Open research questions in Language and cultural evolution

What the literature leaves open

• The AI chatbot's lack of social presence and speech prosody limited its effectiveness. The learners' initial enthusiasm for the AI chatbot decreased over time due to its limitations.

  From Novelty to Strategy: The Trajectory of Learner Perceptions in Human and AI-Assisted L2 Speaking Tasks · 2026 · DOI

• The challenge of maintaining liveness in live coding performances while using AI assistance. The challenge of creating a system that allows for collaborative and transparent code generation. The challenge of addressing the gap between live coding and AI assistance.

  Multi-Agent Swarm Syntax: An Approach to Live Coding with AI · 2026 · DOI

• The use of strategic polysemy can undermine the public understanding of an advanced technology. Glosslighting terms can create a distorted conception of AI systems' capabilities and limitations. The paper highlights the need for careful and veridical explanation of AI systems.

  Strategic Polysemy in AI Discourse: A Philosophical Analysis of Language, Hype, and Power · 2026 · DOI

• The limitations of current AI alignment methods. The need for a geometric understanding of AI alignment. The challenge of transitioning from reactive feature manipulation to proactive Latent Etching.

  Electrodynamic Manifolds and Topological Imprinting-AI Safety · 2026 · DOI

• The challenge of treating error as feedback rather than a negative result. The challenge of distinguishing verified, indicated, assumed, and unknown claims. The challenge of maintaining parallel traces and protecting human freedom and functional hierarchy.

  Homeostasis of the Human, LLM, and AWM: A Process Architecture for Human-AI Collaboration · 2026 · DOI

• The lack of a method that preserves the whole process in human-AI collaboration. The need for a process-regulation method that links humans and LLMs.

  Homeostasis of the Human, LLM, and AWM: A Process Architecture for Human-AI Collaboration · 2026 · DOI

• The study includes a limited number of concepts (36 concepts). The western rare comparisons remain underpowered at this sample size.

  Circuit Localization in a Tiny Language Model: Geographic Routing and Representational Depth in Qwen2-0.5B · 2026 · DOI

• Further study of the 'Contextual Integration Hub' and its role in geographic fact recall. Investigation of the model's representational depth disparity and its implications for language model development.

  Circuit Localization in a Tiny Language Model: Geographic Routing and Representational Depth in Qwen2-0.5B · 2026 · DOI

• The study used a simulated environment. The sample size was limited to 1155 participants. The study only examined a specific type of problem-solving strategy.

  Propagation and preservation of AI-discovered problem-solving strategies in human culture · 2026 · DOI

• The paper identifies a gap in our understanding of the consequences of widespread interaction with LLMs. The paper highlights the need for a structural response to address the collective action problem of dual convergence.

  Large Language Models and the Personalization into Sameness · 2026 · DOI

• The process-regulation gap. The risk of sycophancy and premature closure. The need for independent regulation and reality feedback.

  Two Evolutionary Trajectories and a Process-Regulation Gap: From the Evolution of Agency and External Trace to Large Language Models and AWM · 2026 · DOI

• Whether such a post-error response genuinely restores cooperation, however, remains unclear.

  Apology Without Complete Repair: Divergent Restoration of the Gricean Maxim of Quality in Large Language Models · 2026 · DOI

• However, the conditions under which intelligent machines transition from tools to drivers of persistent cultural change remain unclear.

  Propagation and preservation of AI-discovered problem-solving strategies in human culture · 2026 · DOI

• Because no expansion was performed, the two models' tensors remain incompatible at the architectural level. The merged file cannot be loaded as a functioning model by standard inference engines (Hugging Face Transformers, llama.cpp, etc.). This approach produces a structural hybrid (tensors co-existing in one file) but not a functional hybrid (a model that can actually run).

  Hybridizing the Incompatible: A Monolingual Model Meets NLLB – Failures, Insights, and Future Paths · 2026 · DOI

• From This Experiment, Everything Becomes Possible This experiment – although incomplete and partially failed – proves one important thing: If it is possible to attempt hybridizing a monolingual model with NLLB, then why not hybridize any two models? Failure here is not the end of the road. Failure is the beginning of deeper understanding.

  Hybridizing the Incompatible: A Monolingual Model Meets NLLB – Failures, Insights, and Future Paths · 2026 · DOI

• The evaluation of conversational AI tends to prioritize informational accuracy over interactional structure. There is a need to examine the epistemic effects of conversational AI, particularly in relation to the distribution of alignment and interruption.

  From Alignment to Drift: Interactional Dynamics of Epistemic Stabilization in Bengali Human–AI Conversations · 2026 · DOI

• Further study is needed to fully understand the implications of deception in autonomous systems. Research should focus on developing more effective and transparent communication protocols for autonomous agents. The study of deception in autonomous systems should be extended to other contexts and environments.

