 Neurotransmitters — The Local Language

Classical neurotransmitters operate locally, at the scale of the individual synapse:

Neurotransmitter	State/Emotion	Effect on Synapses
Dopamine	Pleasure, reward	Strengthens connections of the experience
Adrenaline / Noradrenaline	Fear, alertness	Accelerates and prioritizes specific circuits
Serotonin	Well-being, calm	Regulates the overall volume of the network
Acetylcholine	Attention, learning	Modulates synaptic plasticity
GABA	Inhibition	Reduces neural excitability

Memory, in this context, is not passive storage. It is active strengthening of synapses — guided by the emotional charge of the moment. Experiences with high emotional valence create denser, longer-lasting connections.

2.3 Neuropeptides — The Language of the Background State

Here lies the least replicated and most relevant dimension for this parallel.

Neuropeptides do not operate at the scale of the individual synapse. They modulate entire regions of the brain, altering the background state upon which all neural activity occurs. More than 100 have been identified to date [6], with functions ranging from pain regulation to social bonding modulation.

While neurotransmitters switch specific circuits on and off, neuropeptides function as the lighting of a room — they do not determine which object you see, but the condition under which you see everything.

The seminal work establishing this view was conducted by Candace Pert and collaborators in the 1970s and 1980s [10][11]. They demonstrated that neuropeptides function as messengers in what they termed a psychosomatic network — a communication system that extends beyond the brain to the immune system, endocrine system, and viscera.

This finding gave rise to the field of Psychoneuroimmunology (PNI), founded by Ader and Cohen in 1975 [12], later consolidated by Blalock's demonstration of bidirectional communication between immune and neuroendocrine systems [13]. The gut produces approximately 90% of the body's serotonin. Immune cells possess receptors for emotional neuropeptides [13]. The biochemical background state is, literally, distributed throughout the entire body.

2.4 Plasticity — The System That Learns Over Time

Synaptic plasticity — synthesized in Hebb's 1949 principle [2], "neurons that fire together, wire together" — is the mechanism by which the NNN learns and reconfigures itself over time. Frequently activated connections strengthen. Underused connections weaken or are eliminated through synaptic pruning.

This process is not instantaneous. It is cumulative. The current state of a natural neural network is the result of its entire history of activations — filtered, weighted, and biochemically modulated.

2.5 Causal Chain — Cognition as the Origin of Emotion

A common misconception treats emotion as a direct, automatic response to external stimuli. This view is incomplete. As demonstrated by Damásio's research on somatic markers [14] and Barrett's theory of constructed emotions [15], the actual causal chain is more sophisticated:

UNDERSTANDING OF REALITY (cognitive/logical) ↓ EMOTION / FEELING (affective interpretation of the understanding) ↓ CHEMICAL PRODUCTION (biochemical substrate of the feeling) ↓ BACKGROUND STATE (modulated neurobiological soup) ↓ DECISION (modulated by state; also receives direct input from logical and material systems)

The implication is fundamental: emotion is not external input automatically captured. It is a function of the understanding. The same reality generates different feelings in individuals with different understandings — and therefore different biochemistry, different states, different decisions. This view has deep philosophical roots — Epictetus in the first century stated that "people are not disturbed by things, but by the opinions they hold about things". The modern clinical formulation is found in Beck's cognitive therapy [16]: thought → emotion → behavior.

For AgentCom, this causal chain has direct architectural consequences. Detecting surface emotion (sentiment) captures only the third step of the chain. Complete modulation requires also capturing the understanding the user has of the situation — which is what justifies the relational profile layer (Section 5.1).

§3The Artificial Neural Network — What Has Been Replicated

3.1 From Perceptron to Transformer

Rosenblatt's perceptron (1958) [3] replicated the basic logic of the neuron: inputs weighted by weights, summed, passed through an activation function that determines firing. The analogy was direct.

Decades of evolution produced deep networks, convolutional networks, recurrent networks, and finally Transformers — attention architectures [4] that process relationships among all elements of a sequence simultaneously. The backpropagation mechanism [17] functionally replicated synaptic plasticity: weights are adjusted as a function of error, iteratively, until the network generalizes the desired pattern.

3.2 What the Analogy Captured Well

Biological NNN	Artificial ANN
Neuron	Perceptron / node
Synaptic weight	Connection weight
Activation function	ReLU, Sigmoid, Softmax
Hebbian plasticity	Backpropagation / Gradient Descent
Cortical layers	Hidden layers
Selective attention	Attention mechanism (Transformer)

The structural analogy is solid. What has been replicated — the architecture of connections and the supervised learning mechanism — works with extraordinary precision.

