Salvato in:
Dettagli Bibliografici
Autore principale: Senyi, Frank
Natura: Recurso digital
Lingua:inglese
Pubblicazione: Zenodo 2025
Soggetti:
Accesso online:https://doi.org/10.5281/zenodo.17626590
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901932908478464
author Senyi, Frank
author_facet Senyi, Frank
contents <p>The Reflective Genesis Hypothesis (RGH) proposes that reflective feedback—systems that <br>model both the world and themselves across time—is a structurally special ingredient in complex <br>adaptive behavior and, potentially, consciousness-like processes. This proposal outlines a fourexperiment program using artificial agents to test a core architectural implication of RGH: that <br>explicit reflective self-models confer distinctive functional advantages over non-reflective <br>architectures.<br>Experiment 1 compares reflective and non-reflective agents in single-agent environments <br>requiring self-commitment and resistance to temptation. Experiment 2 extends the comparison to <br>multi-agent settings, testing whether reflective populations exhibit more stable, coherent social <br>organization. Experiment 3 examines temporally entangled decision problems (Newcomb-like <br>scenarios, precommitment, and two-boundary tasks), probing whether time-extended self-models <br>improve cross-time policy coherence. Experiment 4 evaluates introspective self-report, asking <br>whether reflective agents generate more accurate, calibrated, and stable reports about their own <br>internal states than non-reflective baselines.<br>All experiments are designed to be implementable with standard deep reinforcement learning and <br>simulation tools, under tightly controlled capacity and training budgets. Together, these studies <br>do not attempt to confirm or refute RGH as a cosmological theory, but they do test a backbone <br>claim: that reflective architectures are functionally distinctive across individual, social, temporal, <br>and introspective dimensions</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17626590
institution Zenodo
language eng
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Reflective Architectures in Artificial Agents: A Four-Experiment Program to Probe the Reflective Genesis Hypothesis
Senyi, Frank
Reflective Feedback
Self modeling
Reinforcement learning
Multi-agent systems
Temporal decision-making
Reflective Genesis Hypothesis
RGH
<p>The Reflective Genesis Hypothesis (RGH) proposes that reflective feedback—systems that <br>model both the world and themselves across time—is a structurally special ingredient in complex <br>adaptive behavior and, potentially, consciousness-like processes. This proposal outlines a fourexperiment program using artificial agents to test a core architectural implication of RGH: that <br>explicit reflective self-models confer distinctive functional advantages over non-reflective <br>architectures.<br>Experiment 1 compares reflective and non-reflective agents in single-agent environments <br>requiring self-commitment and resistance to temptation. Experiment 2 extends the comparison to <br>multi-agent settings, testing whether reflective populations exhibit more stable, coherent social <br>organization. Experiment 3 examines temporally entangled decision problems (Newcomb-like <br>scenarios, precommitment, and two-boundary tasks), probing whether time-extended self-models <br>improve cross-time policy coherence. Experiment 4 evaluates introspective self-report, asking <br>whether reflective agents generate more accurate, calibrated, and stable reports about their own <br>internal states than non-reflective baselines.<br>All experiments are designed to be implementable with standard deep reinforcement learning and <br>simulation tools, under tightly controlled capacity and training budgets. Together, these studies <br>do not attempt to confirm or refute RGH as a cosmological theory, but they do test a backbone <br>claim: that reflective architectures are functionally distinctive across individual, social, temporal, <br>and introspective dimensions</p>
title Reflective Architectures in Artificial Agents: A Four-Experiment Program to Probe the Reflective Genesis Hypothesis
topic Reflective Feedback
Self modeling
Reinforcement learning
Multi-agent systems
Temporal decision-making
Reflective Genesis Hypothesis
RGH
url https://doi.org/10.5281/zenodo.17626590