Salvato in:
| Autore principale: | |
|---|---|
| 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 |