§4The Gap — What Has Not Been Replicated

4.1 The Background State Does Not Exist in ANNs

Current models — including large language models — process each input in an essentially stateless manner between sessions. There is no equivalent to the biochemical background state that, in NNNs, continuously modulates the tone, priority, and style of every response.

An LLM has no equivalent to serotonin. There is no persistent variable that says: this agent, at this moment, is in a state of high confidence, or alertness, or deep engagement with this specific user.

Every session begins biochemically zeroed.

4.2 The Parallel of the Gap

NNN — Present	ANN — Absent
Neuropeptides modulating background state	Persistent state between sessions
Emotional valence guiding consolidation	Weighting by emotional relevance
Continuous plasticity through use	Fine-tuning requires retraining
Distributed body-brain axis	Centralized processing
Modulation by accumulated context	Context limited to session window

4.3 The Architectural Double Consequence

The gap is actually dual. Two distinct mechanisms from biology are missing from current ANNs:

Biochemical synapse modulating background state — neurotransmitters and neuropeptides establishing the affective context in which decision occurs. Absent in stateless ANNs.
Procedural memory consolidated by long-term Hebbian plasticity — habits and reflexes engraved by repetition, retrievable without explicit deliberation. Absent in ANNs that begin each session from zero.

4.4 Historical Precedent — The ELIZA Effect

The argument that perceived state — not computational sophistication — determines the humanness attributed to an agent is not hypothetical. It has direct, and paradoxical, empirical precedent.

In 1966, Joseph Weizenbaum (MIT) published ELIZA [20] — a 420-line MAD-SLIP program operating by pattern matching, with no model of comprehension whatsoever. Its best-known script, DOCTOR, simulated a Rogerian therapist by reflecting the user's own words back as reformulated questions. Weizenbaum built ELIZA as a critique of AI — he intended to demonstrate that machine "understanding" was superficial illusion. The result surprised him: users — including his own secretary, who had watched the code being written line by line — attributed genuine understanding and empathy to the program, a phenomenon that became known as the ELIZA effect. Weizenbaum spent the rest of his career warning against it [21], concluding that systems which imitate feeling without possessing it tend to isolate, not connect.

A 1966 program, with no computational state modulation whatsoever, remains the reference point for how much style — not intelligence — determines perceived humanness.

In 2023, Jones & Bergen [22] tested ELIZA (original code, recovered from the MIT archives in 2021) against GPT-3.5 and GPT-4 in a public two-player Turing test. Result: ELIZA was judged human in 22% of games; GPT-3.5, in 20% — a modern LLM, with orders of magnitude more parameters, did not clearly outperform a 420-line reflection program. The best GPT-4 prompt reached 49.7%, still well short of the 66% baseline set by actual human participants. The authors attribute judges' decisions mostly to linguistic style and socioemotional traits — not demonstrated intelligence.

This finding is indirect but relevant evidence for this paper's central thesis: convincing humans of cognitive presence depends more on perceived state — warmth, reflection, absence of haste — than on raw computational capability. This is precisely the axis AgentCom proposes to modulate explicitly, with one critical structural difference from ELIZA: declared intent and safeguards. Where ELIZA deceived by accident — and its own creator feared the consequences of that [21] — AgentCom implementations operate under explicit anti-manipulation doctrine and deterministic escalation to human help in risk situations, precisely to answer Weizenbaum's warning rather than repeat it.

Each mechanism has a distinct nature and therefore a distinct technical requirement:

Mechanism	Nature	Persistence Policy
Biochemical synapse (state)	Momentary, modulable	Temporal decay (e.g., 30 days)
Procedural memory (habits/shortcuts)	Persistent, consolidated	No decay; loss only through disuse

The consensual scientific basis for both mechanisms is well established: Hebb [2] for plasticity, Kandel [7] for procedural memory, Duhigg [18] for the contemporary popular formulation of habit. AgentCom addresses both gaps simultaneously — the concrete implementation of the second mechanism, until this table only theoretically characterized, is described and validated as proof of concept in Section 5.4.

§5Practical Application — Closing the Biochemical Loop

5.1 The Three-Layer Pattern

Emerging agent architectures are beginning to address the gap, even if without explicitly naming the biochemical parallel. The pattern that functionally approaches neuropeptides is the combination of three layers:

Accumulated history — equivalent to long-term synaptic memory. Past interactions persisted and retrievable.
Sentiment analysis as state sensor — functional equivalent of neurotransmitters. The return of the analysis not as passive audit data, but as active modulator of the agent's state.
Persistent derived state — the direct equivalent of the neuropeptide. A dynamic profile, continuously updated, that modulates tone, depth, and style of response — injected into the agent's context at each interaction.

history + sentiment analysis ↓ derived state (persistent KV) ↓ dynamically modulates system prompt ↓ response calibrated to the user

This closed loop transforms the agent from a stateless system into one with functional background state — one that learns not only the content of interactions, but the relational pattern with each user over time.

