[ai] Daily AI Research Digest — 2026-03-04

[jfomhover]

This digest covers research published on 2026-03-04. I analyzed 369 entries across three feeds (arxiv-ai: 303, arxiv-hc: 44, arxiv-cy: 22).

High Relevance (4)

• Architecting Trust in Artificial Epistemic Agents

  • Framework — Proposes a governance framework centred on three mechanisms: epistemic competence of AI agents, robust falsifiability, and epistemically virtuous behaviour. Argues that LLMs as epistemic agents create new informational interdependencies that risk cognitive deskilling and epistemic drift if poorly aligned, and calls for "knowledge sanctuaries" to protect human epistemic resilience.
  • Related to: AI Impact on Human Cognition and Psychology, AI, Identity, and the Digital Self

• The Geometry of Learning Under AI Delegation

  • Framework — Uses coupled dynamical systems modelling to show that adaptive AI delegation produces a stable low-skill equilibrium: AI assistance can strictly improve short-run performance while inducing persistent long-run skill loss, separated from the high-skill equilibrium by a sharp, effectively irreversible basin boundary. Identifies stability—not incentives or misalignment—as the central mechanism by which AI assistance can undermine human skill.
  • Related to: AI Impact on Human Cognition and Psychology

• Reassurance Robots: OCD in the Age of Generative AI

  • Empirical study — Qualitative analysis of 100 Reddit posts finds that GenAI (especially ChatGPT) is creating novel OCD obsessions and compulsions, with users engaging in repetitive AI queries as reassurance-seeking behaviour. Argues that current GenAI design harms individuals with OCD by becoming "Reassurance Robots" and calls for OCD-informed design of future GenAI systems.
  • Related to: AI Impact on Human Communication and Relationships, AI Impact on Human Cognition and Psychology

• NeuroWise: A Multi-Agent LLM "Glass-Box" System for Practicing Double-Empathy Communication with Autistic Partners

  • Empirical study — Between-subjects study (N=30) shows that an LLM coaching system for neurotypical users reduced deficit-based attributions about autistic partners (p=0.02) and improved conversation efficiency by 37% fewer turns (p=0.03), while control-group participants shifted toward increased deficit framing. Demonstrates AI-based interpretation can support attributional change by framing communication difficulties as mutual rather than one-sided.
  • Related to: AI Impact on Human Communication and Relationships

Medium Relevance (8)

• The Invisibility Hypothesis: Promises of AGI and the Future of the Global South

  • Position paper — Argues that as scientific discovery, economic coordination, and governance become automated, the relevance of human individuals in the Global South may become conditional on infrastructure access and geopolitical circumstances rather than skills or intelligence — rendering already marginalised populations not merely economically invisible but functionally irrelevant to global systems.
  • Related to: AI, Identity, and the Digital Self

• Policy Myopia as a Mechanism of Gradual Disempowerment in Post-AGI Governance, Circa 2049

  • Framework — Models three coupled mechanisms (salience capture, capacity cascade, value lock-in) through which policy myopia under advanced AI systematically removes humans from meaningful participation in resource allocation, formalising these as dynamical systems to show irreversible disempowerment can be the rational outcome of institutional optimisation.
  • Related to: AI Impact on Human Cognition and Psychology

• Delegation and Verification Under AI

  • Framework — Models workers' rational delegation and verification decisions under AI and shows that AI induces phase transitions: workers with strong verification reliability are amplified, while those with weaker verification over-delegate and suffer institutional quality degradation, even when baseline task success improves. Identifies a structural mechanism by which AI amplifies quality disparities between workers.
  • Related to: AI Impact on Human Cognition and Psychology

• AI4CAREER: Responsible AI for STEM Career Development at Scale in K-16 Education

  • Framework — Workshop paper examining how AI systems influencing feedback, advising, and opportunity access are becoming part of the developmental infrastructure that shapes career identity formation in K-16 students, with focus on appropriate AI roles in career decision-making, developmental alignment, and preventing structural inequity.
  • Related to: AI, Identity, and the Digital Self

• AI as We Describe It: How Large Language Models and Their Applications in Health are Represented Across Channels of Public Discourse

  • Empirical study — Analysis of five discourse channels (news, research press, YouTube, TikTok, Reddit) over two years finds AI health coverage is overwhelmingly positive and episodic, risk communication is shallow, and TikTok/Reddit exhibit higher anthropomorphism while attending little to risks — identifying public discourse as a diagnostic tool for literacy and governance gaps.
  • Related to: AI Impact on Human Communication and Relationships

• The Gen AI Generation: Student Views of Awareness, Preparedness, and Concern

  • Empirical study — Survey of 250+ students finds readiness to use GenAI is increasingly tied to curricular exposure, creating a new divide; while most students express enthusiasm, an even larger proportion voice concerns about ethics, job displacement, and adequacy of educational structures, signalling a complex psychological relationship with AI in education.
  • Related to: AI Impact on Human Cognition and Psychology

• Orality: A Semantic Canvas for Externalizing and Clarifying Thoughts with Speech

  • Technical system — Lab study (N=12) comparing an LLM-powered speech-to-node-diagram canvas (Orality) against plain ChatGPT interaction finds that Orality better supports users in clarifying and developing thoughts, pointing to distinct affordances of graphical versus conversational AI thought-externalisation tools and implications for cognitive augmentation design.
  • Related to: AI Impact on Human Cognition and Psychology, AI, Identity, and the Digital Self

• Changing Pedagogical Paradigms: Integrating Generative AI in Mathematics to Enhance Digital Literacy through 'Mathematical Battles with AI'

  • Case study — Describes a competitive format at ITMO University that deliberately uses an AI agent with elevated hallucination likelihood to train students in critical verification over blind reliance, reporting a significant shift in student mindset toward digital hygiene and prompt engineering skills.
  • Related to: AI Impact on Human Cognition and Psychology

Low Relevance (2)

• What Are You Doing? Effects of Intermediate Feedback from Agentic LLM In-Car Assistants During Multi-Step Processing

  • Empirical study — Mixed-methods study (N=45) finds intermediate progress feedback from in-car LLM assistants significantly improved perceived speed, trust, and user experience while reducing task load; user interviews reveal preference for initially high transparency that decreases as reliability is established.

• Shared (Mis)Understandings and the Governance of AI: A Thematic Analysis of the 2023-2024 Oversight of AI Hearings

  • Policy/Governance — Thematic analysis of US Senate AI oversight hearings finds technology-industry participants construct dominant narratives about AI's past, present, and future to argue for particular governance forms, examining the power arrangements upheld through these shared (and contested) framings.

---

Summary: Analyzed 369 total entries. Found 4 high-relevance, 8 medium-relevance, and 2 low-relevance papers related to our topics of interest.

Filter statistics

• Total entries scanned: 369
• Passed Phase 1 (title scan): ~28 (~8%)
• Phase 2: High 4 / Medium 8 / Low 2 / Excluded 14
• Top matched topics: AI Impact on Human Cognition and Psychology (9), AI Impact on Human Communication and Relationships (4), AI, Identity, and the Digital Self (4)