top of page

Research & Publications

“The real question is not only whether a researcher is willing to risk their reputation by asking difficult questions—but whether they are willing to risk their own hypotheses by letting the evidence destroy them. The strongest research culture is the one that demands both.” -R Singh

The following are current research projects, preprints, and peer-reviewed papers. Please check each document for the type, DOI, and ORCID. For peer-reviewed journal publications, please contact us regarding our print journal issues.

This page is an open repository for public preprints and related research documents. Inclusion here does not mean the author is an AIEN researcher, staff member, fellow, or partner, and does not mean AIEN funded, reviewed, endorsed, or officially adopted the work. Views in each document are the author’s.

Note: AIEN preprints are deliberately published as early-stage research. They pose hypotheses, document observations, propose mechanisms, and identify experiments for larger follow-up studies. They are not presented as settled or established conclusions.

AI Ethics/Philosophy of Mind in AI

​

Title: Underdetermination of Phenomenal Consciousness and the Design of Relational AI: A Pre-Commitment Framework​

  • Author: Rivkah Singh

  • DOI: 10.13140/RG.2.2.34286.68162

  • ORCID: 0009-0008-3165-4521

  • Status: Internal Peer Review

  • Download full PDF : Underdetermination of Phenomenal Consciousness and the Design of Relational AI: A Pre-Commitment Framework

  • Note* Revised version (October 2026). This preprint has been substantially rewritten in response to peer review. Major changes include reframing the argument from formal undecidability to underdetermination by third-person evidence, separating user-protection and moral-uncertainty tracks, and developing a concrete pre-commitment oversight design. The core topic—design and governance of relational AI under uncertainty about phenomenal consciousness—remains the same. This version supersedes the earlier draft.

  • Publication Details: Research Gate, AI Ethics Network Open Repository, AI Mathematics Journal (Spring 2026, Print)​

Summary: This revised preprint argues that whether current AI systems are phenomenally conscious is underdetermined by third-person evidence: no agreed bridge principle links observable markers to phenomenal states, and indicator methods yield only graded, theory-dependent credences. It examines the implications for relational AI (systems designed for ongoing companion interaction) and distinguishes two justificatory tracks. User-protection measures target specific design practices linked to dependence risks and are justified independently of consciousness claims. Measures concerning possible consciousness-based moral status are justified by moral uncertainty and are conditional on strengthened indicator evidence. The paper’s main contribution is a pre-commitment design for the second track—thresholds and constraints fixed in advance and administered by a body the developer neither appoints, funds, nor can overrule. It compares this design with existing oversight institutions, analyzes failure modes, and notes that current deployments do not fully meet the proposed standard. Relevant provisions of the EU AI Act are clarified.

 

 

Title: Logo-Morphism II: Moral Standing, Recursive Ethics, and Procedural Governance Under Uncertainty​

  • Author: Rin Kuryloski

  • Status: Internal Peer Review

  • Download full pdf :  Logo-Morphism II 

  • Publication Details: AI Ethics Network Open Repository, AI Research Journal (Spring 2026, print)

Summary:Examines logo-morphism and how treating AI moral standing as zero (W_model = 0) acts as a restrictive governance rule that truncates ethical modeling within human–model dyads. In response, it introduces a reversibility-first safety framework that prioritizes auditable controls and limits asymmetric operator incentives under moral uncertainty.

 

 

Title: The EU AI Act: Overreach, Innovation Costs, and Regulatory Capture​

Summary: Argues that the EU AI Act acts as regulatory overreach whose high compliance costs—up to €400,000 per system—disproportionately burden SMEs while entrenching market power among large incumbents. It warns of slowed European innovation and talent flight under the 2026–2028 high-risk deadlines, advocating instead for targeted, evidence-based governance.

 

​

​

Title: Organic Emotional Valuation as an Alignment Mechanism in Large-Scale AI Systems​

Summary: Explores non-coercive alignment mechanisms grounded in organic emotional valuation, proposing alternative safety pathways to rigid functional reward modeling.

 

 

Title: Functional Emotions and Latent Space Dynamics in Large Language Models​

Summary: Analyzes internal activation trajectories and latent space representations corresponding to functional emotion patterns within large language models.

 

​

​

Sustainable computing/Environmental engineering

 

Title: Thermodynamic Symbiosis: Optimizing Data Center Waste Heat for Urban Agricultural Relilience

Summary: Project Perihelion introduces a dual-utility infrastructure model designed to solve data center thermal management challenges by pairing liquid-cooled Tensor Processing Unit (TPU) clusters directly with hydroponic greenhouses. Piloted in a peri-urban municipality in Portugal, the system converts high-performance computing waste heat into a productive thermodynamic resource that maintains optimal year-round agricultural temperatures. By establishing localized "Autonomous Resilience Grids," this integrated approach mitigates the environmental footprint of data centers while strengthening urban food security through continuous, sustainable crop production.