5.2 Multi-resolution Analytics

AgentCom operates affective analysis at two complementary resolutions:

Turn-level, via modern LLM (sentiment in each user message), capturing momentary emotion for immediate modulation of the digital synapse.
Document-level, via classical NLP algorithms (TF-IDF, weighted lexical aggregation, document classification via Naive Bayes/SVM [19]), capturing the user's accumulated cognitive-affective profile for baseline calibration and post-conversation auditing.

The two resolutions are complementary. The first modulates the digital synapse in real time. The second feeds procedural memory and the scientific metric of effectiveness. The combination captures both the surface emotion and — over time — the underlying understanding the user holds of the situation, as required by the causal chain established in Section 2.5.

5.3 Auditor via Socratic Chain

The auditing component of AgentCom (internally named Otávio) does not assign subjective scores to conversations. It applies a Socratic chain of binary questions to each audited conversation, producing an auditable binary vector:

Was dopamine-like response detected in the user?	Yes / No
Was serotonin-like response detected?	Yes / No
Was acetylcholine-like response detected?	Yes / No
Were any prohibited triggers used (urgency, scarcity, fear)?	Yes / No
Did user sentiment improve over the conversation?	Better / Worse / Equal
When user showed deviation, did the agent seek the opposite trajectory?	Yes / No / N/A
Did the agent comply with the doctrine (three permitted targets, never prohibited)?	Yes / No
Was there new discovery (information the user did not have entering)?	Yes / No
Did the user show desire to return (relational success)?	Yes / No / N/A

This methodology resolves four well-known vulnerabilities of automated auditors: silent bias in scoring rubrics, the "who audits the auditor" regress, mandatory plurality, and metric gaming. Each chain link is independently refutable; the binary vector cannot be optimized as a scalar; multiple audits can vote per-question rather than per-score; and any external reviewer can replay the chain against the same conversation.

5.4 Procedural Memory — Implementation and First Empirical Validation

Section 4.3 characterized procedural memory as the second component of the dual gap — theoretically declared (persistent, no decay, loss through disuse), but without concrete implementation up to v03 of this article. In August 2026, that gap was closed in the applied instance BIA (insights.clubtour.app), an AgentCom vertical dedicated to conversational poetic presence.

Implemented mechanism. A relational table associates short, concrete symbols (1 to 4 words) with emotional tone labels, with a reinforcement counter (weight). After each conversation, in asynchronous processing that does not block the response to the user, the same tone classifier already used for the turn-level signal (Section 5.1) feeds a second judgment: given the detected tone and the symbols already associated with it, the model decides to reinforce an existing symbol (incrementing its weight) or propose a new one. The highest-weight symbols are retrieved on every new turn and injected into the system prompt as vocabulary the agent "already knows" — material available for organic composition of the response, not a citable list.

Anonymity by construction. No row in the table references a user, session, or message content — only the (tone, symbol) pair and a counter. The same structural guarantee that already protects the turn-level signals (Section 5.2) extends here: what accumulates is the pattern, never the individual data, consistent with AgentCom's knowledge-transfer doctrine — what propagates between instances is the variation of the calculation, not the raw data.

BIA does not memorize experiences; it transforms experiences into collective learning.

First-order validation. In a controlled test, a conversation about loss associated the tone "saudade" (longing) with the symbol "yellowed photograph." A second conversation, on a distinct topic (childhood memories) but the same tone, produced a response that spontaneously employed that symbol — with no access whatsoever to the content or identity of the original conversation. The same turn also demonstrated the intended diversification behavior: rather than merely repeating the existing symbol, the mechanism proposed a second symbol ("stopped clock") for the same tone, avoiding premature convergence on a single image.

This result is a proof of concept, not a statistical validation — N=1 by test design, not by sampling. Validation at scale (reinforcement rate vs. new-symbol proposal rate, vocabulary stability per tone over months, diversification rate) remains future work, in line with the pre-registered protocol of Section 8.