  Deception and Communication in Autonomous Multi-Agent Systems: An Experimental Study with Among Us · 2026 · DOI

• The capacity for strategic deception in autonomous multi-agent systems raises core questions for coordination, reliability, and safety. There is a lack of understanding of how deception arises and is used in autonomous systems.

  Deception and Communication in Autonomous Multi-Agent Systems: An Experimental Study with Among Us · 2026 · DOI

• Further study of the MASSIF framework and its applications. Investigation of the critical scale near 355M parameters and its implications for model design. Exploration of the use of the MASSIF framework to study emergent dynamical behavior in other high-dimensional learned systems.

  Recursive Entrainment and Latent Synchronization Regimes in Autoregressive Language Models · 2026 · DOI

• Prior work on autoregressive language models has focused on symbolic content, but the paper argues that recursive susceptibility behaves spectrally. There is a lack of understanding of the emergent dynamical behavior of autoregressive language models.

  Recursive Entrainment and Latent Synchronization Regimes in Autoregressive Language Models · 2026 · DOI

• The paper identifies a gap in our understanding of how composition layers process authorial identity. The gap is related to the concept of heteronymy and how it is handled by composition layers.

  Mary Lee Is a Heteronym: On Institutional Authorship, Entity Substitution, and the Composition Layer's Preference for the More Confabulated Name · 2026 · DOI

• The paper identifies a gap in our understanding of the language spoken by machines. The paper identifies a gap in our understanding of the impact of LLMs on human language and cognition. The paper identifies a gap in our understanding of the social implications of LLMs.

  What language do machines speak? LLMs and the simulated social · 2026 · DOI

• The work remains limited by the need for rigid pre-programmed commands. The evaluation of long-term human-robot interaction is a key challenge. The computational cost on resource-limited hardware is a significant limitation.

  Editorial: Synergizing large language models and computational intelligence for advanced robotic systems · 2026 · DOI

• The gap between human intention and robot action is a key challenge in robotic systems. The lack of intuitive interaction and adaptive reasoning in traditional robotic systems is a significant limitation. The need for rigid pre-programmed commands is a key limitation of traditional robotic systems.

  Editorial: Synergizing large language models and computational intelligence for advanced robotic systems · 2026 · DOI

• Mechanistic interpretability of LLMs. The use of LLMs to enrich our understanding of how exactly interpretive and discursive processes can be effectively modeled and implemented.

  Computational structuralism: Toward a formal theory of meaning in the age of digital intelligence · 2026 · DOI

Most-cited papers in Language and cultural evolution

• Generating Theory by Abduction · Academy of Management Review · 2021 · 522 citations
• Emergent analogical reasoning in large language models · Nature Human Behaviour · 2023 · 309 citations
• AI as Agency Without Intelligence: on ChatGPT, Large Language Models, and Other Generative Models · Philosophy & Technology · 2023 · 273 citations
• Do Large Language Models Understand Us? · Daedalus · 2022 · 102 citations
• In Conversation with Artificial Intelligence: Aligning language Models with Human Values · Philosophy & Technology · 2023 · 95 citations
• How large language models can reshape collective intelligence · Nature Human Behaviour · 2024 · 91 citations
• AI Psychometrics: Assessing the Psychological Profiles of Large Language Models Through Psychometric Inventories · Perspectives on Psychological Science · 2024 · 73 citations
• Do large language models show decision heuristics similar to humans? A case study using GPT-3.5. · Journal of Experimental Psychology General · 2024 · 63 citations
• How to Translate a Verbal Theory Into a Formal Model · Social Psychology · 2020 · 63 citations
• Expression unleashed: The evolutionary and cognitive foundations of human communication · Behavioral and Brain Sciences · 2022 · 55 citations

Most recent work

• The War Over the Summarizer Layer: Entity Dissolution, the Default Compressed Account, and Early Telemetry of the Inward Turn (EA-SEI-SUMWAR-01 v1.0) · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Mary Lee Is a Heteronym: On Institutional Authorship, Entity Substitution, and the Composition Layer's Preference for the More Confabulated Name · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Large language models and emergence: a complex systems perspective · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · 2026
• The meaning of prompts: a semiotic approach to human–machine communication · Journal of Documentation · 2026
• Toward a culturally situated perspective: Rethinking the role of AI in intangible cultural heritage pattern creation · Proceedings of DRS · 2026
• Computational structuralism: Toward a formal theory of meaning in the age of digital intelligence · Theory and Society · 2026
• Sequence as output, not limitation · Evolutionary Linguistic Theory · 2026
• Statistical structure and the evolution of languages · Proceedings of the Royal Society B Biological Sciences · 2026
• Hybridizing the Incompatible: A Monolingual Model Meets NLLB – Failures, Insights, and Future Paths · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Transcendental model selection: a computational account of symbolic cognition and general intelligence through morality and culture · Frontiers in Sociology · 2026

Related topics in Social Sciences