Anthropomimetic Uncertainty: What Verbalized Uncertainty in Language Models is Missing

2025/07/11 by Ulmer, Dennis, Lorson, Alexandra, Titov, Ivan +1 · 4 citations
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2507.10587

Abstract

Human users increasingly rely on natural language interactions with large language models (LLMs) in order to receive help on a large variety of tasks and problems. However, the trustworthiness and perceived legitimacy of LLMs is undermined by the fact that their output is frequently stated in very confident terms, even when its accuracy is questionable. Therefore, there is a need to signal the confidence of the language model to a user in order to reap the benefits of human-machine collaboration and mitigate potential harms. Verbalized uncertainty is the expression of confidence with linguistic means, an approach that integrates perfectly into language-based interfaces. Nevertheless, most recent research in natural language processing (NLP) overlooks the nuances surrounding human uncertainty communication and the data biases that influence machine uncertainty communication. We argue for anthropomimetic uncertainty, meaning that intuitive and trustworthy uncertainty communication requires a degree of linguistic authenticity and personalization to the user, which could be achieved by emulating human communication. We present a thorough overview over the research in human uncertainty communication, survey ongoing research, and perform additional analyses to demonstrate so-far overlooked biases in verbalized uncertainty. We conclude by pointing out unique factors in human-machine communication of uncertainty and deconstruct anthropomimetic uncertainty into future research directions for NLP.