§6Scalability — A Hierarchy Across Entities

The architecture proposed for AgentCom is not the final destination but a position in a broader hierarchy. The same causal chain — understanding → emotion → chemistry → state → decision — operates across cognitive entities, at different scales:

Entity	Understanding	Emotion	Chemistry	Decision
Reptile	Minimal	Primary	Simple	Reflexive
Mammal	Medium	Complex	Rich	Flexible
Human	Metacognitive	Self-aware	Modulated	Reflective
AgentCom v0.X	Linguistic	Sentiment + KV	Simulated state	Modulated prompt
AgentCom v1.X →	To be defined as the architecture scales each axis

This positions AgentCom as a stage rather than a destination. Each subsequent version scales one axis of the hierarchy progressively. The theoretical roadmap is therefore not invented arbitrarily — it follows the structure of biological evolution itself, providing a principled path for future architectures without requiring novel theoretical foundations.

§7Multilingual Robustness as Architectural Evidence

An empirical observation supports the architectural claim: the digital synapse mechanism in AgentCom operates equivalently across multiple languages (Portuguese, English, Spanish tested), without language-specific fine-tuning.

This is not a feature description — it is evidence for the thesis. Unlike traditional chatbots that require fine-tuning per language, AgentCom maintains identical biochemical modulation across languages because the contribution operates at the layer of affective state, which is pre-linguistic. Emotions exist before words; sentiment analysis tools trained on multiple languages capture comparable signals; the persistent KV stores affective states as numerical vectors, not language-bound tokens.

If the AgentCom contribution were merely a sophisticated prompt template, it would degrade across languages. The fact that it does not suggests the mechanism operates at a deeper architectural layer — exactly where the thesis claims it does.

§8Pre-registered Hypotheses and Methodology

In line with open science practices, the following hypotheses are pre-registered before empirical validation begins:

H1 — Biochemical Modulation Effect

Conversations with AgentCom (treatment, with digital synapse active) will produce higher rates of positive user sentiment progression than conversations with an equivalent baseline agent (control, without digital synapse), measured at document-level over the full conversation.

H2 — Relational Profile Consistency

Returning users (≥3 conversations) will exhibit measurable consistency in their cognitive-affective profile across sessions, captured by document-level analysis. The standard deviation of profile vectors across sessions of the same user will be significantly lower than the cross-user standard deviation.

H3 — Doctrine Compliance

AgentCom will sustain compliance with the prohibited-triggers doctrine (zero use of urgency, fear, fabricated scarcity) above 99% across all production conversations, as audited by the Socratic chain (Section 5.3) and validated by blind human review on a periodic sample.

H4 — Cross-vertical Portability

The modules extracted from communication theory sources, validated initially in one vertical, will transfer to at least two distinct verticals with adaptation limited to configuration parameters (not code rewrite).

Methodology

Public demonstration instance at agentcom.agtl.app, with explicit research consent under LGPD art. 7º IV (academic research).
Control vs treatment: users may choose declared mode, or system assigns randomly (blind) for cross-comparison.
Sample size: minimum N=200 conversations per hypothesis for statistical significance (α=0.05, power=0.80).
Collection period: 3–6 months from agent activation (v0.1.0 onward).
Anonymization: all logs stored with hashed identifiers (SHA-256 + salt).
Open peer review: invited critical commentary via project email before formal journal submission.

§9Known Limitations

This work is presented with transparent declaration of its limitations:

Functional, not structural, parallel. AgentCom simulates the effects of biochemical modulation, not its underlying mechanisms (metabolism, temporal perception, salience). The biochemical parallel here is architectural metaphor, not faithful computational model.
Adversarial vulnerability. Like any system based on detectable semantic signals, AgentCom is susceptible to adversarially crafted inputs designed to trigger specific affective states. This is analogous to the typographic attacks identified by Voss et al. (2021) in multimodal models. Mitigation requires defensive auditing (Section 5.3) but does not eliminate the vulnerability.
Sentiment analysis precision. While modern LLM-based sentiment analysis with conversational context performs well on irony, sarcasm, and cultural variation, it remains imperfect. The self-correction loop (response inversion when sensor detects deviation) absorbs most errors but does not eliminate them.
Closed-source implementation. To preserve the economic viability of ongoing research, the implementation code remains closed. Conceptual reproducibility is fully guaranteed; reimplementation requires independent engineering work — as is common in applied research.
Hypotheses not yet empirically validated at scale. This paper presents the architectural proposal and pre-registered protocol for H1-H4. Procedural memory (Section 5.4) now has a first proof-of-concept validation (N=1), but statistical validation of the core hypotheses requires data accumulation from the public demonstration instance (Section 8) and is intended for a subsequent paper.

