Artur Ziganshin

The relationship between symbolic and connectionist approaches to artificial intelligence has been a foundational tension in the field for decades. This paper proposes a hybrid account where symbols can emerge from structured transformations over distributed representations in contemporary AI systems. I develop operational tests for symbolic competence and evaluate evidence from large language models, retrieval-augmented generation, and neuro-symbolic architectures. The framework distinguishes between functional symbolic behavior and genuine symbolic representation while providing criteria for assessing when AI systems exhibit symbolic competence. Case studies in compositional generalization, systematic reasoning, and grounding problems demonstrate how hybrid approaches can advance our understanding of language, meaning, and representation in machines.

Artificial intelligence is increasingly embedded in public services where decisions implicate rights, benefits, and burdens. This article proposes a practical lifecycle for responsible public‑sector AI spanning procurement, pilot, monitoring, and redress, with ethical checkpoints and governance artifacts at each phase. Using concrete examples and visuals, we distill lessons from applications such as predictive policing, welfare eligibility, and school placement. We define metrics that go beyond accuracy to include equity, participation, service quality, trust, and cost‑benefit, and we outline implementation guidance aligned to Technology in Society criteria. We conclude with recommendations for dignified, accountable digital public services.

The accelerating deployment of AI and automation across sectors raises fundamental questions about human dignity in an age of machine capability. This paper develops a dignity-centered framework for AI and automation grounded in the capabilities approach, contributive justice, and principles of fair transition. I propose concrete policy tests and dignity indices, applying them to three domains: elder care, education, and logistics. The framework shows how automation can expand rather than diminish human capabilities and contributive opportunities when designed with dignity preservation as a core objective. Through analysis of economic displacement, deskilling, and social devaluation, I demonstrate how institutional support and design choices can ensure dignified transitions that preserve human agency and social participation.

Large language model evaluation has become dominated by single-number leaderboards that rank models using aggregate scores across diverse tasks. While these leaderboards provide useful high-level comparisons, they obscure critical details about model behavior, capabilities, and limitations that matter for responsible deployment. This paper critiques current LLM benchmarking practices and proposes a framework for comparative analysis built on three principles: parity of information (standardized evaluation conditions), uncertainty and risk reporting (confidence intervals and safety metrics), and benchmark cards (comprehensive metadata about datasets, tasks, and protocols). Through case studies in multidomain evaluation, I demonstrate how transparent, multidimensional comparisons can support responsible model development and deployment while enabling fairer assessment of competing approaches. The framework provides both theoretical foundations and practical tools for moving beyond reductive metrics toward evaluation practices that capture the dimensions of model performance most relevant for beneficial real-world deployment.

Current large language model (LLM) infrastructure optimization focuses narrowly on provider costs while ignoring substantial social and environmental externalities. This paper develops a framework for responsible cost curves that internalizes environmental, labor, and equity considerations into LLM serving optimization. I formalize total-cost trade-offs, provide implementation guidelines, and demonstrate the approach with a multilingual deployment case study. The framework shows how infrastructure decisions can minimize not only provider costs but also social and environmental externalities while ensuring fair distribution of AI benefits. Applications to energy optimization, labor conditions, and access equity demonstrate practical approaches to responsible infrastructure economics that can guide both industry practice and policy development.

Artificial intelligence systems increasingly function as epistemic infrastructure, mediating how individuals and institutions access, evaluate, and act upon knowledge. This paper develops a comprehensive typology of epistemic risks posed by contemporary AI systems, organized around seven core categories: hallucination, error amplification, spurious coherence, authority drift, opaque provenance, filter bubbles, and miscommunicated uncertainty. I argue that addressing these risks requires recognizing three epistemic duties for AI system designers: provenance duty, uncertainty disclosure duty, and contestability duty. The paper presents concrete technical and governance mechanisms for implementing these duties, including retrieval-augmented generation, calibration interfaces, metadata standards, feedback loops, and epistemic audits. Case studies in medical advice, legal research, and news recommendation demonstrate how these mechanisms can safeguard epistemic integrity while preserving AI's beneficial capabilities.

Protective interface "guardrails" are increasingly used in digital systems to prevent user harm, but their ethical foundations remain underexamined. This paper develops a normative framework for UX guardrails built on four principles: transparency, proportionality, reversibility, and contestability. I distinguish between soft constraints (warnings, friction) and hard constraints (prevention, blocking) while providing operational criteria for each principle. The framework includes a design-and-audit checklist, impact metrics, and governance practices that reduce harm without undermining user autonomy. Applications to digital health platforms, retail investing apps, and creative AI tools demonstrate how protective design can preserve agency while enabling safety.

Current AI agent evaluation overwhelmingly relies on win-rates and aggregate performance scores that obscure crucial dimensions of agent behavior. This reductive approach poses significant risks when agents operate in complex, real-world environments where multiple objectives matter simultaneously. This paper introduces a multidimensional evaluation paradigm that assesses agent performance across eight critical dimensions: calibration, robustness, goal alignment, uncertainty handling, fairness, interpretability, adaptability, and resource efficiency. I propose Agent Evaluation Cards—standardized reporting templates that make agent capabilities and limitations transparent to developers, users, and stakeholders. Case studies in navigation, healthcare triage, and multi-agent games demonstrate how comprehensive evaluation can support responsible agent deployment while enabling beneficial innovation.

As artificial intelligence systems become increasingly autonomous and capable of making complex decisions with minimal human oversight, fundamental questions about agency and moral responsibility demand urgent philosophical examination. This paper argues that machine agency is derivative rather than inherent—AI systems exhibit functional agency through their capacity to act autonomously, but lack the moral agency necessary for direct responsibility attribution. Instead, moral responsibility must be distributed across the socio-technical networks that design, deploy, and oversee these systems. We propose a framework for responsibility allocation based on control and epistemic access, implemented through institutional mechanisms including traceability systems, auditability protocols, responsibility ledgers, and escalation frameworks. This approach preserves human agency while enabling the beneficial deployment of autonomous AI systems in complex social contexts.