Citations

• Teaching Language Models to Faithfully Express their Uncertainty
• Can Large Language Models Express Uncertainty Like Human?
• ConfTuner: Training Large Language Models to Express Their Confidence Verbally
• On the Robustness of Verbal Confidence of LLMs in Adversarial Attacks
• Humans overrely on overconfident language models, across languages
• Reconsidering LLM Uncertainty Estimation Methods in the Wild
• Revisiting Epistemic Markers in Confidence Estimation: Can Markers Accurately Reflect Large Language Models' Uncertainty?
• MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs
• Revisiting Uncertainty Estimation and Calibration of Large Language Models
• Position: Uncertainty Quantification Needs Reassessment for Large-language Model Agents
• Read Your Own Mind: Reasoning Helps Surface Self-Confidence Signals in LLMs
• Seeing is Believing, but How Much? A Comprehensive Analysis of Verbalized Calibration in Vision-Language Models
• Reasoning Models Better Express Their Confidence
• The AI Gap: How Socioeconomic Status Affects Language Technology Interactions
• Not Like Us, Hunty: Measuring Perceptions and Behavioral Effects of Minoritized Anthropomorphic Cues in LLMs
• Large Language Models are overconfident and amplify human bias
• The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers
• Object-Level Verbalized Confidence Calibration in Vision-Language Models via Semantic Perturbation
• Thinking Out Loud: Do Reasoning Models Know When They're Right?
• Know What You do Not Know: Verbalized Uncertainty Estimation Robustness on Corrupted Images in Vision-Language Models
• Uncertainty Distillation: Teaching Language Models to Express Semantic Confidence
• Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations
• Measuring What Makes You Unique: Difference-Aware User Modeling for Enhancing LLM Personalization
• GRACE: A Granular Benchmark for Evaluating Model Calibration against Human Calibration
• A Taxonomy of Linguistic Expressions That Contribute To Anthropomorphism of Language Technologies
• Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
• Influences on LLM Calibration: A Study of Response Agreement, Loss Functions, and Prompt Styles
• On Verbalized Confidence Scores for LLMs
• A Survey of Calibration Process for Black-Box LLMs
• A Survey on Uncertainty Quantification of Large Language Models: Taxonomy, Open Research Challenges, and Future Directions
• Flattering to Deceive: The Impact of Sycophantic Behavior on User Trust in Large Language Model
• Tulu 3: Pushing Frontiers in Open Language Model Post-Training
• On the Way to LLM Personalization: Learning to Remember User Conversations
• Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation
• GPT-4o System Card
• A Survey of Uncertainty Estimation in LLMs: Theory Meets Practice
• Do LLMs estimate uncertainty well in instruction-following?
• How Do Multilingual Language Models Remember Facts?
• Accounting for Sycophancy in Language Model Uncertainty Estimation
• Atomic Calibration of LLMs in Long-Form Generations
• Taming Overconfidence in LLMs: Reward Calibration in RLHF
• Calibrating Verbalized Probabilities for Large Language Models
• MAgICoRe: Multi-Agent, Iterative, Coarse-to-Fine Refinement for Reasoning
• Finetuning Language Models to Emit Linguistic Expressions of Uncertainty
• Programming Refusal with Conditional Activation Steering
• Are Large Language Models More Honest in Their Probabilistic or Verbalized Confidence?
• The Llama 3 Herd of Models
• PersonaGym: Evaluating Persona Agents and LLMs
• Rel-A.I.: An Interaction-Centered Approach To Measuring Human-LM Reliance
• The Art of Saying No: Contextual Noncompliance in Language Models
• LLMs instead of Human Judges? A Large Scale Empirical Study across 20 NLP Evaluation Tasks
• Can LLM be a Personalized Judge?
• External Invariants: A Cryptographic Trust Architecture for Institutional AI Inference
• Large Language Models Must Be Taught to Know What They Don't Know
• SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales
• Can Large Language Models Faithfully Express Their Intrinsic Uncertainty in Words?
• Can We Trust LLMs? Mitigate Overconfidence Bias in LLMs through Knowledge Transfer
• Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Models
• An Evaluation of Estimative Uncertainty in Large Language Models
• Believing Anthropomorphism: Examining the Role of Anthropomorphic Cues on Trust in Large Language Models
• Interpretability Needs a New Paradigm
• Overconfidence is Key: Verbalized Uncertainty Evaluation in Large Language and Vision-Language Models
• Conformal Prediction for Natural Language Processing: A Survey
• When to Trust LLMs: Aligning Confidence with Response Quality
• Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations
• Calibrating the Confidence of Large Language Models by Eliciting Fidelity
• Can Humans Identify Domains?
• Linguistic Calibration of Long-Form Generations
• Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraph
• Calibrating Large Language Models Using Their Generations Only
• Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context
• Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks
• Asking the Right Question at the Right Time: Human and Model Uncertainty Guidance to Ask Clarification Questions
• Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning
• Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Models
• Deal, or no deal (or who knows)? Forecasting Uncertainty in Conversations using Large Language Models
• Rethinking Interpretability in the Era of Large Language Models
• What large language models know and what people think they know
• Understanding User Experience in Large Language Model Interactions
• Are self-explanations from Large Language Models faithful?
• Combining Confidence Elicitation and Sample-based Methods for Uncertainty Quantification in Misinformation Mitigation
• Relying on the Unreliable: The Impact of Language Models' Reluctance to Express Uncertainty
• State of What Art? A Call for Multi-Prompt LLM Evaluation
• A Survey of Confidence Estimation and Calibration in Large Language Models
• Quantifying Uncertainty in Natural Language Explanations of Large Language Models
• The language of prompting: What linguistic properties make a prompt successful?
• AI Supported Degradation of the Self Concept: A Theoretical Framework Grounded in Established Cognitive and Computational Mechanisms
• A Diachronic Perspective on User Trust in AI under Uncertainty
• Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting
• Cognitive Mirage: A Review of Hallucinations in Large Language Models
• Explainability through uncertainty: Trustworthy decision-making with neural networks
• Uncertainty in Natural Language Generation: From Theory to Applications
• Shifting Attention to Relevance: Towards the Predictive Uncertainty Quantification of Free-Form Large Language Models
• Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
• Explaining Predictive Uncertainty with Information Theoretic Shapley Values
• Direct Preference Optimization: Your Language Model is Secretly a Reward Model
• A Survey on Asking Clarification Questions Datasets in Conversational Systems
• Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback
• Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
• Generative Agents: Interactive Simulacra of Human Behavior
• ChatGPT for good? On opportunities and challenges of large language models for education
• Hallucinations in Large Multilingual Translation Models
• Navigating the Grey Area: How Expressions of Uncertainty and Overconfidence Affect Language Models
• Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation
• Discovering Language Model Behaviors with Model-Written Evaluations
• Language Models (Mostly) Know What They Know
• Teaching Models to Express Their Uncertainty in Words
• Adversarial Training for Improving Model Robustness? Look at Both Prediction and Interpretation
• Self-Consistency Improves Chain of Thought Reasoning in Language Models
• Challenges and Strategies in Cross-Cultural NLP
• Training language models to follow instructions with human feedback
• Datasets: A Community Library for Natural Language Processing
• Machine Learning with a Reject Option: A survey
• Alexa, Google, Siri: What are Your Pronouns? Gender and Anthropomorphism\n in the Design and Perception of Conversational Assistants
• Multilingual LAMA: Investigating Knowledge in Multilingual Pretrained Language Models
• The Pile: An 800GB Dataset of Diverse Text for Language Modeling
• Reducing conversational agents' overconfidence through linguistic calibration
• Formalizing Trust in Artificial Intelligence: Prerequisites, Causes and Goals of Human Trust in AI
• Measuring Massive Multitask Language Understanding
• Do We Trust in AI? Role of Anthropomorphism and Intelligence
• Fine-Tuning Language Models from Human Preferences
• SelectiveNet: A Deep Neural Network with an Integrated Reject Option
• Proximal Policy Optimization Algorithms
• Explanation in Artificial Intelligence: Insights from the Social\n Sciences
• Explanation in artificial intelligence: Insights from the social sciences
• On Calibration of Modern Neural Networks
• Deep reinforcement learning from human preferences
• TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for\n Reading Comprehension
• A NEW MEASURE OF RANK CORRELATION
• DebUnc: Improving Large Language Model Agent Communication With Uncertainty Metrics
• Epistemic Vigilance

Cited by

• Patterns vs. Patients: Evaluating LLMs against Mental Health Professionals on Personality Disorder Diagnosis through First-Person Narratives
• Failure Modes in LLM Systems: A System-Level Taxonomy for Reliable AI Applications
• Reasoning's Razor: Reasoning Improves Accuracy but Can Hurt Recall at Critical Operating Points in Safety and Hallucination Detection
• Teaching Language Models to Faithfully Express their Uncertainty

Related

• A Survey on Uncertainty Quantification of Large Language Models: Taxonomy, Open Research Challenges, and Future Directions
• Perceptions of Linguistic Uncertainty by Language Models and Humans
• Believing Anthropomorphism: Examining the Role of Anthropomorphic Cues on Trust in Large Language Models
• Rethinking the Uncertainty: A Critical Review and Analysis in the Era of Large Language Models
• Uncertainty in Natural Language Generation: From Theory to Applications