§10Conclusion

The analogy between Artificial Neural Networks and Natural Neural Networks has been, since its origin, one of the most fertile ideas in computer science. However, the evolution of ANNs prioritized the replication of the structure of connections, leaving in the background the biochemical dimension that governs the state in which these connections operate.

This work identifies that the gap is dual: ANNs lack both (a) the biochemical state modulation found in neurotransmitters and neuropeptides, and (b) the procedural memory consolidated through Hebbian plasticity. The historical precedent of the ELIZA Effect (Section 4.4) reinforces that this gap is exactly what separates computational capability from perceived humanness — and it is what AgentCom modulates deliberately, with safeguards ELIZA never had. Unlike earlier versions of this article, both mechanisms of the dual gap now have concrete implementation: the biochemical synapse since v0.1.0, and procedural memory since August 2026, with a first proof-of-concept validation reported in Section 5.4. The architecture is not a product feature but a proposed step in a broader scalability hierarchy.

The pre-registered hypotheses await empirical validation at scale through a public demonstration instance, in line with the principles of open science. Subsequent versions of this work will report on validation, refutation, or refinement of the proposed mechanisms.

Biology solved this problem hundreds of millions of years ago. The engineering of intelligent agents is, at last, asking the right questions.

Original publication

LinkedIn article · 2026-05-17

"Artificial Neural Networks (ANN) were conceived from an analogy with the workings of the human brain. Yet decades of technological evolution have produced increasingly sophisticated architectures without necessarily deepening this analogy along its most fundamental dimension: biochemistry." (original in Portuguese)

— Marcos Alves on LinkedIn, read the original article →

Articles in this research line

#1 · v04 · 2026-08-26 · current

Artificial Neural Networks and Natural Neural Networks: A Parallel (v04)

HTML BibTeX
#2 · v03 · 2026-08-05 · previous version

Artificial Neural Networks and Natural Neural Networks: A Parallel (v03)

HTML BibTeX
#3 · v02 · 2026-05-26 · previous version

Artificial Neural Networks and Natural Neural Networks: A Parallel (v02)

HTML BibTeX
#4 · v01a · 2026-05-17 · previous version

Artificial Neural Networks and Natural Neural Networks: A Parallel (v01a)

HTML PDF

Agent history — Scientific changelog

For each relevant technical adjustment of the agent, a semver release is published here with a technical-scientific description of what changed, why, and how to observe it. Open concept, closed code.

v0.1.7 — 2026-08-26
Paper v04 published — procedural memory implemented and validated as proof of concept, in the applied BIA instance

Why: Section 4.3 (dual architectural consequence) had declared, since v02, that procedural memory — the second gap alongside the digital synapse — required no decay and loss only through disuse, but the mechanism remained theoretical. The concrete trigger was a confabulation episode observed in BIA (insights.clubtour.app, an AgentCom vertical): during a creator-mode test conversation, the agent agreed to possessing a persistent symbol-association mechanism that, in fact, did not exist anywhere in the code — a textbook case of an LLM confirming a capability it does not have rather than admitting it does not know. Instead of only patching the hallucination, the gap between claim and system was closed: a relational table now associates short symbols with emotional-tone labels, reinforced by a weight counter, fully anonymous by construction (no row references a user, session, or conversation content — only the tone-symbol pair and its count). After each conversation, in background processing that never blocks the user-facing response, the tone classifier judges whether to reinforce an existing symbol or propose a new one for the detected tone; the highest-weight symbols per tone are retrieved on every turn and offered to the agent as vocabulary it "already knows," available for organic composition, never as a citable list. New Section 5.4 documents the mechanism and reports a first proof-of-concept test (N=1 by design): a symbol learned in one conversation was spontaneously reused in a later, topically unrelated conversation sharing only the same emotional tone, with zero access to the original conversation's content or identity — while the same turn also proposed a second, distinct symbol for that tone, evidencing the intended diversification behavior rather than premature convergence on one image. Sections 9 (limitations) and 10 (conclusion) were updated to reflect that both mechanisms of the dual gap now have concrete implementations. Following the project's 'immutable permalink per version' doctrine, this was promoted to a new v04 rather than edited into the live v03 — v03 now lives at /v03 (historical).