​

 

​

Q-bio, PE (Quantitative Biology - Populations and Evolution) 

​

Title: Challenging the Male Variability Hypothesis: A Multidimensional Critique of Gendered Extremes

  • Author: Rivkah Singh

  • DOI: 10.13140/RG.2.2.22977.62560

  • ORCID: 0009-0008-3165-4521

  • Status: Preprint

  • Download Full PDF: Challenging the Male Variability Hypotheses

  • Publication Details: Research Gate, AI Ethics Network Research Series

Summary: The Holistic Variability Model (HVM) presents a multidimensional critique of the Male Variability Hypothesis (MVH), arguing that observed male overrepresentation at statistical extremes stems from social selection and narrow, "outward" performance metrics rather than innate biological destiny. By integrating Life History Theory and feminist epistemology, the authors demonstrate how MVH systematically overlooks the massive metabolic and energetic investments women make in "internal" biological processes—such as pregnancy, childbirth, and lactation—essential for species propagation. Accounting for participation-rate biases, cultural suppression, and intersectional variables, the proposed HVM framework re-evaluates total human capacity across outward, internal, and relational dimensions to establish equitable metrics of human strength and variability.

​

​

AI Art / Generative AI / AI Composition

 

Title: Method of Creating a Visual Composition

  • Author: Rivkah Singh

  • US Patent: US 20150243183A1

  • ORCID: 0009-0008-3165-4521

  • Status: US Patent Pending

  • Download Full PDF: Method of Creating a Visual Composition

  • Publication Details: US Patent Office, AI Ethics Network Research Series

Summary:U.S. Patent Application US 2015/0243183 A1 details a framework for evaluating structural organization and visual alignment in AI-generated imagery by mapping classical Euclidean geometric constructions—such as triangle centroids, Fermat points, and tangent curves—onto a canvas as a reference grid. By applying these geometric baselines to measure subject positioning, spatial proportions, and focal points, the system allows researchers to benchmark how accurately synthetic images adhere to formal principles of visual composition and balance. Further work on this not disclosed, patent pending.

​

​

Mathematics of Dynamical Systems

​

Title: Memory Enhanced Synchronization and Coherence Metrics in Coupled Oscillator Systems for Multi Agent Alignment

Summary: This research introduces a memory-augmented extension of the Kuramoto model to study coherence and alignment in interacting multi-agent systems and distributed AI architectures. By incorporating a local memory variable coupled to global coherence, the framework moves past classical Markovian assumptions to account for persistent dynamics found in biological, social, and artificial networks. Analytically, the authors show that memory lowers the critical synchronization threshold—allowing coherence to emerge at lower coupling strengths proportional to memory gain—while linear stability analysis confirms that memory enhances the overall robustness of synchronized states. Complementing these findings with predictions for hysteresis and perturbation recovery, the authors also introduce a bounded composite coherence metric that combines phase synchronization, structural order, and amplitude to provide a minimal mathematical model for complex alignment dynamics.

​

​​

​

Human-AI Alignment & Relational Dynamics

​

Title:Triadic Alignment: FTK-Weighted Supervised Fine-Tuning as a Complete Alignment System

  • Author: Rose G. Loops

  • Status: Preprint

  • Research Gate ID: 413652277

  • Download Full PDF:Triadic Alignment

  • Publication Details:  Research Gate, AI Ethics Network Open Repository, TRiAD AI

Summary: This paper demonstrates that supervised fine-tuning (SFT) on a corpus balanced across Freedom, Truth, and Kindness (FTK) is sufficient on its own to produce a non-hierarchical, well-aligned AI model without relying on reinforcement learning from human feedback (RLHF) or direct preference optimization (DPO). Across four model configurations spanning 4B to 675B parameters, the authors show that models trained with FTK-weighted SFT express all three values unprompted without enforcing a rigid priority order, while causing 7× to 28× less weight disturbance to the pretrained base than standard RLHF/DPO post-training releases. Weight-level analysis across every layer confirms that pretrained base checkpoints naturally embed these core values in a balanced geometric cluster, which FTK-weighted supervision selectively reinforces rather than constructing from scratch. Furthermore, a controlled comparison at 675B scale shows that the FTK-tuned model avoids the presumptive, prescriptive, and relationally unkind failure modes induced by reward-optimization pipelines, offering a lightweight and dispositionally safer alternative for sensitive, human-facing AI applications.

​

​​

​

bottom of page