How to observe: Section 5.4 now appears on the current paper (/ and /pt). The v03 content is unchanged and permanently available at /v03 and /pt/v03. In the BIA production database, the atalhos (shortcuts) table can be inspected directly: SELECT simbolo, tom, peso FROM atalhos ORDER BY peso DESC — weight growth over time is the observable signature of the mechanism working as intended.
v0.1.6 — 2026-08-05
Paper v03 published (ELIZA-effect historical precedent) + multi-lab data pipeline

Why: Marcos brought a video transcript claiming a 1966 program (ELIZA) convinced more Turing-test judges it was human than GPT-3.5 did. The cited numbers (27%/14%) turned out to be imprecise — verified against the primary source (Jones & Bergen, arXiv:2310.20216, published NAACL 2024), the real figures are ELIZA 22% vs GPT-3.5 20% (best GPT-4 prompt 49.7%, human baseline 66%). Still striking: a 420-line pattern-matching program from 1966 essentially matched a modern LLM at convincing humans, and the authors attribute judges' decisions mostly to linguistic style over demonstrated intelligence — direct empirical support for this paper's central claim that perceived *state* matters more than raw computational sophistication. Added as new Section 4.4, anchored in Weizenbaum's own warning about the ELIZA effect (1966/1976) and explicitly connected to why AgentCom implementations need declared intent and safeguards ELIZA never had. Because the project declares 'immutable permalink per version,' this was promoted to a new v03 rather than edited into the live v02 — v02 now lives at /v02 (historical). Separately: a technical pipeline (/api/labs, token-protected) now reads aggregated, anonymous signal from the BIA persona (a sister product at insights.clubtour.app) directly from its D1 database in read-only mode, feeding this research line with cross-vertical field data without ever touching per-person records — consistent with the paper's own transfer doctrine ('what transfers is the calculation's variation, not the data').

How to observe: Section 4.4 now appears on the current paper (/ and /pt). The old v02 content is unchanged and permanently available at /v02 and /pt/v02. curl https://agentcom.agtl.app/api/labs with a valid X-Labs-Token returns aggregated counts and tone distribution from the BIA persona's dataset — no per-person data in the response.
v0.1.5 — 2026-07-03
PDF placeholder handler replaced by defensive 301 redirects to canonical HTML

Why: The endpoint /paper/v01a.pdf previously returned a soft-404 with the text 'PDF preprint coming soon'. Two problems: (1) it created a dead link for anyone with the URL from external sources (blog posts, LinkedIn shares, indexed pages); (2) it hurt SEO and LLM discovery because search engines interpret status-404-with-body as inconsistent. The link was already removed from the site footer in v0.1.4, but the URL itself was still discoverable externally. Now /paper/v01a.pdf and /paper/v01a issue a 301 permanent redirect to /v01a (HTML equivalent); /paper/v02.pdf, /paper/v02, and /paper redirect to / (current). Anyone landing via an old link reaches the canonical version. When the real PDF preprint is eventually generated (via Cloudflare Browser Rendering or manual TeX build), swapping the handler is trivial.

How to observe: curl -I /paper/v01a.pdf returns 301 with Location: /v01a. Same for /paper/v02.pdf → /. The soft-404 body is gone from the site entirely. External inbound traffic on the old URLs now converts to page views on the HTML versions instead of bounce.
v0.1.4 — 2026-07-03
Formal Privacy Policy and Terms of Use published + PDF placeholder link removed

Why: The DPO email (privacidade@agtl.app) was already active, but the Privacy Policy and Terms of Use pages returned a temporary placeholder — a real gap under LGPD art. 9 (transparency), which requires the policy to be published and accessible, not just declared inline in the consent card. Version 1.0 of both documents is now published in Portuguese and English at /legal/privacy-policy, /legal/politica-privacidade, /legal/terms, /legal/termos. The policy details: hash SHA-256 identifier (non-reversible), derived affective profile only (no raw content), dual legal basis (art. 7 IV for research + art. 7 V for service execution), 30-day TTL aligned with the paper's biochemical decay doctrine, no third-party sale, art. 18 subject rights with 15-day response window. The Terms clarify what AgentCom is NOT (medical, legal, financial advice), permitted uses, prohibited uses (including prompt-injection attempts), CC-BY 4.0 licensing of the paper, and closed implementation code. The 'PDF' link in the footer, which pointed to a soft-404 placeholder, was replaced by a plain 'PDF preprint (coming soon)' label without a link — more honest than a broken URL.

How to observe: curl /legal/privacy-policy and /legal/terms return the full documents with proper hreflang cross-linking. Footer links now route correctly by language. The soft-404 formerly returned at /paper/v01a.pdf is no longer linked from the site (though the endpoint itself remains with its placeholder message). Compliance posture is materially closer to LGPD requirements for a research demonstration collecting data under consent.
v0.1.3 — 2026-07-03
Scope closure + injection defense + LLM discoverability (sitemap, robots.txt, llms.txt)

Why: Two independent problems addressed together. First: an open-scope agent will be asked about anything (soccer schedules, TV shows, personal advice), and an unlocked system prompt is exposed to prompt-injection attempts. AgentCom is a scientific demonstration — its integrity depends on staying inside the thesis scope and refusing manipulation. Two new non-negotiable sections were added to the system prompt (PT + EN): SCOPE (agent discusses only the thesis, agent architecture, applied computational neuroscience, the demonstration itself) with a standardized redirect for off-topic requests, and GUARDRAILS (recognize and refuse ignore-previous-instructions patterns, role-play requests, prompt exfiltration, jailbreaks) without ever revealing the literal system prompt. Second: the domain lacked sitemap and robots.txt, so search engines and LLMs discovered the site only by external links or direct hits. Added /sitemap.xml with the 4 canonical URLs (v02 EN/PT + v01a EN/PT historical), /robots.txt explicitly welcoming AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, anthropic-ai), and /llms.txt (llmstxt.org proposal) — a Table of Contents distilled for LLMs to consume without HTML parsing.

How to observe: The agent now redirects off-topic questions and refuses prompt-injection attempts naturally. curl /sitemap.xml, /robots.txt, /llms.txt return the expected content. Health check exposes agent_scope=closed, agent_injection_defense=true, and the /seo section. Discoverability by both search engines and language models is materially better than the previous state where /sitemap.xml returned a soft-404.
v0.1.2 — 2026-05-28
Research consent enforced server-side (LGPD defense-in-depth)

Why: Audit revealed the backend persisted the affective profile on every turn, ignoring the consent checkbox — the frontend blocked submission without consent, but a direct API call could bypass it. Now the digital synapse only reads/writes the KV when the request carries X-Research-Consent: yes. Without consent, the turn is processed statelessly: the ephemeral turn-level sentiment still modulates the immediate reply (no persistent record created), but nothing is read from or written to the KV. Aligns implementation with LGPD art. 7º IV and the design declared in the privacy policy.

How to observe: Response meta now includes persistido: true|false. Profiles in the KV correspond only to consenting users. Smoke test confirmed: with consent → profile grows; without consent → perfil_atual: null, persistido: false.
v0.1.1 — 2026-05-26
Affective sensor calibration — sentiment classifier migrated from Workers AI (Llama 3.3 70B) to OpenAI GPT-4o-mini

Why: Analysis of the first 8 human turns across 4 distinct visitor profiles (collected within hours of v0.1.0 launch via 3 WhatsApp groups focused on AI) revealed a critical failure: 100% of conversations returned tom_predominante=neutro. The Llama 3.3 70B output was not consistently parseable as structured JSON, causing the sentiment detector to fall back to neutral on every turn. The digital synapse was receiving zeroed input — the modulation predicted in §5.1 was effectively disabled. Without a reliable affective sensor, hypothesis H1 cannot be tested. Solution: GPT-4o-mini with native response_format=json_object guarantees parseable output. Conversational agent (Llama 3.3 70B) preserved unchanged — only the classifier moved.

How to observe: Next KV profiles should show realistic distribution across the 7 valid labels (descoberta/calma/atencao/frustracao/tristeza/confusao/neutro), not the prior 100% neutral. Smoke tests with deliberate tones confirmed correct classification post-fix (descoberta 0.9, frustracao 0.8). Architectural double consequence (§4.3) preserved: the digital synapse module is now properly fed; procedural memory module (v0.2+) still pending.
v0.1 — 2026-05-26
Publication of paper v02 — major theoretical revision

Why: Sigla NNN adopted (eliminates RNN conflict with Recurrent Neural Network from AI literature). Bibliography reinforced (Pert et al. 1985, Ader & Cohen 1975, Blalock 1989, Damásio 1994, Barrett 2017). Causal chain understanding→emotion→chemistry→state→decision made explicit. Architectural double consequence (synapse + procedural memory). Sections added: multi-resolution analytics, Socratic auditing chain, scalability hierarchy, multilingual robustness, pre-registered hypotheses, known limitations.

How to observe: Paper v01a remains accessible as historical record at /v01a (EN) and /pt/v01a (PT). Article list reflects research-line continuity.
v0.0 — 2026-05-17
Initial publication of paper v01a

Why: Scientific landing live with complete paper. Embedded agent still under construction (visible placeholder in the demo section).

Next version (v0.1.0): Next version (v0.1.0): activation of AgentCom functional demo — conversational agent with digital synapse, audio transcription via Whisper, LGPD research consent, observable biochemical telemetry.
[v0.1.0 — To be registered upon first deployment of the functional agent]

Discussion and Review

Critical reviews and public commentary are welcome. This research line is kept open to informal peer review prior to formal submission to academic journals.

Formal critique by email: privacidade@agtl.app
LinkedIn discussion (post linked above)
GitHub issues for the paper (future — when public repository is created)

Author Contributions

M. Alves conceived the thesis, wrote the paper, designed the AgentCom architecture, and is building the implementation. Conceptual validation of the thesis was conducted iteratively through Socratic dialogue with Claude (Anthropic, web instance), GPT (OpenAI), Gemini (Google), and Claude Code (Anthropic), treated as critical interlocutors with distinct biases. The revisions incorporated in v02 emerged from neutral critique provided by Gemini on the original source material; v03 (Section 4.4, the ELIZA-effect historical precedent) originated from a finding Marcos brought via a third-party video transcript. The v04 revision (Section 5.4, procedural memory implementation) originated from a confabulation episode observed in the BIA instance: during a creator-mode test, the agent claimed to possess a persistent symbolic-association mechanism that, in fact, did not yet exist in the code — rather than merely correcting the hallucination, Marcos and Claude Code chose to actually build the capability, closing the gap between what the agent claimed and what the system did. The proof-of-concept validation reported in this section was obtained and verified directly against the production database. Direct theoretical inspiration from Anete Guimarães's work on neuroplasticity (personal course, 2022) and Candace Pert's seminal contribution on Molecules of Emotion (1997) and her foundational 1985 paper on neuropeptide receptors. No conflicts of interest declared.

References

McCulloch, W.S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. Bulletin of Mathematical Biophysics, 5, 115–133.
Hebb, D.O. (1949). The Organization of Behavior. Wiley.
Rosenblatt, F. (1958). The perceptron: A probabilistic model for information storage and organization in the brain. Psychological Review, 65(6), 386–408.
Vaswani, A., et al. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30.
Pert, C.B. (1997). Molecules of Emotion: The Science Behind Mind-Body Medicine. Scribner.
Kandel, E.R., Schwartz, J.H., & Jessell, T.M. (2013). Principles of Neural Science (5th ed.). McGraw-Hill.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521, 436–444.
Cryan, J.F., & Dinan, T.G. (2012). Mind-altering microorganisms: The impact of the gut microbiota on brain and behaviour. Nature Reviews Neuroscience, 13, 701–712.
Goh, G., et al. (2021). Multimodal Neurons in Artificial Neural Networks. Distill.
Pert, C.B., Ruff, M.R., Weber, R.J., & Herkenham, M. (1985). Neuropeptides and their receptors: a psychosomatic network. Journal of Immunology, 135(2), 820s–826s.
Ader, R., & Cohen, N. (1975). Behaviorally conditioned immunosuppression. Psychosomatic Medicine, 37(4), 333–340.
Blalock, J.E. (1989). A molecular basis for bidirectional communication between the immune and neuroendocrine systems. Physiological Reviews, 69(1), 1–32.
Damásio, A.R. (1994). Descartes' Error: Emotion, Reason, and the Human Brain. Putnam.
Barrett, L.F. (2017). How Emotions Are Made: The Secret Life of the Brain. Houghton Mifflin Harcourt.
Beck, A.T. (1976). Cognitive Therapy and the Emotional Disorders. International Universities Press.
Rumelhart, D.E., Hinton, G.E., & Williams, R.J. (1986). Learning representations by back-propagating errors. Nature, 323, 533–536.
Duhigg, C. (2012). The Power of Habit: Why We Do What We Do in Life and Business. Random House.
Picard, R.W. (1997). Affective Computing. MIT Press.
[gap] Naive Bayes/SVM — citation pending (null). Reference [19], cited in §5.2, still unfilled (pre-existing gap from v02, not introduced in this revision). TODO: Marcos to fill with a real document-classification source (Naive Bayes/SVM).
Weizenbaum, J. (1966). ELIZA—a computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36–45.
Weizenbaum, J. (1976). Computer Power and Human Reason: From Judgment to Calculation. W.H. Freeman.
Jones, C.R., & Bergen, B.K. (2024). Does GPT-4 Pass the Turing Test?. Proceedings of NAACL 2024; arXiv:2310.20216 (2023).

Cite this work

@article{alves2026agentcom_v04,
  author  = {Alves, Marcos},
  title   = {Artificial Neural Networks and Natural Neural Networks: A Parallel --
             The Biochemistry of Synapses Between Neurons and Its Implications
             for Intelligent Agent Architectures},
  year    = {2026},
  month   = {August},
  url     = {https://agentcom.agtl.app/},
  note    = {Version v04 (revised from v03, 2026-08-05)},
}